From 8b1950a97df16c974578328a345d44f27c7cbe27 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Fri, 21 Aug 2026 08:22:43 +0100 Subject: [PATCH 01/34] Add a TTS benchmark template, and say on the page that it can be run MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The open-weight TTS benchmark is the first one here that needs no agent, no provider key and no orchestrator: it measures speech engines rather than coding agents, so the whole run fits in a terminal's boot command and the report it writes is served to a browser block beside it. Two things the sandbox forces, both verified rather than assumed: talkito's .gitmodules points at git@github.com: for both submodules, and the sandbox has no SSH key, so --recurse-submodules cannot work. The three public repos are cloned over HTTPS into the layout tools/paths.py resolves. The image installs the talkito package but keeps no checkout, so all three have to be fetched at run time. tools/benchmark.py with no arguments opens an interactive picker that reads stdin and exits on EOF, so it cannot be driven from a boot command. --only names the models and skips it. The web server starts before the clone rather than after the run, so the browser block has a page from the first second instead of a connection error for the length of the benchmark. Samples are synthesized before the sweep because benchmark.py builds the page at the end, and a clip made afterwards would not be in it. Pinned to piper: it needs no torch at all and faster-whisper scores on CPU, so the arm installs in seconds. A full sweep of all 73 providers is around eleven hours and is not something a template should make easy to start by accident. Not yet validated end to end. The clone layout, dependency install, report serving and browser wiring all work in the sandbox; the run itself stops at benchmark/tools/paths.py, which requires tests-repo/test_tts_asr_roundtrip.py — a file that is not in the public talkito-tests repo. That blocks anyone trying to run the benchmark, not just this template. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- benchmarks/README.md | 8 ++ benchmarks/talkito-tts/README.md | 91 +++++++++++++++++++ benchmarks/talkito-tts/template.json | 41 +++++++++ .../content/benchmarks/open-weight-tts.md | 7 ++ frontend/src/data/benchmarks-posts.json | 2 +- 5 files changed, 148 insertions(+), 1 deletion(-) create mode 100644 benchmarks/talkito-tts/README.md create mode 100644 benchmarks/talkito-tts/template.json diff --git a/benchmarks/README.md b/benchmarks/README.md index f6d57dc8..4d81dce0 100644 --- a/benchmarks/README.md +++ b/benchmarks/README.md @@ -10,6 +10,7 @@ results surface in a browser/note block. | Benchmark | Dir | Runnable? | Execution model | |-----------|-----|-----------|-----------------| | slop-code-bench | [`slopcodebench/`](slopcodebench/) | ✅ yes | **Host** (no Docker) via the fork's tmux executor | +| Open-weight TTS | [`talkito-tts/`](talkito-tts/) | ⚠️ blocked upstream | **Host**, one PTY; no agent, no keys — see its README | | SWE-bench (Lite/Verified/Full) | [`swebench/`](swebench/) | ✅ yes | Inference on host + **Docker-per-task** eval (Modal or DinD) | | SWE-bench Pro (public 731) | [`swebench-pro/`](swebench-pro/) | ✅ yes | Inference on host + **Docker-per-task** eval (`--use_local_docker` or Modal) | | Terminal-Bench 2.1 | [`terminal-bench/`](terminal-bench/) | ✅ yes | **Harbor** harness; Docker-per-task (or `--env daytona`/Modal remote) | @@ -17,6 +18,13 @@ results surface in a browser/note block. ## Two architecture families +These describe the *coding* benchmarks, where something has to run an agent under +test. The TTS benchmark belongs to neither: it measures speech engines rather than +agents, so it has no agent-under-test, needs no provider key, and its entire run is +one terminal's boot command with a browser block reading the report. It is the only +template here that skips the config wizard and the orchestrator entirely — see +[`talkito-tts/`](talkito-tts/). + **Family A — host-executable.** The agent-under-test runs directly in the Orcabot sandbox VM; no inner container. The sandbox VM *is* the isolation boundary. This is slop-code-bench (via the fork's host-tmux executor), where diff --git a/benchmarks/talkito-tts/README.md b/benchmarks/talkito-tts/README.md new file mode 100644 index 00000000..9be77896 --- /dev/null +++ b/benchmarks/talkito-tts/README.md @@ -0,0 +1,91 @@ +# talkito TTS benchmark + +Runs the open-weight TTS benchmark from +[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) inside one Orcabot +terminal and shows the report it produces in a browser block beside it. This is the +runnable version of the published +[Testing open-weight Text To Speech models](https://orcabot.com/benchmarks/open-weight-tts) +table: same harness, same 84-phrase corpus, same page. + +## Why this one has no orchestrator + +Every other template here drives a benchmark through an agent, because those +benchmarks need an agent-under-test, provider keys and per-task containers. This one +needs none of that. The benchmark is a Python program that synthesizes a fixed corpus +and scores it, so the whole run fits in a single boot command and the only moving +parts are a terminal and a browser block. + +That also makes it the cheapest template here to run: no LLM calls, no API key, no +tokens. It costs CPU time and nothing else. + +## Architecture + + terminal block ──runs──> tools/benchmark.py --only piper + │ │ + │ └──> tts-benchmark.html + └──serves──> python3 -m http.server 8099 <──reads──┘ + ▲ + browser block ──opens──> http://localhost:8099/ + +The terminal starts the web server *before* it starts cloning, serving a placeholder +that refreshes every ten seconds. Without that, the browser block would show a +connection error for the whole run and only become useful at the very end. When the +report is built it is copied over the placeholder and the next refresh picks it up. + +## Two things the sandbox forces + +**HTTPS clones, not submodules.** talkito's `.gitmodules` points at +`git@github.com:robdmac/…` for both submodules. The sandbox has no SSH key, so +`git clone --recurse-submodules` cannot work. The boot command clones the three public +repos over HTTPS instead and arranges them as + + /workspace/talkito/ talkito + /workspace/talkito/benchmark/ talkito-benchmark + /workspace/talkito/tests-repo/ talkito-tests + +which is the "submodule position" `benchmark/tools/paths.py` resolves via `../..`. +The sandbox image installs the talkito *package* but keeps no checkout +(`sandbox/docker/Dockerfile` clones with `--no-recurse-submodules` and deletes the +tree afterwards), so all three have to be fetched at run time. + +**`--only`, never the picker.** `tools/benchmark.py` with no arguments opens an +interactive checklist. It reads from stdin and exits on EOF, so it cannot be driven +from a boot command. `--only` names the models directly and skips it. + +## Running more than piper + +`MODELS` is the first line of the boot command and the only thing to change: + + MODELS=piper # the default + MODELS=piper,kokoro + MODELS=piper,kittentts + +piper is the default because it is the cheapest complete arm in the table: it uses no +torch at all, and scoring runs through faster-whisper on CPU via CTranslate2, so the +install is seconds rather than gigabytes. Most other engines pull torch and download +their own weights on first use. + +There are 73 providers. A **full sweep is around 11 hours**, longer on a first run. +Pointing this template at all of them is not a mistake the boot command should make +easy to make by accident. + +## Status + +**Prototype, not yet validated end to end.** The template's own mechanics — clone +layout, dependency install, report serving, browser wiring — are testable and tested. +The benchmark run itself is currently blocked upstream: +`benchmark/tools/paths.py` requires `tests-repo/test_tts_asr_roundtrip.py`, and that +file is not in the public `talkito-tests` repo. Until it is published, a clean clone +fails at import with + + RuntimeError: cannot find talkito. Set TALKITO_ROOT to a checkout containing + tests-repo/test_tts_asr_roundtrip.py + +This affects anyone trying to run the benchmark, not just this template. + +## Seeding + + benchmarks/bin/seed-benchmark-templates talkito-tts + +See [`../README.md`](../README.md) for the general path and the remote/cloud +alternative. diff --git a/benchmarks/talkito-tts/template.json b/benchmarks/talkito-tts/template.json new file mode 100644 index 00000000..239bf006 --- /dev/null +++ b/benchmarks/talkito-tts/template.json @@ -0,0 +1,41 @@ +{ + "_comment": "Importable Orcabot dashboard template. Unlike the other benchmark templates this one needs no orchestrator, no agent and no API key: the terminal's boot command runs the whole benchmark and serves its report on localhost, and the browser block is pointed at that server. Seed with bin/seed-benchmark-templates.", + "name": "TTS Benchmark Runner", + "description": "Run the open-weight TTS benchmark and read its report", + "category": "coding", + "items": [ + { + "placeholderId": "runner", + "type": "terminal", + "content": "{\"name\": \"TTS benchmark\", \"agentic\": false, \"bootCommand\": \"MODELS=piper\\nPORT=8099\\nROOT=/workspace/talkito\\nBENCH=$ROOT/benchmark\\nSERVE=/workspace/tts-report\\necho \\\"== Orcabot: talkito TTS benchmark ($MODELS) ==\\\"\\n\\n# Serve a placeholder first, so the browser block has something at a stable URL from\\n# the first second rather than an error until the run finishes.\\nmkdir -p \\\"$SERVE\\\"\\n[ -f \\\"$SERVE/index.html\\\" ] || printf '%s' 'TTS benchmark

Benchmark running...

This page reloads every 10 seconds. The report replaces it when the run finishes.

' > \\\"$SERVE/index.html\\\"\\ncurl -sf -o /dev/null --max-time 2 \\\"http://127.0.0.1:$PORT/\\\" || ( cd \\\"$SERVE\\\" && nohup python3 -m http.server \\\"$PORT\\\" >/workspace/.tts-serve.log 2>&1 & )\\n\\n# talkito's .gitmodules uses git@github.com: URLs and the sandbox has no SSH key, so\\n# --recurse-submodules fails. Clone the three public repos over HTTPS instead.\\n# benchmark/ living inside talkito/ is exactly the layout tools/paths.py resolves.\\n[ -d \\\"$ROOT/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito.git \\\"$ROOT\\\"\\n[ -d \\\"$BENCH/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito-benchmark.git \\\"$BENCH\\\"\\n[ -d \\\"$ROOT/tests-repo/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito-tests.git \\\"$ROOT/tests-repo\\\"\\n\\n# piper needs no torch at all, and faster-whisper scores on CPU through CTranslate2,\\n# so the whole piper arm installs in seconds. Heavier models would need much more.\\npython3 -c 'import piper' 2>/dev/null || pip3 install --break-system-packages -q piper-tts\\npython3 -c 'import faster_whisper' 2>/dev/null || pip3 install --break-system-packages -q faster-whisper\\n\\ncd \\\"$BENCH\\\" || exec bash\\n# The listening clip has to exist before the page is built, or the table is numbers\\n# with nothing to play. benchmark.py builds the page at the end, so samples come first.\\npython3 tools/make_samples.py $(echo \\\"$MODELS\\\" | tr , ' ')\\n# --only skips the interactive picker, which needs a TTY it does not have here.\\npython3 tools/benchmark.py --only \\\"$MODELS\\\"\\nrc=$?\\nif [ -f tts-benchmark.html ]; then\\n cp -f tts-benchmark.html \\\"$SERVE/index.html\\\"\\n echo \\\"report published -> http://localhost:$PORT/\\\"\\nelse\\n echo \\\"no tts-benchmark.html was produced; measurements (if any) are under $BENCH/tools/\\\"\\nfi\\necho \\\"[tts benchmark done rc=$rc]\\\"\\nexec bash\", \"subagentIds\": [], \"skillIds\": [], \"mcpToolIds\": []}", + "position": { + "x": 0, + "y": 0 + }, + "size": { + "width": 560, + "height": 560 + } + }, + { + "placeholderId": "report", + "type": "browser", + "content": "http://localhost:8099/", + "position": { + "x": 600, + "y": 0 + }, + "size": { + "width": 900, + "height": 620 + } + } + ], + "edges": [], + "viewport": { + "x": 0, + "y": 0, + "zoom": 1 + }, + "setupGuide": "# talkito TTS benchmark\n\nThis template runs the open-weight TTS benchmark from\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) end to end inside\none terminal, and shows the report it produces in the browser block beside it.\n\nUnlike the other benchmark templates there is **no orchestrator**. Nothing here needs\nOrcabot chat, an agent, or an API key: the terminal's boot command is the whole run,\nand the browser block is pointed at a local web server the same command starts. Start\nthe dashboard and it goes.\n\n## What the terminal does\n\n1. Serves a placeholder page on `localhost:8099`, so the browser block has a stable\n URL immediately instead of an error until the run finishes.\n2. Clones `talkito`, `talkito-benchmark` and `talkito-tests` over **HTTPS**. talkito's\n `.gitmodules` uses `git@github.com:` URLs, which cannot work in the sandbox (no SSH\n key), so `--recurse-submodules` is deliberately not used. Cloning `benchmark/` into\n `talkito/` reproduces the layout `tools/paths.py` expects.\n3. Installs `piper-tts` and `faster-whisper`. Piper uses no torch at all and\n faster-whisper scores on CPU via CTranslate2, so this arm is small and quick.\n4. Synthesizes the one listening clip per model (`make_samples.py`) **before** the\n sweep, because `benchmark.py` builds the page at the end and a sample produced\n afterwards would not be in it.\n5. Runs `python3 tools/benchmark.py --only piper`, which sweeps the 84-phrase corpus,\n transcribes each clip back with Whisper, scores word error rate, then rebuilds the\n page and the export zip.\n6. Copies `tts-benchmark.html` over the placeholder, so the browser block shows the\n report on its next refresh.\n\n## Running more models\n\n`MODELS` on the first line of the boot command is the only thing to change. It is\npassed to `--only`, so it is comma-separated:\n\n MODELS=piper # the default: minutes\n MODELS=piper,kokoro # kokoro pulls torch, so expect a long first run\n MODELS=piper,kittentts\n\nThere are 73 providers. A **full sweep is around 11 hours** and much longer on a\nfirst run, because every engine downloads its own weights -- which is why this\ntemplate ships pinned to piper rather than to everything.\n\n## Reading the result\n\nThe report is the same page the published benchmark is built from: one row per\nconfiguration, with word error rate, real-time factor, compute per phrase, PESQ and a\nplay button per row. `tts-benchmark-export.zip` beside it is the shareable archive\n(page + measurements + audio).\n\nMeasurements are saved per unit as they complete, so an interrupted run keeps its\nfinished work and re-running resumes rather than restarting. `--keep` adds to existing\nmeasurements instead of replacing them.\n\n## If the run stops early\n\nThe terminal drops to a shell either way and prints `[tts benchmark done rc=N]`.\n`rc` is `benchmark.py`'s exit code -- non-zero means at least one provider failed, and\nthe per-provider output above it says which. The page and the zip can always be\nrebuilt from stored measurements alone:\n\n cd /workspace/talkito/benchmark\n python3 tools/build_benchmark_page.py tts-benchmark.html\n" +} diff --git a/frontend/content/benchmarks/open-weight-tts.md b/frontend/content/benchmarks/open-weight-tts.md index a95a19a1..1eb9d186 100644 --- a/frontend/content/benchmarks/open-weight-tts.md +++ b/frontend/content/benchmarks/open-weight-tts.md @@ -10,6 +10,13 @@ We tested all the open weight TTS models against the same corpus, using Whisper Every row can be played, so you can hear what a word error rate actually sounds like. +**You can run this yourself.** The harness, the corpus and the scoring are open, and +[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this +page from nothing — there are no stored results in it. A single engine takes minutes +(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it +spent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template +does the whole thing in one terminal and shows the report next to it. + ```chart tts-vendors ``` diff --git a/frontend/src/data/benchmarks-posts.json b/frontend/src/data/benchmarks-posts.json index ab282645..79692bde 100644 --- a/frontend/src/data/benchmarks-posts.json +++ b/frontend/src/data/benchmarks-posts.json @@ -31,7 +31,7 @@ "slug": "limitations" } ], - "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nEvery row can be played, so you can hear what a word error rate actually sounds like.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for twenty-one of the thirty-six\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones — so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the twelve\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2x for readers who only care about what can be spoken\nlive, and sorting by RTF or compute per phrase draws a line at the cutoff so the two groups\nare visible at once.\n\nThe line falls at 2x exactly. Nothing sits awkwardly against it: the slowest engine that\nkeeps up is OmniVoice at 1.51x, and the fastest that does not is Chatterbox Q4 at 2.03x.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes, feed-forward over one.\nDifferences under roughly two points are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" + "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nEvery row can be played, so you can hear what a word error rate actually sounds like.\n\n**You can run this yourself.** The harness, the corpus and the scoring are open, and\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this\npage from nothing — there are no stored results in it. A single engine takes minutes\n(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it\nspent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template\ndoes the whole thing in one terminal and shows the report next to it.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for twenty-one of the thirty-six\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones — so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the twelve\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2x for readers who only care about what can be spoken\nlive, and sorting by RTF or compute per phrase draws a line at the cutoff so the two groups\nare visible at once.\n\nThe line falls at 2x exactly. Nothing sits awkwardly against it: the slowest engine that\nkeeps up is OmniVoice at 1.51x, and the fastest that does not is Chatterbox Q4 at 2.03x.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes, feed-forward over one.\nDifferences under roughly two points are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" }, { "slug": "agent-skills", From 5f3808ce8f90606158a3e44cefe04dc8b3c77e22 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Fri, 21 Aug 2026 08:54:09 +0100 Subject: [PATCH 02/34] Add a social preview card for the TTS benchmark Same shape as the agent-skills card - 1200x630, title, finding, chart, footer - and the palette is sampled from it rather than guessed, so the two sit beside each other. Generated rather than drawn. The y-axis is reader preference, which arrives by ballot and keeps moving, so a hand-made PNG is stale the week after it ships and cannot be re-cut without redrawing it. The subtitle's r is computed in the same pass as the points for the same reason: a re-cut must not leave a stale number next to fresh data. The x-axis is logarithmic, and that is the one deliberate difference from the on-page chart. The page fixes x at 6.8s, which covers the engines that keep up with real time; ten of the thirty rated engines are slower than that, out to 51s a phrase. A card has no filter controls and no hover, so it has to show every point it counts, and log x fits 0.12s and 51s in one frame. The flatness of the cloud across two and a half decades is the finding: r = 0.21, and the sub-second engines alone span 1 to 94. Two layout bugs found by looking at the output rather than trusting the code. Markers and labels are drawn in separate passes, because in one pass a later marker paints over an earlier label - "NeuTTS-2E Q8" rendered as "euTTS-2E Q8". And the y-axis runs to 105 while its ticks stop at 100, so the best-rated engine sits below the frame rather than against it and its label has somewhere to go that is not the densest part of the cloud. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- .../content/benchmarks/open-weight-tts.md | 1 + frontend/package.json | 1 + .../public/benchmarks/og-open-weight-tts.png | Bin 0 -> 78514 bytes frontend/scripts/build-og-tts.mjs | 250 ++++++++++++++++++ frontend/src/data/benchmarks-posts.json | 2 +- 5 files changed, 253 insertions(+), 1 deletion(-) create mode 100644 frontend/public/benchmarks/og-open-weight-tts.png create mode 100644 frontend/scripts/build-og-tts.mjs diff --git a/frontend/content/benchmarks/open-weight-tts.md b/frontend/content/benchmarks/open-weight-tts.md index 1eb9d186..628200db 100644 --- a/frontend/content/benchmarks/open-weight-tts.md +++ b/frontend/content/benchmarks/open-weight-tts.md @@ -3,6 +3,7 @@ title: Testing open-weight Text To Speech models date: 2026-08-17 description: Every open-weight TTS engine that can run on a consumer laptop, tested for WER using Whisper. author: Rob Macrae +ogImage: /benchmarks/og-open-weight-tts.png toc: false --- diff --git a/frontend/package.json b/frontend/package.json index 2ebbc0b0..d179ced7 100644 --- a/frontend/package.json +++ b/frontend/package.json @@ -9,6 +9,7 @@ "start": "next start", "lint": "eslint src/", "check-catalogs": "node scripts/check-catalogs.mjs", + "build-og-tts": "node scripts/build-og-tts.mjs", "test": "vitest", "test:e2e": "playwright test", "test:e2e:ui": "playwright test --ui", diff --git a/frontend/public/benchmarks/og-open-weight-tts.png b/frontend/public/benchmarks/og-open-weight-tts.png new file mode 100644 index 0000000000000000000000000000000000000000..4496b60d647dc81af10a2b106eaa3edd5caef8f1 GIT binary patch literal 78514 zcmeFZRahKR*ENW{yK8WFcXxLW!5Ry$jk_l}1a}A$+}(n^JHg$(nSTH8doO0L=5j8k zpStL$>glekQ~T_*_F8LKw7RMsDiRSA1Ox=Cg1odQ1O!YS1O${P0u*>pFV#6R1Ox_z zg0zIT_pkFDSaa-U;@&GNm9+(vik6hUOYn!CpbpZq4)T+Ox0wT<*PK6ecVKgq)R5kz zU}5to_v7bJrV*Z{Vknl6fN{l3*W1g@WVNj~GP#|#BykKFiT~LE)0x*BSzOza9vYgHQys7V zWnyAN)ju`mN2UrzlttaDL!g+m5HWE#Lv6V{4l#ZrF7|((K7{6f;st=ij8~-<$!i&b zdb#ljHHv>C4+m*za{bBcO3WWLrxPu$NO2VvO9#69y{`(~?gT!K<$Zo(oanwd3x)bP znZ(O^jgh7*WNh<$C}xb<&k>gg@oJvNll)bkl z$g~NB5)=quAr7H}tE+jsGPASOv(wcma7#nLm#Ph(D@$MC?oZlo&xR;U(|w8?k_Qqd zt9m|85=s5gFhQrpW^*lfm{Zew+{qHk2;qtR+Swft$6k!Pl^g7edGD1^pLg-op8XJ0ny zMd*o?3S&Q$+cT_`H{-_-Leg||OUnb3GPIQMzGC{Ofp4IyDPZJ20*Q2d#KRdU-<62} z`_bAYc+_Vdl>U<>iakS+EGB!$=JrOM(<>C%@mWo_pQoqj3Z`RXX-LrM=x3bQia9>~ zp3KRb{ZGog$b}VE6{wS1(^R$dWUu(By3#LFhg-_Q+(EIFy75V$0WY!OZ+n?1}_|lM*Lz3Ya z@cd94|9M`>SE6Y*+JJv2W)$!+oI~Dis_2I-k^Js@6Jl_OPwa{>n&)iqnEEpf%~Y1Jd6<5-pNy*LauJ|8Lb$G)@c<|6y z0fDu$QY!+T!yFRQn3%8^SXk+{x@YHV_?4BvYtdDN7q>CG@KZ99;V<5dN? zzp_rLzP!A&AF>fXOJ5^pX0sI*B6exlse~@W*hrY#ec)yZ2^nR_O*j;&%M+ld@_r8` z<|!MWTEg2h_mzcffJ;??O*V8Y@n`2dVaw; z*U<-5QF&=cf-c~+w6Zypw3j7_1}I}xy@naow0_A?bshbJdtUUgZ1#~t-Nb!xIXJG& zPLhR4M8YN7~e*Ql*`EusE*5sGAi3Sk%n0xHJmtp7Qp&?5fkdOfP(#kcI! zP&(_^?g_B5w@gm0J{LIkZvA+cLTCI zNWVWEVp{%u_v5;|KhYX~yjl$1djEFgx<&}=?c*Cq$kA$6dr7gTiMB#jaQ%1Q#t)rD z2nZ_75(d~=`~mJ~*A-gqVk)HtP76c4-XdB?B$5cu46Xi9uT&_+-_;2Gz4iU$QDig? zT8M8}ZkVp~*VAfxm{SDWIhW%e!P*5}`$6s}n)6C~sH$cQ>xEcu1vAR^*m1)SYvhwd zNjKZa>)Vk9rZmsM`g~}c&)XJsb^ZCDWE&fm>~D5DR5wC`YuE>p4gS3LbJy8cd9bf< z{bc3T#LegV=)9He?1UWO#A8e2p`hR2{X50QBcI4H39<3W5Yeiz%cjo+s~oEixys_5 zXB!a@kb$YcTrc-hw?Efxqc$Ks8uXr6lhz_0Ld)Am)QbwT#R6V8wxH!QX@2e9sU0@j zE&kg2)o~oVx3}k<4%}+9H1eKfT6R{;d5B&jw#&`9zQz5wwWlTq*kI3yxyVdShQ6}D z6YJQ;%s6_ySVUbT_r+FZ6W+1On;_(Kr5ENL<>+G(Oy7?d=U=~@-o75%oDBEr2n2zo z7_|idu$gw<9If$rOqTPi*h?Ytyw3Y@T{xz4c|<1UfWG`A1d5Ma$B!fs%@sJ>w-bd- zxIYX#mREaSicu*RWm3{j8<*qbT+dE0mjkHL#?kMy`Cg_>FJ% z9ZaXq?$3v(9<46+g40Aiwv735jk@Y4o<_1nvPenn7Q1y;Yo@X$g5H}vmW<6mj`nn^ z!30xX0e6?LET5tQuW^~haD5&<29E+=KBpT)580nON{^BAw2zmw;al%thtR>8fUg=f zw6&E$r~wz0wNhkH_k7EJ|B}89nt1=KuOOa0_ERAakQQIvlbUNtYvs9jAkThJB+}uD zgmNG13u6kIsf>r6hw(_o`GwnIWDB|-MN0jAz0pB;d3$+e9r={=o4L32{T)6gpVOw0 z+b(B8sH<0$yazOHKau!y!UZ#dtP^|V=i{=V$cXdG?bUGsbdISXjsS^jUakqfOziYN zpE*jtqOxKH+{bo&U~r?vRhV^`M(9pYz{}%8n^K-XsKgrRwA?;anb!kZ9npq86;f(* zQrN~-rk2rjtkfi7ARz&oN!HkwU=8cpX(18wr3XSpF-v%S==uLC_e)3HnP!fprsCtP zm8f;Px88&hE4`m-x6(Qc4PDe#p*&xzfCfS!pxz)x->Z=;lOdBCHZf#T#D!o$GN-afDb_L zb}CkYWrgQ2JU_C2H&0`x<>&CbnmC1wQ%>RvCnF1UF>&IzW|}hwMiUuac*97qhG8I7 zu=he#FooAvC-7=}^Q|DA#=S^a6xQ0>KF|m;1hXkp0B(JyUmP;^AU$t{kJrEoJT&AXlg`4 zt}DF(T!#UkM_W?9Vyy)kNBxJzH`J6q?jhq((hhh^_J7ON660b@yolCI?Rbk(&z6l| z7QYHMxmQZCeOL_rgarCl(m)fLstfG6{Ts|IG5hHK^;VflDZ1q$K3iY!-(U^LebtrI zE?$q%@^QFPQU$>n0oOs6p_Bo?zPi%ryon{6JRENf+GKk}Bcnlq$aQ~k zW-4&4xQMj2Q13ClX72fI*y42OIf-4sDVbTOV~cbw1?&!v1IcQ@&VDx_lhD=!zv}Vp$yR(dA;}H9&Fh3CH;_jVj=IxiFH4v z86fW4L^?b|64#j^i08do!n7nQS zxI@cB)lu;zI4A=K%Gd~n2! z&E9pn)tpv=${FlHI8%0>pM_&SYWCRX={30n>guyqS@=dKjAmiTA%VQs`N-$zXEadD z{#kSqEA~7 zXYtBI%N(~{<=oj7{g(?!ziSn|fHiLlPboi9Ro-DpV}2(P4RaD~bX2epmBqw4hS%iL zA})qBn@nr!(Wud@URl$helf2<1S$GZRcm*uK+m29=1nffX+p#0m>4bDR_ZH@i>-8Wr$;7$Wt2`u@cS7p!AY8w`<8j*AO$b~`F8Sz^dA zE_hWb2=}Y+tJ^GQze`a$AxA*yOlcw3^2czORZSRzf@{;+Je%uPQ+i7HfG*gmeX+n6 z35`3J4k@u5mCF90VRH%8d%agEM;2T2fhE)F`K#52h&~ zH!eFic4=U((E;v|>a=6YVw(c85wI1u6&A6-eh{1aKu*GDUccG>)>EFr?@LE=y}2Aq zO4^vX-O$_!b4U3gLQSpZ*7z||fqm`}kMY6avn)`o{BeW zG=Pigo%n($CY%`j-5)z4{QmFUZq~@Hq5%eN$|}t>V@G)tr6L;K`7bzKxg~cGpLm!N ze@>5ZQ+jiLq?Tkx=EDY)L75>TBS~z9Br5YLCq%3vlK*HGj5vT{8%&?7tqe$zxZ6xNdje_VYeo zVeodDIFpKo<-_kn6ufqD^u;t$li(o|0w5Hiw!X1#1OGDH?c9q3L36K@pQ8RG{lDs9 zhSSj;jW7kDy_d8Ljs4rcK^|;G_~GJJPOm{#_+~o&2^b60m^j#H7`u^n4G&@5g@r8{ zUXo0fxtJc+J`5~rwA*2{+baHX2vR$o_*O|XrWg3sw@ut)((c<(emptYIgA>5G@Yw@ zZ2ow!u+Uy=Xp5HH&OKRu7a?OEGMddEk}7 zNb#aJucxFE4##}*hadFo{S5Ov4p(VhS$fOjnn7q$gaBK(YnV`cQb8f#3XO!x-y^5qPdF_Fdl z;{-1uDgl!3`IlN+|N1vUtlfnDk<* zznB*;$!BZN*Lm(Y=lG!)7nk0>_#c~{U?Zk7G+KmGU?3n^1|2>d=)+XQ^OipiOCi8n zmd#Q61mR5*&kyV^Ld|b?>g+kZ1OApxd?@w229wpnFpe~%RvEU~8oD->#`obMVKEjE z`w0>ARQZ3^y!8^RX%Bc~wY9G{;x+0tR}IQzvml`06PNmYI=pbXVUy^O$laMwwBhtS zj_7;HC+7FK#PT=4n>xUr@IdH5Cy>#04tND#95(fE4pDmx9$%EBp||Ia9uw3-8F%}h ztxXC6w0Oj&LL<@)Wdwp`Y0fE`by!DVm^7K1y%JHpz@ zq$dDmP5ETmhfF?FfdJZSC2?Qg8c;YZ$-p>P$OzVd3XiD=A5{kI##zCvTXA zI1{AgzZS7~$kQrVDbS9^RR-oW|!I$r7VL@uf-uk7mK$-0e= z7%o|v^m)5kPYJ!fAcWzb@l>6{8_%icuu{Ljmr880`$JbZEy=_jKTp&L1&BgFTtpj4 z@K{)E1xMUhU}AJUszCAn(co6QL_n!dgl*dW>c858o);$B=Y&B5gM_Z78CDv4!_SqO zn4Fm-;n&b)e zf_6o685>sR3Zrrv2Y2+mn4vr{@)8j+zg-oO9|$`VyYym$6Z)3U$T-8v{JI@eL^rL0 z$I(>CX)$sAVC1O|ds0=B$0^kPmY6$X)EEcgqkvzchtE&LG3Q57ChLNWpUr|$w+ENO zcg;C(`AWBUkA@wm@Mg|xr*!4F<|B|UK2v2%eMyj67GORs(D+8o&^UsL&(8RBtjQ=V zJ$xBy`t~gEOhPIkH3g+Ronc=Vo%DYA9CoT5a{$g^qdvJD!m@0FmV1~3ba!RiQv}6^nlvX1YLcx ze;2e5W1X+KVW5@18ru5qydBJ$5-mI#N*Uuk=yaC=3Pw{{aX=x;1WJFj(a#4TtnZ?byC{_4VEyn8;xluy>{ zg#l$fY4+9oi!4SHKARaLDPjT97nGWxo1df>Wa?tAtQyIX!FKPBi4}xpS>mc8x59Sk zqt#~tJ>UJe!SHAiO~2>BZ0rgtWarSaBR(d@8#LD>fShL9?TdYxo0$qLQ@9b{H#>{w zuJqdQ{V&p5AQbKHMaLsDAy>WeFOn03A``(zOIk$VdED(!)rU(8s`ux&Q!vGXAxN;$ zga1KVBL;kwv~Fv6{or+jVnNvRfNspD%&%8|HfN;?>c=hbw???O&0__j@>_8*>gD7>S6!^JrCI;$0+!`2x4G-7$ ze7M@ovBNjJU3_YcfekwO*bjG(Nba+GAm7mR_no?SAahayn8GPGdT*rK&9d)Hy`R`)Q@_3?)nHnCq24FUylY%-<#fuDJ zpB4*SNkuS3)^%S@2%Sto26$gv$ZacLQEEQk`8zROotq%nH-H8~io)eZ`PI65mSb)O@b^A&gu4Q8k&!q7< zm7jC=p-qyMdA+QuOk%R>)ozydtgo+~#=zD;?O-;E*i4yOJGHk7%x`-+n)rcMOil7~ zvi0WWOnV+mu{BvH^f@=9mOjMQM1eJf%CJ3I$cmEF9GlU6ruL~|O2C@-<#uopz}bLBk9gsGc#`@g8pA5GCPY%7r2*Qy*>L?EOBbhS zIrry}){i(iWytBZ6{HY}e^Z2nPBN@~kG80JbV%MUl@s}%3Cr!muKC^hRU}KCOj_M@ ze5mM8#XfKN&dWp-^F-Y=VGpapv1!~DkS#EtBOr@|<53E{d#?Do)2_kcGEp!j1^<{M zYR9al^uD{i7~8Ec0qoVOIa``3ZgE`w_0rk%(Ijz^A-3#T;07V8&5qkt;p7Jr`i&QX zk_(&bv>KC_ z;g`kZlph#jQ=%UIGe{RDU4gMGamg@{Nu(+3KhQtRTz>fcMd`TvkYeCk!lkzHrFY`@ zmt2zSL&z;tkM2)jM7Ic7TVMA@Ban$g+^S-FLeeP&8euhLvCTK%)}G8%Vw|1YLx_bi z0sB04RLHMWXZgWSM)RlWn-j64^wuzeaytG{O%-tsq8SUOmXALS-(bnmC;y0;<5?l~ z;ruP}Gedh*X+-h%oNTq>p6JCzCld>g_}M%hoF#{gzx9x#U>Y^P3Ur_O z`n@A8Om9M^3Tc|I0xd{tKNeRkpy_rQ?ZPmayhx^D*ok$@Fp-3})l6D>qO{ss@*?f-rzZNKk9f{jCRVBbY4uTW&{hlU0yQy3L-e&gkL912{tj0>f^`(o1z+yFE;e zlH2)79X}}?Qms?tq~``VSC|m2U7=m8pKCV}!uSQ1qFM73g_irh7OIKI(2f}Qe{ZV` zZ1_`kz{osBS77Cst=lGa%r!@$sDF8sd#uDTvAr=%NKs=mtww1w@4+j&9-XsprQ z|5aoLnAsX7lL}(>v%9)3(g_eW9lB0RGNzigyM|`zB9;7cDcbR@&VIVw(otb2o$Eo@ zqGTo)@bSsoS7Op95cG2Bu)T}~1&CUq_ zaPYAGHs_<16rEeZ>*v9fyPet@?2+vX_kNTKNXeo&I~{}~!d*=S)GE|?5YON7F58Z! zDWV#<=}~{P=4PE53|(C?uyf&5pm~_hdRP{ZQ>>;1D@PAs2#RxR(^(zXiXpkAlXZ^~ znTp%T&iF$i|kuUuSi213E@ zk5JNJq8W1Qfd)y~h%iZ|#_12dO?mjji2#5)-M``mdKuzDzt7KO(`y6o2FP>XP;`(z z#98{4v>^{zQD#n9pS=%D7kd^^@A>?xP_wm8Q=e@g8VQk9IYME zi*+PJ^;A*r?n{8P4cEBLYCX|PZLD%YrgR{=BeQJDBg;&OU;02tOQZdKp|WEVVw5a0 zUSOB+hi++bM08%_a;L#O+WCt7G_n7YK%zG6F&Hx#xB3Isd-ynTi56;=5_@N3q7hqU z8Z_q@r_4G_A4c0j0ftX+Z(#YeLZ^S8OL3Ky&tcW)V3BXhR(W~>RefA3Z6 zpoD#~d>@7{78>I^JKs0n->U;ZYuZfBbagA#1t9}O4V^K|S!k;bHNknUj5__4N>1)y zuB#Sb@70ag1LjxJtz6|)ex~)AY~4+49g}Q0Mwb8yNVklGRy*<&F&Wcc4Xg^g5;D2z ztN`UostSs)xq@yk{VTf**6^v8bw0_mjlYmYKe>)rt8!qQbHSCq|I5X6m!JE|LJg01 zPo|V5eeccfMS32Gr6H$+3Y!`l#+U?mCV)50+OmpeQ;JDM*qH94`;SheQoe_weAxyZ|17F;2Yok^yngWEpz%_(2()F^}JYz z84Zn|jAmkvs2o!mg`%&IFFPHEq*4`A5WBl4eB!S#^nw`CwI-!E7wSmfZ!jUH@9~BT zOeWf$r{3ah`MeKkx$b=%eD+<&t=<=}7k&WTjpY%oyFGxA*4&ste!D-+;^G_+zmPzA ze0096j>hW`(x&+QM|0_D?ij{up5eb=Qa9y@V>K$O0mJdEP0SbM_PmucT4!JVu;x2kp zCKBT0y*x+|f#*iQT(4Sx7bvlwFzX61TF%l41615krvb+|p!vSS?ZEvFiUE9on}TLP zo0FITGe^yy`@Bt6ZTv!y7pw7G!U3=Tr|TCy1@RpoPZHa?CKW2SpyMEpmdjLk!^GUM z8SI<&COZWq)7_KHEkEktW<5SO@5g-ObtOa*=E%zPP)&fWDtf7oG+`k{TBoM#|`5=Z9L<90|FcN1+;d+LFQbtFO<| zPN&x@s*FYMF{YmWa{0!8z=U!L1?hG>aaV`H;$FNIJ~0m3@QX=l~<)ZWnvnSf6nB^;$=a zjoSF!z=hM^BEC}gTR4tGa4?ZdYm!C`+WMU=& zyBN=Eqi0{g+3d)~yi!@WAm4`1nH%a6H8T-lR1Ml=bMtqD*49In-RC45wqptMnK4xI zjXXjO@(fN^u8Fom^?w=r?g%=in9i8q66)0p zl5zpBH_V5$jqEAA%!0fLZ9N*H6yDX!*@mR(EwqA?W`IIoP+cVrZ;^|*NGH!U(za;>5`d6pV(_7+>%PQro`n8U>9 z6s#^;`|)uTIWIog*+-y-v{c{J15}G;NUz&@E83TD-JynLoRl5xMLIZQWh5=YncA`c>FLe_sb2j$jU~6tgB#yN zPLLVCzz~(>z6vb+1P`K*p?{kyY-?~Yc|5LYXng(VXw~+Zm^yFG_Kk7)&RWd|?Hehl zHIYU|MrLNs3-I9gOPwO?P!Hqi-Rj<9tGnFrU>sz3*LOo-PV8 zUk(GFm+d-WQH|*s>U!X+Dmn)8FZ7)m*lBVySo`{NR$wWXkDE41;H^JLMHsBSUrq>U@-Np<+y+*%+=eC^w>h|ZIAeMw6;fo)YzMa znk&}eV;hYr4Vadei}(~y5|oaNhjG|{A9l(R_4A)qD0*Zj9p`nV_YT`N3;aE zg8MjSAt7QeN2Te&yLs~H+MoAk{(nOfk0!IwKE9d+iW%DEueLo+*0%*#ucPRiwqHO{ zqJ-TVd0mdsbb?Jra3%}mcatHvpQM71s{ZN|iwGyjK48m5CMzs^s0I=xSLfEG&{O7e zaJSp!a3F-INRZMZIuZ$3DRSFW$pn&hR_NIC0&iaYeG^;saxREJ zVM@C684Ef^CS;pkqZ8G>O_c{9At7C($42?zKNF^~7H0i9>fHq!o93wV3jClWPctw> z&1gynhv}|g{F7FCHS#@|N&`J-r8dQhi>Zo(na}?Ho3Xj(FN8{|We>`HT-to=>hLO< zUaP-+nm=L<7xly3xtebA5+dA zSiOz1C*=4K3iJAL$WqT~BOxlThWL29%BZ5`j@N0^phQ8$^Od<816&FxZUutvK~3R3 z)26K`0Jw`AG9g#_KhjVvFR|D1HgqF@s3SRTW`aY?cmS9~V8;_(0V6dkHgkATTag(v#r<#aCkjW$;^K}xy^%f@=)21($PgA776ydo|y8`TW z&>@y!ggw@Ie?ytv%Y4ImHeK-bw=H@k^~DYpzJEcs#yHd%fb+aqx%BZa4sXjh-Gnev zD$y!uDS-eCvoZC+B+_&DED#mwupko9Cu{qXn~0*M=U1lK^Igp$w?K@iO*k(V?}GIt z5>k?UpYyGf(4~gnOXy~SSLt%w{Lvg?zum0?Iy}4(R?@Y2BEcw8f^WDp&e4G6JNW~! zUo;-_UEuTWHS$17REO(aV!TG+E;oOP!t~(@BsKD2d^eX$Jo6Nq4`G7rIBj`*tZ=-$ zUzV9|&*${^JWz3;J4L$YzSJzwalhJAyCvR<47T_7$+4&{&rYTUXnifE)Z+r{?0KMwYhHXc9`ts`1Sa*f1m(HFe~L8*tz#jX%T){AQ@ z$VWi=tKh>Dvs$NX4Ix*s+!aZ%k6uMpF+D)lD}&NZ-1GsskwHxLM-;^a0Qd?x1B1vD zPsWNcW2C{9dBlv0gvD-Q36rOTsY(^o3=!CiXUz3$coROc=j~)t`<>db zr65b#9|tN8r}`--T1lNt1&we9UX;&yJMTJJ6TZCN&=M@Ho&Xy)2(s50VLLt56K!8G z;x@5|QHtOtL+84fU?*Bl72iSiwrRBXa@4byJcbkV`)VfoT%@Zy1mKc{K zxSllBC$9hNc9+hu%!C6I7k~QmovH5C_5&B@82u=)ruif;p*9?Ad!35%ne76vKat&-h^;jogW zgi=;!!>a%8u{fy&=_2QbxOqAOo-A~M{N(O$oQYw^|1lq^rPt*1P#wQtiuiLGq2Je( z&)G+%hTh^vNR6nT5t+dkUQGDpRSglFPu-by-=>cX4(s$&2i`0uRZ97wgrjov5L~$DTi#FOYa(WYrF1!dHbFSoO zjhJK`xBq?8?EDxB2sLYKknoQ49#Bm;Wt_on$x1+!A!IXU;(vUsAjzg)CuaC8wpWdf zpV#`{k#``dD=WPd?NCdl_>M13sVChjlN7;0G3kGkf13ryLR!EBa5;cFUn*P}CYoINfdbLvLJUgK zgXgJM+0d@&r!*A@f&#eE^L?Yv8AFv15P?g01sv3>pEuDGH`-nYqZ4o-x})$QK2P_w zNSD)&gz`QQI9D+pRE+Bp@H(v`;^H9@n3z#;&`q13i+CH4Hn5AdgzE9SUE1WURkENz zeEIjX!=Y7?+WqkT;xGr2n$~Hho{L%v=)u6w_;Gz^CIS~%xuMUgj2GDE3YIq(zxIkE z`oK{k+g?ex{g*(Rx7fGwf0}1(+F=dj6t;wvS2B4+g-6EJsBqZ)<2AZsw_~AeG>@)W zswtcIv|8%e5!Lc@0h?|Sw;xPpA&0L~FN6-c9cP>zv6(3Orw;y>-g!ds!$dThOX8R> zkHWWt`mp@m#RkjfDsC~T3Khn%LlJqa5ErflJbI`sbj<#0i?s$rA@qlhIJFAq=8f3E!IJ<9UR3+G?A=^U!7 zH#dHn{Q8yu85}KU&h7ecwe&tDUGmk+Zuj!= z^z?j?XZP^nP{f}IFxnAtByNX(|J3xSxxT(0Cw9odV71~T(@HV&JJK^;l_Wf?oP~MA zst5G@H%^8Hwjuyh`uNSwt-F^OCSG(uwPhTs5yDgT{_f%4KZIH`f7?GUY?*z@jId4e z?!bt+vQ(tjrk=bOdHM)tMmN!hU~MpAWtRTdvT9N`p-8k8r8@2|?PE=LnFGi=eTgYjZcE_lPl{9cH1 z=KRIbR_Bx$sJ5fS)APeqD9#A|F9}NW3v2&$io0u>*^chqT}_l zYmI10&_>o95?adS6b*g9x&LOpizm{_xf?Zt!0lfP&pkw`0Ei0+_*|!a4eS)b^0b z;2{Jm;e7Ky1mVmPTYZDu>4FP6mp?n#S0oi;S|ErORzx-o}6V+e)4}>^5k{E zJ!EU)^31+ucoNREw7RNs@Pps$U0v}&SJ~ic^KiDgIkG;#JTgBu9PwHqPv7bGwEHr? z43?cA(C{zBF)V}A!dB~|Vz*TYWMO=qN5`ftmKZ*E_W-OsO0*J)NIr>u`J4#134g$0 z*4Q{YHpanAMDV{ogxVPYw?0AsZ#e%S?li;;UVvd{sh*rEM=7A^KouoN~K}Fvr{#gM7Ok=J_4FO4(vafB%1(g z86pgtGC@&$d#6)}`C=zP=xtjMc*B-vM?rmk{r;!U&W-W(sqI#ua3AqJQRr-LCnqaM z6JzFp8>;h_8)%=lwm~D9FU0bp`pmxHKhZRx?Pfc9VuvbI2H##_J$G168UDFiPA}%Yq#_|+-RuoD%+CIh_UNCQ zLc>bKAkWUpIoaD=D^WtXCtiPU&EQl68vrR+&;J!m@0mcePzp-|0E)K9*oCxQ?xWx# z#2VL#Lmj*(D^N<$1z!;l4^QZoj7=KA|X`U@r5)ST6k z6G?^S3E*rKRi{`_>?&kr!WVK|^a-2x&+uou{G5*!4qORuwk)*Y%oqq3bHT|fH>KQa zWnCzNOk{STXexo!NK%r_o-E&hFup`8k~K%rv!PBqrU)zg-+3FhUac_|Lipj2Nux$G z4u58UgPzOjj1_vQ3PdnJT209QU4NbTg3YY^n(AF3dkC=Bc+e{`pzpBMWS?FRz1DOA zGY0J->hHe2KHg!UZb~f|x7}!mdV>e~_wdT2>YFA*ur?f@v$2YTYD`Q5MI}KlqVwrL z^lC>(a*vet~wXzCq;QNipO^w|~tB-^6eS41{CQqhpe3q}@+vQeMd^r)y6(*QrKHfTd zf01nv!5u*#rBUHy3Z z^0YK*Khe;;l9!&oTyJFrC1I+Cy`KDpSa*&>%=`Lyapw1^omvt-Y3+5fw!hj?*5TzG zDoeq^vHN@zs#k03u-%Q!Ed+o^C+2Lg%iq~(!-ml;5=XaqfE5llANNVk=-n{TPqMMI zQvpWiZiU&q?!Mc->S4gPdt7PNnnK$WHJWm<*O|_jviJj`=eXwj&4eRgl~J=15fDz- zZFZ(ijI#NhP1st4GczHyXNu&|5C47BpmS}b%@OiI{!Q#)Z;ww*>~X$QkSp@q3j%&# zg!NdCsIoFL3i^AK`{(hWg#MmTmRCq5>k<*}yt{LA-e`3_nnD#sol<43Hf)JZk^iz6 z6@|ZCV^%1~otcqwf5J67E!Jyn{^RFQ_nW=8(NXrvvg5TTx-Z^f{TCjU8;H@D+<>10zdsG&3^s@v!eW}$OaPtMr^(Xa{~ZpE08R~__nC7WS=ggYb@Obo z%GuHq9~;|eZxFMeotc>#8alLQyA=zxN9Y#l@6Q|+WUlx{0gvap$w{4k1?t29{qEMS zsF14GtVdZ0y&^7rBJ!7)i%Z}O=i^MlXABau?-QYZPk>aLF`9tq^Jrj-t5R1}AZT@_ zY&vMi7gdI8_O5QGN{QeOexLA{!Ny?l?VDxRd?Aq864tNGJVF2&bkz$%1g_<2^t@no)!%@-?2czfv2HFBY zoeqDPfBV=8O=L9_HT%MNy4a9cT`gtoa331!WhYd#1mz5{Ug|bEYWFxZ$S{(X?rwE^ zf{S{%=jlUTtk>xX3Th7z59g~fo0?EA3k)&j^;@D_b~c3+LvMm7D^{4t3+M)^#}+{R z67;~qIypV%a2?b)67s|QnBj zFpDs6B=w71005^42P%4+g6R`1%!Z8)(5$aIRRL6tLAxSQdCsrTe+SfuTRFpxbJNlQ zSI6RxhX)5C9~@jiX*WH3w+Kyb)Py>ph?Omz%yh zLV-vyil=mP@N~APB3+$e#7!pB9=Jbj^~yYOH$SxnX-d+GBPr48pZfiF&_bU5{AxKc zDktaT(UJIsx-SrFYl_D{5Z;Zr2_686Bl&$rk_-27Z?&l^J>P(0tk5#D29=nn$?N;* z*hJ8`c_|Ns&5~GQMz69`)(r zCVi~9U^+Kf0QhT)72{Q?@rp#hg)DN?2kZ9ueKonkda}pwik5&h;{ws7*nq?{Tlaz# zTzarY&{Fa8t|i6IJxKD{ZFQ=Qr#X_4lDfIMeYLdgg_5MxLA4d3SiQXT`1#XRMz$*m zUuClSV;?HU}UoEr+Vir&m zzMZkCp+V_vi8TNDkHC3(`E|#W6UQ4S5xSQeHZ~cLe-K<}IH{0lxT5>XoPn>+zU*W^ zaHY4{nlHZc9|N0>@AaQOY8ya=pl=$Ir`IVqUFWmRy2G#L!m|ezfB| z0XIn&{{2&}kkuH9!k>SC=OamYI~=poc2?1mk;C)y+x{==-mQ6U3{kIwl9F#Ss5Dk*TSmD*AT6W9l~VE=o)mGFmiq>Am?S2XyH=<5Z7|}# z49FW#RJ*!-9XC4A^nyY{G+G_Hv5bD%x+CU=NSwn~6BnUm9JaNky}WdlYgqBvP0?^( z!G5(}Sl|VsX$WM1l@AM zMCnB~5yWD%Z6_nf`0}`l2r|-+YfEK)JjEZXQe>IwNO@w74Nd-U!l5AdR`)#pf#k*p z&VJj)e5urD^z%Rh%{?DNL08uZv}_Ia#N2RUD8e$sP8B;e_?xU}3+3s{tE-@zGayTx zng>kyqmzRX9_7bUW1=>HT`}--h5cYeNOpX?^GrOaIn$#@b(s|_J0xz+3#c__6N$O4 zPtM69evXfhW*iBK2g6T%XELADuhbZt{k4pykY_*%GzJJuZJi8>xJL+ss+ldc+Q4Ve z3F-}28QGMeR-aYh%+AQvzwi_uruAg=Txx5xE+ZB3xjSC-5qDO&y0)U+oIhDAXGSPEfuLj zHcpg6U0PN)Fz-*d&w10$P0=Fu1;wjWcyzFAVCRSLL)(|nx8&ny^~=rYDEpG9cQ;oK zvn8)%sfbBbrJptd`#zm&1}4cWghLVr`7u5SKQ_WQFG*Eac0ZkKKUfd;5hc;~So^fP zs+uxtvw1-33;${mbVNV2x{E&)R7i$-mp@7M@sETl?2+Y1{M(^6bbAICxJalJQ?^QPJ~o*gs~!jDqyxaOohPT z-%K){taxDPhEbTtk*&jbm${^#QCvxeh##ElBd%$o2zngwl`3V!Y7akz>UvVwu$`R> zxx&qJf8v2^bY2M%L8a3o&o_wtK_v&5wl6-y&BNo9&xtRIEa39XK-JOlfl-hS=>%bDd%J6_a{RZb&7TySj>uTjYn-u`r;QMWDL9;R@= z`Ic=LZh>0ss6 zAZjZ}OY^r`GaPFwRhH#T$%^=pNj`rq!XjKs-weiQ!+3>9fsLa0dOjsAVTE1*=h@fC zrb3Fy>I_uG=`GxJzuDMq7i+v9InjEIJ3QV0G@OgSh07rE%Epqge;-?@d5z(umvRxa zy%WA*&ldI~C?Y?2zdKvG1kE>Wbo;nmY$l+&$6|)!(Nhk8rKD7?F%VF-4-SQ#8R>a< z2YIx6ZjH;5`ypkoK3m|47ymcJwXZ#Mh8Ysp7pSBW(5p|RHE`>BDT(Jv^6 zJ9ap`4+94?jUU`P#`y!gva#u$QVJZJz0U)_Wg~OPnPkr z(x58Wwwtl*?-%n@8oC~XRtMa9Gf(fw@h2yvM!f1;qjrdS_^Z`9P4UG-%L#o;E34U_ z)U+izjC^qc$S9?G5zH=x2iLPzQf_r7+VCbzlkz`pHmF(-%rj3x0eOmxIg2%hqW)dd z(*=7oZ9blKJ<*G^($fuEo&lDB$=KpBPa1vD5K%H+nv5-B~4%acF`_Fp)_Y z1ng#F1@B=*qHJ~bX`eKdaUUN7Zzqnu1`@eQ7tp$At;&lTRFpoOBCfX4U_mOkb@6)g z2YW2_9ioAbR-8B|_x6|vXq#f#d?Y(+2#3UEEyTuFFDNXu z<})d7dbi+B1CbBrL#RBq)j^e!)EaYu`yIi#cWC-6R`@s;5!tZ(*1v$$DGfZN&N%Vx8+Sc5gUWf7y zCN`%)eJ#v+O2lfcK0i~~t5vmq==OKVv?4Bq55WsfWi5WHbjmwv?UsO|mEh~bC_nfH zW~y)T<(6wVG~!6i@vPu-yX5uixD{MV|6@qpqFld~>~x1di8HRj#R`oOmNsROL*UEI zRDyzBGt@qy`3pXphwoIIrJIoGGXG`UP}jsVmlg#T?v5A5`hibp1)a{`+CKkF(Gb&W zyGvo4?#bz)GWqppvoS)R%-Np`nl!UMxm*F-9@_;2$G~Z^ zCLj%!wvAN$BoChTKmiN&px`_GBn%*4M7$|on4cH~D(bUUc|kpj8|)_@4$lHEtBvRi zF`qk{lS^I-z1Et0sB^TdEf({lxAev#3-u7m{XfeLAJYxG6U_6*1CtBPY(qqr18qA7rb*>U<@bD1p z)EM@T^K1?HOnZ0bjKg0u;gIPm;p5>-G-PKRrccV*hCC^T5yg?Og}4UWN!SNlQE;m7 zzi`y)YY@tn?Ii@mSy)=`?Hov+sxe>q_dwQvnRd0Mz(PD<@7>)iF|Y|7+#QJO^m~RY zMuocAU6`oUb40>I%$*eT*@LNO?fd@r_KZI1Dt(nSvSSIwH5Fg33@W8OZUTn@PL$Rm0yLVT~*8y*jE-LORYGIHN8=Ma@$WhlU}~18yEO*5C9S-#d|)%DJ}LaI5~*fKK4p~Ja%t)FLf^>igu z<8w=mDAzPk{cG1>e@bF@&;AJAKiX4qwERFvWMRqB+bCeL+g@{KbI4enstzimIgaK` zO;7&48ann=wkIMe&>rkC)o8UL82ve9LjIN=(fcQmmFqQwb=V1Z_V6*;2PAVurnAgs zDX}8UzmRepu(7up`Jt<_6WbD7S1T+8Q_H`~7hGIjr3+k}561vUawU7xIR0B<*Q}wg zK=@3l>g(%T!AD4VI1b{Ny6+>y$M+~;aqY%;qWJvTQclsBON2W41H@>Dqm_ATO3k$= zpN>Vknvcu{It)A?H_!}cPi|+wmt$1@3Sgv)?&XRq7Vn=}#YdDBnRz66cxVYVzYKUD zd3yG*q;r1}>>dgo3hAw>IRkle$1UbQrf6$Rj1ro;B^hRyGs zySdzk-;|3m&K33ojL)j7=iNT@T3yPNkEUu;}bGF-y9hY=}-i+SUY z8yyboS~9U7VO*W(bdDCH3St%+u-lt6*X<5)D^Cn z!z3jRD7dyKp2Gx<;&W9Q=byMn%kYUX2wPx7a1j}7A zTRP$H>>jE5HOxv$Q91vlfrhp4_^g{jT|;ASO(YzRxZPv2H@qliOzn+{b?r~mftX-C z_yx%aCEc_7xI2(tPOr*KIb`5FUFD6}FbBXe4uw{FNkmFRJ;6l`&XaHHF>|(S+NYrH z5JXV=5TaQZg{mj9%o|%;S(KF81dxcNqeER5$#be99{n2VR%KV)Q!(%dA>=!YS_V<)+kx}(iWU8yfJ4P&nt_Ib;LsUtNk!e6ItTqQ{ zYn`~n#Hg?A))YW(ak}DokwE?2|D~wc^|0LRqNI-zsThz5tSURFXW3r;;xJ-7?S(`w zx$6`MOH(EE%|*A*&do*B$V=i2VpaVc-_3t1fmhoI56BQNIX#z~CH&W@^M_>}Ib^}r%m9EfCk z`?D~%G_Lji_Bb`oMJz`QJq#^UuT)fo8dVVI11()otTYlS3c*X__a;y1#KdIFQXvn= z6!VzFA3suWkNFzv>kGH*0Oo&5h@XI1m^2CF1It>c&+*gkSVwFSSt^kyqTp6&zRSmSP{XIza#c&n;Cq!xgk zi`ZXob)?7LgJesK$B=zLI@rnj{@t+WCq#Z~b91vf8{x^$fl_-$xugSIoe7B@lzLu` zVz-i#le>F0vLShvA3roki0k;0+-=Nm$V466{h z7=U%0FZ?K!YSb{I)|_5Y%@*<%j!I6&A$DPi57-ejE8ec_FudtSKpHX9$)(7n)s)vN zjGOZtZC?mQODrv+pHOYq3vk=`sR{=onS7{}qwObv@J!q7{inyziib67ARaD8-c$`z z6Vge7psg&rGs3Eoj8}W(>r&n^4t}IV- z2=D&mGc49G4NIQ6TgmjHKz*{N{zfZRqc#whax1`PFob6K`Kiy00gQ3QV+tv$p&5$q z_TmyA05g=e;E}g+0|=GI8N;)2VMmSvJQ->QG56Sb&Vbsy1}KDo4#Z=?Sq4Q|3q9@S z>3QiXJesZ)^AxUA;=VA9xxIPbV1YblDj~DZ$sbXtH`{vNCxe>Fc^_yGN2FX~3z^#k z(w!@mHGW#eo6rx@+KX)2e6GbxPVV)(*1jQY0cVGMOQ3#j@WgwPX2FXi7yJ3!h%m{( zLNkXgnZTYhT$`O*gjCOkcaU|VLQ4odFT}2zd$#sG#RQ3KH#JumBoAPv@vN>ltZn{z z;Jk#Ac@XoTKck;crusu8B2W*+<_vjE^*HbzoZX!-?=I(E@Ud&;qwuN^ zCxBXXAg4%Y*H{l#&>w>|gD=2E`9UR}*Tz*@AQ0_jE2L97lN;VJqn*5C#i;wKeA(e= z<&V^hXnTT5gz&(Qpj#uN&BpnUnJuJ;=Q&2AQ&Ax5&q?M!pE=1l_@Eu~>9Avq4@2)0`vq24`vFA#K_m(IV8r*rfB&=9j3MV7g^muN;hp1au_QKzgE+C> zQdxOioEUDn4!&E+5x}U=WLE1pN1#kMzwpVuh_3g{x#A;XOncvx_3d71N{Di1a@+U& z+3$w2PY~ed%B!1EApj|=Tfu$bcEIIN4*)bRHSD;to66(3C1rVPJ!4R& zFM*Os7ERde|J7oNYx8Q|+=JL`Bis>@VCr zVZqcW3;J`Sh{QV?%I7v3v3ao4_2}5Vb^lbaj;hDyl!39!)!M*mot3Hfh={5)gl1Si z6vXL!yx`_Ur}2&4vl|8~XpH!pgmAOArmx9nCFt$!;NawRJGH8nv8AaY_$-oa1C8@? zHhzeO7a7LvwOBC#H)Lja_w;z@req#m{VmRZxQ#$+`JE4*_z&R z5Tw5ShJOgc_H?eAw-Ez2xDLyu!S;1CJhY{UO!JGyTa%}Y5Qwz8y4pC!Xta_9K->Z< z5e5*iSr{fmkJd0{+(mF@t2w_#{j@o>Q6?0DJ=g;Bbw(Bz6C!&?PvN%)_8tKF!g8^C ze+%gsB*Dxlaemnu8D!O8;1HV=p8QT9H{daf`r9rb9*p3PP{|Gyh2~}j;ppVH=bHrB z*dvxM!6C#g;CVbEg$me_$w{Fve}^x&%)`-0-OkrZr(^^@j_~e;HIKn|US0yFD&)s2 zRuseilerO2{3igXH($Ped*7jDQ9{Ei^3OlyVfeRLe^N`B8Xvz|ZXTmb`1+G`InQZx z^9(?ll+s_eLf~7YzJ2SvL_o!8t{@QB1gO@R7w8&;_W0~eL8v823}yt}hw&{5#~ zGbx?=L*05jCAE}PZt&NIE-%HBYHlf>^D@U=$oumRL}Z`0ox{Rk${8_aB5$L(iw|uC zfF`BJ&_^<;%X6!0LH~Hg#gWg@mWKohQ{IE4->vM0K8iW?)?q^{=QT%WwJoZ^JfB_^Bj@qEJ1sGTtdiGaR6KG*DMlPDVoWZLp9fL1c-qA z{J8!owC28<0Zu`><+ZiK5U?o(crRHkEU3XRaN?uJ`V7*l&TLRE{HsVXC;!_Hdfs#3 zU&x%nOYG9lj&?Cmr`AuiDdSN9RwHG=vfzTR39Dgb7GUiSKlNPTRkfV{(P%?(J=`udPQ0-O?05`RW| zdVY%}!HD{iKM*d;K&In&5#klJWf=Z@#qOmO=d`ht9A=>>ZHq3mP|~h1im7}&_2z(`4WnVB;oGM2;asFdtWb` zE6o(u(9faqECHZ#wRJ3z(yiA^rD9V5s{dnW{RU{D~AYe!OVkzWqm(FARcEEwn z=V(+)-`sb|Mc+n8w>c1&qQ-!gBkG2tCoh5obcaS8R-%@P4r@gumHGVnAz$hl2bX{x zYCpIEe!w^)*08fu#^GeW5!IXeDX{Cs;6uX^(K=(YLaXD3u-%gN1vam31xY$xf*ci< zG0B4?zKH$y(R`FiB*Y43@ zVj&g`F3te%-#C?}lKxsXSHRNLbh^i|R3*D(Z-19mxUChCEX=vmRUfe9W(fP=KqQi0 z-PFHn{XSdkg{DGVGLS%ly1fMyapM-X^AH?BMTH9J3!%t~9vJe*Hv*g@wC5yX!%Cr% zVR6jc4%a$cb3~BSdF@jc6xX`ElE-*#mYW1{p!xX)1OQOC@&IsvUwHx0=&zUzxJBEG z*ZVObs?jo%Efo5k`}@HM>83nzn%)}sPT`$Z{GTNE^w0BoUeI6$uPq`RMH#FbCMHnP z$;U`>ReYeK`4M8+H#jJzF<#AbbAuqMhS>z?Rlxw-+rvwmt@_lM0@m@wx&rk|9&BmT zi;AZBAEMpGveh^M;-Xpa4+mwku_y2!%^vDFp3zTw-j}EHkO!}?7piSH-w#*Ig@jr1 z8w~88t*CLsVNlA4V#!XHyYUZxfr=#Jl|2aoW-(yzaOHiC+iq!TdgvI+|oIyP{05VS<9!FYu zr7hz3^l-8Dp7&*TVUgMWd|A|U%BfA~{%q7U&E!W=2r5aPy=PQ@uNKhe9FCf9$G6sK zVY4qHHkyt17mzndH*yA3wA${AWea)`i<+@`()ByGpZ6N-E$Y!|GXIb(PIOrF*04v` z48o>IBPJV({PRXJkNxj#h6)jFZC=iF%|!-MWJP9ppNQZWUlB;jv@~RvKB6MU=vn8k z_OQZ24d*N7tz(G9I;+dSwR!2xt@ROBQ16#ba8>)44-~RT-z;kJXO@@Ylm;4*V1%c> z<;-DK;JRHqU%Q6cPr!q_Y5(^Y<0CqgT z@4$FeNo@1UcB$@g(U_f1lM+dS6xTKm-)vwz6YPhF#FkpR#}oX*&cg7)2`i~)+{%R? z#O}Cg$4o{vb zrWAIg-kdvQW*MHeP;lbcuy%`R=K_Th>O~FKKc%)Ok(37`3?l4Fh&Wef^-F!zzcVUv zh+_L$!&9^`XT#nV6_p?FHzq>Ca~uTx+{`8d9XUt^p9u3;Eb5oS0eIN%_wO1fCp~aO z&;m@xw^lK^O3>e(&H!^8WJLtc3kq96YeF&?7t914wi`nk0BO{RnG>)P|1bThy+K z*LA8GShbi@CMyBa!9OK(v`CJ8&Y7DKDkDnY&oL`}--aK9gNiP!F$&T;aiA|YD05&HbPdAR$eWXkwWwINh4@6VL67nWtUDGhD3bZp?vU1YxAXkn#fqx9Avzjo^D8$^ z?|K|9gU;^mgkt{7K+tno??BRq3+U}rG3pGfko4| zj*Nb1*MDC+BC<0Y)K>k#ChY{pGy{S^!G8je6!BR4qQjlW#-Q6KEG4|TpGTz>THB2D{ zVa{8IJ&dIF9lV=oMiHMA7)M5))~IPl8d_W7J0E#4e)^3M)b(k{-ZB@q_J<6-l_15zeL=(<*qDxF$Wb#*FJ0;nJ_Aeql5`A5w370JZLLaSF|d#6Z8 z%dZgW@7dPj&Q3;hvV#Md zNyPeoi*QY8bdqbyla>Tp)<>K+}$faBawlJtG4Ev?hTp`?WFxK~HqBwm*V z0Cg#uaP{>&?@omxdOdSeaw?9~=wdSEEoREf5Dej)tf%Rin1qFdkb8W5o?C1eAKzZ} zLruri9=}NVzX3Yzuiv%~Hs@;3^MIZXnzu1ba@~1Pv9T%%38vB(o6B@%e_kl*_I$y! zJvsEgzVJ={5Bmv4nUikWyIWg4zCQ?%T@>qeVZ5&n?@9uy z9yQ3cT|k#P@_DLn3t%KE^WzX3B*8@T4dEzcg!tI3EKC(M#9Hy3V_j;l_1{_1lXRUG zc{6nxGj!!;J25vxUDz!fZgYWaX^)zz{NG@IC{8T zof_A=QKD=+cQ=`P-`*bfvvq?_gIVc+fZad)xs$XQFkW%uz!0o;=C+jIWFq+W?Y_;g zBt2oWlVJ&+X5X$|k30d9`9E6H2f{3p?4w0TVcWZxwa%8X2(d+%&J@<=z%-*nbxmi5 znfnd)#ei`@UUDRsr1o5j3YLeRpuUY z#Fs}p@Us_Q$%GiGHzjZT2Lb(S7T!!CoJ0WXnl`U%qWmjPT{>;<8LUn> zi%-~56Q-9F*Jcjj*Jhj96hKW~Uq4M@frEqd|;UmyR z=c+V}tgf!ci5aD3WN?1@+XSF$LFvHEE0=*#8vxQR({DBUtl6eZ(@Ea<$$AMoBcFd#Wfvfgtf-Fc@BQEfWG%c~vmGLZB7+STUxO^lBt>~3oH z12Vp^NVC@30s3M()nfGmb~WubGpwxsa!#FqROxj%;Z<_iL=DQ92T;TTgTnTl+Q7By zUQsg2@aTkpYR^sBRX4_~P(1ODs8CJcH!)wI(NR%Q6adB3x~BU%Qd zQ~NZblFlbU4UjKii{;{+Y>9wxwMfXdzuw^%BM70FEZPk(>UAYMsOo9UUYKy#CthXQ*&D9!U;o1b zMB++ad4qNT6v?|>h>-}l4Ndm44gamXUYZV}m=;C%X))i|PIi@yCFL3!SjFP4HSV0N zeK)}{*!<82H5}FB9|RO>{#Or}nG+4I9@dLJvp@;1k|VfWY1Pa4-D+4qgah zM#czx-$3{VL9-bNmmdFd6Jhxpy*;D~t!z2_i>IDLEide7)To&_$lU#PEB^>T9FALWl)zm1vTqkr6&y77EzSgUfi7j<`*~?fi@5pnB+XlZ< zoX3)gveE2J@3GqUN8WEmxJ-=0%>CR#6-7s0Yq2*sHSPZqNe&X~^f=w!CvS7yh!yE< zmB2$K$z$C?AgMPQt{REqXLai?&s5`AEOq4?PIn3ncR~dP%?cM z)igB~%~GeqG3s!2xX%eNMHePzPyxUI;4IKlpOw=-;5KKRYDyc(|zRd`_P8ozBnia%X zP{~q0Z#_8HRS)y&fcWvqvE*DwR@O=m81e;ECa;{?oaa7ZzFbKb6g!olT*?m)wH`Kt-rreD+>#$;ZR8&Y;6_S5KL^15Eos!O#kp`m@mJ%FA|-J8(!@_GDG zMiPRv2?5E&$flCsvgXqgVt~dM95rTfHC{EIl%y{8d2LmsFY4iSbXPM8^Md%{5snib z0}r>jm~lUxmbSN`z{A?lT?e198Xftsp||%kM{-GOTAGQu)rRyC0&_F5M90zGr9u1E z7#9Agdh5|0bO4=_uvnhZU+m+N&L$wmJ>1CF&0d~v(p z>y1Dq7IE1#06`57loYA@}yqxTPy`bz>&sgD7XT+elSwYF-s%a zx1*4dCE|Il0mqmI1jkxb+ni-}dTCO#ru z6n1FsXOtDWa2Il`HuFBpa4c!Z*=l<-k#=@s)?B--s_2F$4VJpt*k?|yzZQh)hagu| z)uC9!em9v3i?z=4jdxbVt*x{2N`J4pOIP>c|c2O)U)fZN$@Rt;46l7Zw&}74q5&=-Dr=`i!)8jkw1`lE!4R1+3_oWtBvS3CN ztpSBBH1ImKU2Z-Jh77e`?0WwWQ@?z9)Mz)rySl;jgpDH+q4<6jZBumj%}q<@Xw!TM zT>_HYwbgS9L0mLT*eeZ$t2R87!n!hbZ?<4j{m=97>N?x-{Sg76MN!Zk#M`GNS_7_d z^fX~VOM1%?-F8sQyRC`~J+*T?8Yu3&mW3VdjN~I!dI$#C+@Wn0y1~jh^EJ z1>&&J$1naw8@~(}YyYxH4cbAK_yP(hEs(B>)p_6crT8df%?83<2zlJtuAl&aaT1!q z_R-Pr{cE2)rUYDHzZVnqTT&ce1^1#6U(y{Z7QDS4vMYEtG`5J!^u(kjoTDgBCgn-F z=K4lz0;H|RsZt6nfHOk)jE|8vGoy~M*4ExpT3j;4L2pd=qgONh^IT+&)#7d4@84`m zUngRg?4rh?7(0+Hfg|6{1Wwb7Z*CAQw}lex#Bt)tVb-&#AJiFjG>;4qpM z!8f)Vur*-wnnB|(mXVHDS zIUEEmEK=Io9bcS4DKTwJiHnB1nx20jC?(l92O;S*%fZoX`QJE>kM>fG(`uWe1%%oe zqJA7>aE?3kxE@%e-wV#&pJY{?uJCc3o7FNk#dFO8{VQ!4869dIk|c*&;HK3uMZ6h! zOpP}Jhljs*bqR|Z2TVCuS6AiB1;PQ(BZqBM4yUn%QRJufjekGvhisZoi){y9 z^SRk`_QI0?vX=Lx-&#ej$wcD$`?<1fBNY50Y<1X4%EMueK+KeB>0G=SV1K&7b%H?@ z!_V-*G~p{FN~O4kLX2;m1&AS#TQV$oR-cQ6IvL<0gQLRnASFaGzDa|JGzQFaz?R4t zC!dA^bw~Mm24|?Gsq~zufJXXU8xBI7*!vCYaGqO8u+QCz$L;9HWyk`=ui>2OqC>$% z1~r`g-}V}hy>&4T`wXRW*^;}zCzePojjHn^MzSB(#ki;rc&1d{ZN+iy% zlBFDvVWf5du${eZA-+wnFMn@%?S;Thp(JRnuA>$tT5R>36AI0}pTb_0?--4&b^{@$ zhIoL}BrLU1bSuTHE z9CJCqgzo_^oW>rroU~}UEF}v2+@J7=r>77&A+rZvZ@}IS z-NX=DbjN&95uGF{2aM+-f~*G(^YIj`!jBPM|8-qB^gjRh0pOyiXFk-TvJTwb52b=c zqRA(6W5J3{+htv=n`+|*jUyo;_1_cZVj#Ph0m{5U+z@e$o$i0zThz*Rwyw-{CR-+{ ziPK?P2n$FO0CRLtrwdi>qB%Dey!E~*F(C(BD%n5K!j3T2&(hP=@8RzlcY7b64t24! zut2yWi=q;76LDEyrMu9VZJ8E@_rY@lPUqLHLk6pi;^gE`+w6T)%8r_j4PAiSKs~LB2DXPGlzfo$Ov(b5><5WNA=3jf95c=RYM67smVZ=m+>Ecz0bu zXa)aqSh-W4yJXdCy1cw(l`ql+Y^l%>;(s3Orc;fKM3YLCQ(ZFVnlwE>U3_DlBxXRl z4IvH>3sVmgX~H5UB_zBJMw^q)_dQ%5CE$IueYLLK| z4b)pc7dm1Acu@XsK=+WHZkUi!ieaBHe&)CNWFEbM;Mcz=souYNu+vul2#F5?=Kf3@ z4J(PE_lNc8dVytk()wC4fUnpom2$;wGtqyC)L1mX;ef z2T)!Nm`?Xo#2$$UGE!`y6j)sNEqYmewTp*`v$XuDzcA<>GB#gE5)5T#k@{PtTiCD7 zz^Kzbd(Ajbtcu?Y(I)OqwC0w6DL1#wub&&E|9H_Bp`+u-F9A>$*)O~1r3#z&i~>Bi z+u2;G%d{!#ITP5P7@flvLbYMCKCSv#ZwT3n8%_vsL_wl zgoU&P)s&8+3WY*b z2%uI#k^>5~)xCw3_4bz08k)q#Zt(^PQ_a~gTjA2Y)b|)Va zGBO}MI0Tw%?2o@m=NFP3{S&Y9YZ6CxfI{*uNWz4yU^^J-BY=j05DbPga=p0sZ!nPu z@1qD4I^Ifv+p^#9}*q{^Yn^@ z?Etd|e7pN$Vr{_tD}4NNtAoB6_{YQh>p5|JBXuFw-Ps=%O_5agY+?cD)rNOJ9TwK# zV-sty%VVzFzjlqf;aCiA03`DzPL!ai186J251xA{et!<98NRrv!=@j-$T;$LuSRDt zRS^0Q3&_k$@;sPY2ecsQgq&EMJAjR;+lLq@P=bo_&Fc7Rx{$69Rzf}cE5b@|Kx<7% zAr*C!ZL=)XE3{cX0~d24AnnN~>O^2d;$IMbwj(sy0VJez!yL*E8y#mR=lotrN;!z= zS$Y8Xo(D~-ofUc=*Jw<``;r&;krcisrk$aaLX7hD@RtTaAxRd0vs-#tJedA2ei`Ud z1Y$T_?e;pH@jBAks|4mM$2KQVv|Vb9Pq1=w&fcH>NvOKQ5cPK5SYCF&If$$y7yi}h zZVx?StV)GWwk%}!qYa?eEv&58I=q@)@Qa_1y>!ba&3ofS^$FHI9$fjU&-ad>R&-Sf#FR}O&_>tm&EXm8b80wS%!g8d z#xefwb`}YB2YmsZ^vZ96A5hvhhZDB4C!q$-(n7*wn>+Ser5dpiZ?M`jEYqe&xBPj-JCT}ESV$f zuv|oo<1_qAV$_06P-+WuTha*xeFV{5G9EvQ;HGU$4FT(Uy~lHV#N+hEt=D(AOO6I; zbkeX;1or!9BX%%~T}T{(S@jqhk;j8_@y?)nAiF<%ZPhu49E^y8EWKP4;;S|6tV%~w zWj~+9V=d$ZF!_1NWiWU z77SQ%e&~l5hAP5)&L#5_T||%Ia*71O&rjW+bdcc*!|f7>A|i%-9@*tAfC|ij&%p#A z3W(I{+R61=mUK&rXG1T#dV6PBqLp;DT9>OHT+V#*ui8a%&o3 zETDZ^ZoUQY_SDvJT7tn$INwG&u!z})cb#s8SjT?!HArjbjraPwc6r?D*qK;p+QRf@ zAD^E3yuXo-$^iOO6F^IIy!c*z?D{_#ff45z3{#jcO^r!qFtCZB09oh1{8utG!>{Xo zKQyvo06GNFX?FHp5ubNI%)4ly30HYdp5#}Ge?F*g6ZqWmW!n<)>I(=6U}8?r@)RIZ zBLr(QgD9mjfZ6u{{FE1)fC&Hh?-KEP3c#+Ftkx}$7e!N^Tq-y76}AfVzrR)>B8lI| zk-Ta72ds?oq{{QCBTxqDvfu%8vuoa?{ypVCXWgb?Jjat)CyJG1OYMJm_PYJzt%V#` zCyg8W4#NHKygMSW2F4`8wDkR^43%_X(`FC?qdi7MIMx4LA;Q&qVHRnr3HF@zjK@Uh z{cr+I#SIj^-Yq$7S;+f8*BeYBcPd$q-LUkdbI*_|bd&m{J-~oWF&($eEL(<43{C};y3g*79+kBqzaCc;CU~Tor8^I^@e-4cy-F^ji8SCkxM;M;?-%(-IKBiKxW1icYdPlQ<`|p?q zMR-i12UP#v0~pHna_|kBNa+7fa*k3K{_k4=EO{p^&`a&VOVC65Z~cJ}G~@g~y~+OX zBdD*h4cz>@{)1n>INvyVaJ_f}^IiG)@C(fPNGD*S40{JuQ&RAN@);qR0L~i$Sm1b~ z!C_PA9=)f-w2{VthC6t)*v$JLcWe+fpRwWLWB;oAyd?|F^PO7-1{b^ud5{A=vy1ms zfk8TOaT!R)f^OE=gWqU*7ZxtM^YSh_bKiwsZ*cO7`E>SUDg*8=??TT<%5AvBo`k8G zGQ10Ljak%%+7|Qbfi{?6A!VgD>X$N$z#;8z0kXm{}G9wcwp{%t(gDe z{wdHXV(f=buHWTaA$|V8-~8Zz?oAP0zgfMufAi;qC7WLr0U3cv_ZRi*7S7If#!r7% z&ja3=X<12mv5n{~EiEVYZ)n|?_{Q+E1p;=Tsyi=4#CiikK8#_lKEYh68-qb?>~eA zKZC|K74rV%(RA@(i{Z0sZnx&ppxcir$PKo47Anqgb(Tlpx7djQQAAYzwg=mCrMKw_ z4c_-x{mGz0F3tuZeqq2&qyD?Ta;SrgL)Y8s8Uc?kq-q>S($a(QJ)S=ELA`lbABTR; z_T~r#q(5E#w7Sme{hG%Ygm$W5Vr9wt*zhClVX)RV_;~(wtuTr$Yzi0?J~nGU0Piw8 zo1C%kA@=s=^M+W^o9NtHR21!alx?x$F&_31N`w$Fe4U7wy3m}>u)F;^x4HMN&hDbj zCNGKCUUOq$z;`PVx82Ljsn6>DdGB}f_iBA^^A3!Ef`T-57LZXO1kcgT%y9KSZpATT z%koqis~$sAN1!e>yG>Lp%Ids78NPZ4&HXkK={Q+x$Pk4y@U@CNTj0W0728xBA00P6 zd-UwbuUxdBcKcskeRWh^OA}{;1b5fq?gR@CLvR}gcXuaPaJS$Z2<|QcLa^X2!JXg) z3GM{;Lf+f=?Q-TE{+N5GyQ{0KOMX?=0%FiS`E!)W4Nui3?>Y0Mj|4`6)XC)BmRj+e%VT6TM zxu7Txck##O;lTSY#)r*Z*~x0}R@fKlnGo;?ic^J0-Tan~w7L+4r#hKN1aWV@ffH znrO~>V8R;-csYMysmf5G?WRlixe3f`U6$l5Dm{GM?V=a{Q*(J7&Iyz^2dWVRJsjfU zE{^0O+>O7N9$f|^3O^D7@j@J(IG{7vLanCp_Dd)txA+2Ya7b0-&Li%~-QgI@)SSZi z(V6X&L+A6I%?{sbg*69WA0a)AIQCEKkzvs}iC_nOmfQ73nvZ+07azw4#2+sYW-Qm@ z?SsTH{HN%?dLdKfjyzKRET?X+1Px-^m@yTTKBfs1c%!tEYIqbPvjYg>}YYF62db+u_aYY*P z2W@Qz8#$%y)##D3D-s~G(ydtnSt*k&dUUPKVmJQ*6mSMK-~W!*$Ia^!=jCr-9({V& z_0=khE>lnCe%iac3=gG9Pi5*q>O2C?xz2RrnRX16xr(db0Hw6`8~6GKHY5YYQa^tN zdRQnpoMPi`(P5uP4fBW!0e&g&s=LF7F51<{+D{%BC6LPlkDv#l;1CJFK}q{s4JHQ! z0Q?&8aKj|WckZ&+&5lMIdVjsBwdV&kd`07#C77i7dt2|qJxluh!qe%-cGAW6!)8|D zJZ_+^rrFM>fG06x8^0A=XjcxTy|a@7j$hl^KRde`om*EG3u(xA_0gLUgPBtYYZz~1_eDh%YgPT+dh@Z1TI&3Y97|3$SU9J#>T)Ss+_6& zOH$qjleIpeEju^YR-?tjZfCf+aR^$-pD{g9_5113UUu@O=vfO~PcRjb5B8Ri#FgHbd z+qG4iZry0JdiLi0=F@8X{&cHzoYL3#cPrF@r6go}0Ei}B7X#j(*T0Z--!`PbafJ~e z%eR{Q=@Z+6RWt^KNyy-H@y^koZgH;EGM?WD4@bv9(f*PVVKw{BGzZ^5K23pvB<`?} z^Ea!P%XfypMn)4r^QBMx9~iFZ^{5NVS5NJ`Zbd;J{?<-%yhABW*S{87zc0hJebxPd zR-x4_J$pj32sj8!3B35K&qmnyxgU{Af?3s)lS8lUOi5R*$LFqtK-;8ga@_W7r>2oR zzD@3J6%{Cl#`)?(0?xlNQ@U(f`BPf=l z!9Wi%`MWbtEM#E`gAKLBl)_kP_+wcaf!K$LAd@j16|$NrG%o?YamMVG^(!NMTHC z5AYn=uCA+PI3$Q`L`bpkFmmn*)`S5!uW9p2^+-$@9mreZVg73VA^K!c0wJhSi|Jbt z;c}2Ix|I7(g!?|IFFCcWa9HQ(!GQ>DYL0J#dfjtzSn(&A3a}(As(C)XDVS%?hMs4d zu^2hDtG>2&=A!K5N5TKAfCWG>k9x&-57xO&PmzJd@+wu~_7zD~BB5`9{U|`n`b%~F zWFr(4bA5MXv0;CzP25}-NtP`{ug&WLfTlmI2v0H>9?ZVtAIes&_h-ItJE0uh2pB7x z%h^9ELm%%pu3iJt_*cmIGU(K0uND{PulDAT9$K{lr>wk1e*%ybp{b?u1Za{%0VyM{ zj|d0-kUb-R{!z!32C-e`KIf%5dRhhIHn$y`)`X|phd*|HAEpB?NFcn<@8yscn5UBW zPQF2f4~RMY78rJ=U%!xe6QT1pnt!+Khk{db@wk>TX*?Biy5 z-g(DQ2AkPR&L=8P5z`5xxwMZq#OI7+`uZqb8}vE@Dus&pYnH z4~aCTJvcZxlYtu@9T}*pt1t2a+i|$`EG-e^Mw%KMr)OrWZ6`GqCPQhvbsA_eHIyV_ z4Z5kcyZ`G8Ag?Db*d3(!$iN3LCK}JgJvIe05GSjOlj+XAcfkTXmb@3@g+?;9Qo)gm zb3zb2m3YZMB7QKMVz01SH%H?o8%lUcOC=qhRu9nWTO8viVwzk|ta#na^}c-W3=RA_!728z<#*XJYJoTuxRtX}|UN{DQYD25sNyeLl}Y|s$gWDWNy{4dm&7j1 zr3>EzRR@>1=fSiI^qW3n8#5$?NkQc?lEa@~7PE-ItI}y3uajJjiaO6YD6d@hjJKHq z0DvX>K&<$!#{{wxJY{Fs9*@OTwoe`kk^0lMB2^DZwUCJl7K)_fm>XtRtrn0o6*8;c z6~pyEWfRLgg%yjlFr@x7S2p*V6>m^#ykwXc(V_OU!j2b_sf?QX=U6?5t**DZ{t6Rs z*wdIi6Z$)6>+?zq4j_^DzsLjBDRBK6ohjf?8;Wre$sK@H2yOm50#0G0xbpM!FVJ_c z;UMZd7-QqyQGH%Pi=Pf{C6h_zYyM}m&@WBPZc*h zPy@@6tWl|frE~tv+2>1a;F>E2j0b-#)NDw1v)|B~)%*Y#J5S^A=;-);IF;rlDk@Nw zIH*Ho4lrA8(%xCbKoK?Ow!8_J6C9yv6{%mL|0k%7W>e9CMe z>b!6bqx|g5DcaLU9{75{Y_o`vS39x}$^$w`609mZ?p=hIrZflj0uBR9iNo5^-3m^z z+MAMhK4OEG^Rqeb-$Hq1!LG$r+&RaX{mn}a+0zD4CEkG6n!iqA$$m*wj}(5syh|3M zu;uCb(xN zYbr0~ks*JD^pWgPH(>M^kE-X*tK#WA@K85 z7<%2@r=Ka0I9UXm;PhE|L@d~iu}Ex9P3j#hl?Rk)2Bhvn__Q&|lq%Px4a$Paq;LQ* zc>vioyw*#|fHM{BMHPUhuT_QV*=GxUvDiI4T;En;&!73RZ{Iq{;o>qIa zB1S-<+E3wliJc226Vv{Ed>lkU@QXGUtE!Kv43YfovC)<&V9`&sJEV+`+X)B9Ar4Nv zAD)=^Bbn5+if(Qcpol495*EKm7GiO|*qeLhp>g+ih>F9b<_$juJ3j^e$*D?#Lq2~9 z8A)JAp{Q^3JMlTQnNy=29fnsUu{~ROZHV6Qf15Y)qKR|0kkIY)`qH-ykYeV@ktY)+ zt|m^pT+hsZ z&U+Y26Bn6h4?;mFZ1i{~ZJ|dN&+0aULy|yd`d~?_lCiAf8Rp1n{W$Sn%qI~z;2Jv0(aHgbfSyqwPZ&6(R#-@&OSn76o$uPc zJ+uUZnD@S-h66S+n0+nIZvoO0%{fSppvdWwE%JLbC4AGwot!_{FNxB$mGmk!;?Etb zbP!&=k^nN_Oi+P)>=2j>wY$_)g4D|+IY^@FUZI0xlW{*Hy1;RU_INcw0bjGffs;HQ z^zJk0SR>-DNroIneFKD^Uuu}H!HaPO#XXe$a|p-Z+zi$?GPo%u%gD&8L`OusAVR zlPray(nLfVIjv*)XDGe(Q?-9r@0p(GQqpcwTt_u0wmpyn3%^NXTad66xv$-Cx8*=x zwPUKv^I2&+9oxVj%tv=9$fR^AAh%_TDsaiJGK|I<$hyCW$9IN34^D2_kAvl?jVqYy zf;e)fYe%n(<0yM7)UBK)315Cm`HlfLF2wlch+Q<<6c@|{CAe0PiK_TQ7XgrWC{6dv z=Y;r>SE8|Q1N(8gpynf*!3b)n@bY1m@4gm?8-SrrA}xstA-JyCzgSCM9EuEnr9(cH zcZ3JYC(JKZZlx^1+P30u9FdO%k@NTKYR{FPls5|4hIT%edQ= zFC&O4Ads-!1LlikFx!&VlyGv*iAth0?7^)ekvE~$#zuBvD(potF3-Lqa&Kn@8tKX> z3YiFo@;R9Pj0`y)A5gQh&+U)f!*b$?m_M{F7%9((>=JZlyR;x^!@?$-Q_~@+UA!d( zDkv!wIa4`p3|h1SV8t>mFW7ofKsMqNb3~B!^Gj?YoW7p-H;Hq~$%(YX!3*<(t5^7T z(n80V&M}cdHPm3}Tf#P$eTEaUzJ#6rPP~@%HAs1%IUCZ7L%r~ri)SW~+^J87hThLK z7XiXx%#Hm_MPUka^Oh+e6ARpy=ZN7BIep)M4u<8gjZIx6=QgFkEI3vo_ln6GM8nSN zxEev@V(e!!!?B3u-7{pk;~(7=qf!4JKX!?^BQwP9VIDRRypmfJlaw4*n;sBflb(MR z%RR~r%_K;Iq*5n;$aH3;l3*9oljTwT-i!lh!CjWZH`ZCS5~OUSVL^=)ZKTh^Q;`~s zf(_UQFQ1IsQw3#oGZya3!@YpjvEs7`4kyClU3mn@<4|L>F7`SuuOL#o*ljx$$B$Kw zA%h>)*BP+~!p6NWJFpXMtK7-jXyKDOIO))kwthoD6w^Wxv=)~qy#hzhV=YAzY0A*k zq5yuMVqzzlUH?D7)67d&8c@{jdS8X+EJftGZ~=ndCv5$-!Up|j#P3TmG%axep@ng7 zu|4_h>#+tCZboSh&$+8Pfw5Lu=*nPAQKw`b3dX_yjYVzG3+t7WSxX&fA<=okFU$7{ zJB86Sueg+;8kqEDh^5$-gVg|puxKgalc(FmTnV$KK8tZ|A%hJw81j3rJH^XZWM}cN zpUM;sw;)co#|!Q?lt*(cRV!(J^5-e5gyRj7J|p$E77;Tgv?S%>TfzP?^_VG0qK{ax zhQcr^*J6T+6OUxcnWGeiUjN(pUR$@xdm|dj`TXiI+!=7(u#!Zn#=BRD`KH7b;8U2& z2HW<}bfb*%;Zze8HDZNYhe`}8uR1WL;Zd81vN%f$<#L#}*d-3A*SzdG(;k$ovUi;{ zHlxrvVYPSxn6vg1N>g-&IvX0e?C?3>D#Ld>t3md_nCOH-_nLmY5r(#Vf zQbd)x2Xj+Az#A3aR3~v71UiIT(tRFRO0%} z<@@A_2!xF|JlP5_nRO~W^L7>mf3s<3z;#7J&#lutagdd-N4EC!$J6+E3MCGc&KO zUr2r2uBX*h#$4TcpMjTaqQjA+lb*R*hoV{_rx9U;AREIIO?a^6c{3NVDfcG@Dl3PL zv+CqtG(SXl{XLV(y5`H~9o@*grh~VIPRF=!Pec$B{e@dHgXMC`+GE!Jxx>oxfisXS ziUw#hN5~BCpa;wbsJG&_MejM3C2mycHglu!xmUs9Sj(SHOe~6tgra2A$mssiBU==d zU3Oh3FQNEEe|nWIbnagmNqO#b$;j_XlP)eI+sM7r4EwoREtYg!tn!kw0(v>SBn~@K zkVBJ!$<<=Fr;ZzMXR}OSg3AHIMXz+Co$<@wF(@$H>AgHrr9fdAiC%v$Zh_ycxFI|# zP)JXo7!YeiehRz+lpN3xmY$yh$|%KCHD+-`><2-2Ija%8DghUKcRiiC{B~er^PN)y;&{_9bxOJR&QockPOTiQ)60yP2O>3pV9 zrc=M+4W4}H4ez008h<67^;&B1Utd5CagQa4>r^Et6@%gU4Wsd9jbDX;fhutplk#DZI1pRIhewlzN6B%(%vWr5@^uvXWPz1HY{0M zUGti+UTuMj=WQO~X>1Y*S?OyN(95Ck5}Ousni0oPs&0kf@gnH7Va8WISJ{yoVt+Bn zy3#l?RvOJg`}?TIe7a7-N|GZaE(qx{+v+P)i53AodHFPcPpbA2er%(iZ~f%(%HXle zoIo^0hxld{1=d*CgJI6#Gw1Y`upH(5R8AK{rOld<|Hzar`~22j=`TV0zrBrZQBdt_p}2&c7%FJ)91U^$JW zk$5<1SKeEaHXjKh&F{ZVzhx~Z?)lCE(K9;xS~^W%P1h!iZSP(u3V2DO8^VLJPFM_x zu*llHj)xOZO>5X@k5%O4n5oyr#0*V>_%{UusXPkR-X#&|FDd1w%4smA^dAs;3?Xa# zb}o(lUMNa8HA3jnIEeHjanVb}O~Y~xp{!c`ezZE1*AZ`%T>J{T6e#y$p!n`(2 zncGqjH?kJT)F20T1*w1-zpP7+44K3sdPW!S;dSYGa`|;(iKqv(O52LF`0}*AdRGfD z%rdmD`|uBba80+bhPIhUJ@U!WRbCH3xe< zV3_F)@CU7sslS-Y_H#M5ApACERUD(({9PEq9wYzbKBicl{srQ?QQ<&A`GjzZ51SGr znW|QI1*@B@E9-Ye1dGosW12-apTev<(mh3K!w8KO>CP*^C>!T2xW^xuFj|^X zTN{u&Zx6_8fK15wsj|G8Bmca~5@$9Hrv#w~A$Y>e`jd)*fd8v3Re~QPT6PmV^O;ut zDQWB61gaCIE-@-wlHH&Gs4is5= ztS*_#m&T+BlYFs0M(EY-Mv)(f4I;N%h*YI0yX57lvFC^<`2p^&E`kN9gvwfw!*DsO z0WaL6<*I^wrTl55o^_{T;tH0rRk8s*3>>P4TlahuPG3u$3s`ZZA`ebzX!!85pX=J#3rP;QK^;@2#L?mG_^nO8&oo@=fMk zt?H(3Dda8Ku+=oc9L`oh5J~sOO$|fO|qTUS%OEHX=KI zWeugVI{pEZN)9)k>??w4kWK-?r)um%rr<=f^+=7#t`Yd$x1a9*MqLa@$YQUcRmtnD ze@@>Hu}l2oa|)N^Gtq_p0%1Z;Hy&($xDoFBl7WaT$+8;IvmJp{7ltUp6YotDLED7a ze1hWv&ec$m*@B6YPp!eRiRVojywP{@MSFC4>Vx)*PS%AZX=S7r_ zHu^5;A63h#x>2ytE_lQ+_qol>B5f ziMlB-Y~^JLN_xe$C`ny8a*SSIpLCNt$QdCmnMolj7J^qBUFc9{;#^ko9_T7}g|+}U zZJ_qLh{nfEVdf^B019t0cj_Qt+mN+Ep8>&9wPr#C0ddXxql6?%AP3cl@ zFyh*WaIH8_w>n1mq>_swCl%4)SVGw&8w&o3TzH*b53u(ya)e+;eMgRuXX*m_hMOV_ zR(G~#Bz-9<2!h1o#~_NrM*)@&;F~blLz*Xa^~(DbK#iL5>vyh#D+ij2M?;Xy8vR7X z0Nzcx@VdLSk*oY`-q043T?*-xPAe6o#9%jZXv5MucxT_z)a5QwPc9WF)|fI@4fXfI zkP%+e1E!?@k1{}rM)Q;{MmAR{jO?#u8O0S-dLfBRkD&qJsW+PMFI8N|UM zl8wAm{QZ3x1CCJ4&e1+4;J}V0g#8Ogxgs7vUy{pzYo+G;NBnI7g&1+u0GGmm~UixtCzLL8I1hinlSbyVN z9`-p%V`=P)aSy!f{(NaA35hRTvCbyST#ec^ukHoOju`Gs&EwpfnT8~YzAQ&-X(@94xY4(23nad8lQZ#V2^k({ql5q2l*hVU4D za(z4rPj%AztMGX2AifXgL*imx_kUX2Z&y3H+I@Lk2Z^7Xygf(t+KXFAd|~lgv6Gq4 zx1=<@NvO>^zplZSf7+%Iwc?Y(dWU44mf!N*R;jp!xwuEqY7SpU>NrxY0XjyCm3y5D zD_QJaY%qr!*A+Tgr~N(8TN>0hVqPyOFPG+q${Vka40~T42?#g1WcPWI-509LAhe;P?eYCZQ>3~f7ICU<`@>6th|y@&5MItWH`LL;OsN8i zZc}K$&I)K zem@};qAC6lE?22>fre-n@ryi_-`oy2(kMlM5opWnh zUT>f;AMg8CQ>vDM)_UoU7W4B+>h1zq2+*!}#1*#wdD3qOZsE_S4G;tKQ>Lfog9 zs#Eh+9p(w0r+Y4HIuh|=x|Dq~jM&0GsVf!-FqWN|jzaH8%@6KW$%0zYvoEqR;8YNi7?X zu-l21jMiH(Pc%a9Uq>FJw6g6w0*{MZS!oVUW1SQVnN99Lk55|eq5k9AvEz9}K)zUt z-GDm-S!3wCG2Tdh^k-PDktp{=j3TUGM_&tOsx2I$fb{x;`EG@LJM`BX6>w8NyW_Qa zO;L&te9HrkbWJ6)vT@2Pj!{8KJ6qwRe+V!8UMD({22}P7z@B$$Hy+uqR`0FjjK6-J z+3tekmaOhj@YW}1R$E&r{u|UaBhTa>=6L5u=K<9g_9uT ztBH~!C|hvt$?f(&!zicD4wT_f1(oj)&XJ=7wwgJV*IxNHJkF z46llPQz}=2#(*B`S4T_ZmGY63J1iJLr0W@>vgsWRLw_Sp+ME*jW|i#^suv0_F@=Ig z+8XYPJ;V-IbtRs9vPnt?r;n_{9>Wx!3Krwj*l0qcED4NbEJSBa4fpO7zP|d!^Wg6JF)lk7F>PbKNFXGv$qRvMh%$#IcCgcSMrQeqEFg#D+j|mD*bHSZ zU5gV@rn$t;!%W7gbK~@yGYwYqw*dtjTgv_Vtj#?&GC#V6INurw zYsSVqia8lW9f^`kQom=+b3>P!&eIZCs^MKa&`Q5=feKzqwQ<`RdW_f#-^$OWk(@2S z7EsNhZ2QO&e8jEe;U|K6nH>h@8h3wD#Wn8TdfRyPvr38y)#Gw~<}=&}G*Wp5`8=O* zKKm<=Z4ZXMQzxrTcTD`WJL_9SL+y7-x4$|ItXn_0%(wljLP>wrZM<34f7E(hJ{|KA z$kZVhc3aCuudV#o7qFLQ-H4tP3Zfw4+{(*pzgzF_I1FM4r8G9`Q%I8^fq+VdpL$-O zu>5rT3c@k||E&aa#+@*vO+kMEk>|N--irlqLU5B0`L&!8XcGe@ZmB^FYp+%0X<33l%te#0vE)0(`)&C=RAsglG za40>&ypA(zto<4p4xV%*%Aq_HGxHZK{hw6{Lt8GHbMx~i<~yCFxmIj=@B)I%m{=-K zXAZEPr(|($L?kewPoXxlBW7v^=-}KPk~$IRG{1_;i=EyS-4TdYV8oRKQzb(pv{+K4 z-;ur6&rUeqfX}Sd)Kw(1J&2eBHC3Pkn&gT4j^O+~*zCTG!)ztfi;u*=(Ub*Va}xCV z9kVG2@U;kyzrs2CAxohz&L&g^$JySvd)hN#!m-^9RQ9NRYl}`t$K-V>09D^X_>?vY zpqlTH3lcVE=Dm+TMdAbz$l47krxu%-;^wd!GnVt!zVNpZhdN5p5GrLc&C_bKwa%l+ zky9p{rk4;2SxlqzDc8=&>_gaO7D*RoF-`M29R^||s4K#{aL!UqjcErmH0Bd3FW)+0 zawg>A>B-iPtZM9Xa$4$HgQ* zJyn+B)Wj57W;lf*rDxxUu_>{qV<5;rG~PWOv1h}}wUHlp=sx(+Hs2;5eMl-trwJyK zAf3&W@hcNWWh~cWxMfC|yg`j*tk!ev5Lhc3HC@+Z--LS~WDTEb?$m4E8ZL$5fNZck z#9%P_>V4v-Gzy_%A|*NPH$45(_0sJkjDBsu=thJrdkXoAxl=MEO+e2|DVfkt)s$ST56{muoc?e;0NU6RiEn-w;rhZ#I^_c8)&#fTKsq48UCsvGb! zy_kvEBeOmAG_WwG;s5+F1w<_PlYzfe$Qs?=r=k4b?)?6gc>nFw=6Rdp?M5`gGNoUn zMU=K}D>1Av9*r&BWAwD1biP_yy5~z!wi;~A4lFo^u zi^tHx)g04VnZo=Oy&h#W6NRpllTk2%=xF1c0U#>7HUy*7ysHd!n+yr!ISFsKAK+2&In4xSd?18LVCYY)u(P6nud=SAHTYdXl? zDQVHA8iZG^L;yHh-VyM=69w%559zprIG_L}{r+_*f*XM>!FgSqjyS+=m4fX5?sGM` zMWgXscm%0N(^5?v)eMe=C#Dk&@YzLI%H!?>WdB zjaOzdg20wR{e$vnM=!ATkrx=7>CVnu9>H$!JuTt&2hNg=)K<$(Qziu#Sec4P3`-iL zZpR&G(IIh$&PJ61@86f<2e}{VVw`D14=unU3FALN4fGvUt*H_QHX<;h5T!_c&*LrV z460u}Hb93E>{_*1JfjcbU;m_$gyqDRQs?=Gaks{H)J*1J)k+6*Dz@EF+beP?a&aRU z@W6=DB~CsHono3}rxa<@#RCm!Z0sq?$E7t&{!J~35~+7jHBuFL^e{YuIx>orYm%f= zM}^s_5qYF6Z;F5&lY716tjEgm@ySIhQ#_MhQt(C^qELh#Bmbr*QFm-RS4`s{9wZCH zh%LRK^>ZemndRVHCNUA+1`BiY@j}_}ICLV-7fsS@%KIBjV89qceGI^n>@df|l;BH$ zl@n!SLQm-+F(UJCN<~!Q#F@xG7DNArrOq{*uc~Dr!2R&K;S?fdLn+Da@6iXs-WHaA z<$!5mPkQw%xB#w!TTs=odhIV8L=K;e1y`v<#rSBU=)k-!EzLsDs&)%sQ&K|AR*d}2 zd0Rn{P?0jh0J3j>GvsRh+*N+!5q)3>X(c30Dh2X#f&Y?l@&{j1v*gB_D!&XKPpk-I zGEzZ!K1`eN=zYkz(^Ie?)36^#~t56yft|e%s-QQ@&gDma2*@H4V4@WcZy@B z&fKm;2!?_yQm6PYF9f_rw?n{)U50fX?G<2qU#!Rvc!Y&^JYK@}MXIs<_uT2_HO10Y z*Dx<_4xo+g(D_8HZN#4^q5d~m3l3yxT4!uKLw1KZtshxADaN?K>J&`Ito(MCt|fyGJ{-D zp-$4lt>L*M@6Nw6(n< zn$Oekp}x4~8XZV*%- zv4BvwItXJ_0MLiFO&Rb829CwlOkOriDKN9;DBYkc6O}WRMvC`RvbdJ>!X=Hs;EYhvBvO1tr%dnW}ib3cMzh>yE;* zolYxNf=r1ZP-L+676s@Z=k{N%jo9JKV$ zQ<{20hU^+;YB*84r(ngFqdYT0zN>icTw!uF%q2}&Tr8FLjc|%2Sv?VMC7_EFNh`)-dYcyq=+jXE|cz& z$i0{AG?1$xjzn_CcB1#6GQy^^v^%tc42Ys5VsTnGL713vF{KWBrzWIzQY&JKYPxX) zUJz&c^{itfIRkU*!kUwcl;8j`5&etVp7cQsR|Z%6Je2?l?0mFH@ilu)ctrGzs@k%2 zNjVpjG&~5MiIXO~AunRE4vTU>6%06?T+?8Nr*e}FK3EJDmzqf+td_2y%0dqmB|*O9 zuZ+2f4)SkBIK?f|Lz?8Vo5<+nZ2GYC%6SJ!9Y}$vT5Q#nUUtJ-HpQyxX|z3p~wU{ zc&`&uLi2#M?7v(l$`FJh{!Y8m>f$C>v5A;^)_w&{hwc0Duz!lPD&Tn^RE!MdTLKAC zu6LIx^?w@eb@AYbX%#XQ=+yHKdb&Y}^R1u%Jqa+P3?Vs)ISF|iwFQxlhjSjg$8oml zJRO-|udj`16}@?#dxtfh!v=r;==yN{<90ImtcX?r?at4iiJw1HZ z8xs0evP6qm@6Zz7@Zh^Z5UFD#)tv%gf7MdM&X7GR%DVxVS)fEt~R+ z=S-JpQsx&L7K`DNkZkPi4;dxBZ3E~P6*YV?8z}Xxiab3NOBj@cxRwxG#KdjA#lr{% z4edcA@B!{lPtbGtkHuxu{_Ow!+$*EqS}a)DU|?R?>0A1hw-@GCF=9EqE}L_OBihf& zYk$XmWz_vPs7kN((_7=hva+)FDtiSx1rvyivNFd1EI=+wJC#%l41^p1N_HSAgP0N< zK%^~xkxo!!{*?YHjs5EJ`i%8|K>U+z5TJH!?Zbx)1|p*L)YQ4*a;AUf84y@s!!Jd5 z{k8O?v(soV3eB;F?%!YTzw-yUS%l1;kTBvoffS5^jTpSAkJ$$HEoR>GeV#(gdw`_k zS%Uj>IndAZ+zJ50O2==e@~6Iim8#Ly5Aw(z)=<{U7rRM^@_7QK0B(b3V8{?jkdV*4 z$y$J-LZ;B?w77wLdM@UV&wIK|$0d|RG~RH6E}P}bl-VT4LaAxfBuBwG?B%h}2%D45 z_3nB8v5mnwh=x~{y01`oCB2pJO57g(|H*IsA{xp)$>hB#g;Zyn>*XoTi|pNJ7dC$v zr=xZ62~YpLM0*5roX&lqk#=A2Ggli?$sCbMFT5eN`8BDgRZLo>rKbaOxLN7)jhW|{ z^uK-1p>5Fx)?25lv^sASt;3>X(EbtF3jCme9vqF?LvrScHmf)t3ZoO4WdCI+{fy92 zSjmr1W@Nv1py_pv4$D=brONC>(_VL6pk8Js>F?i2+asK7X-|k__*^2TBlpV{o6M$6 zW1x)d_lg`D93U~|i9J?G0S3=Tc{pBfBw46$eZPZRPs_^ME*f~tS)@q$L{!jgs<8!m z7K}YUF*PQmGiG=Q+!MoY9vT7ctixIy9opBTi1*O!tge>qbYNkp_gmlk;rDzif8+?w z)BAQ*BESFj1%TiX17bIeYIy|&`>N&K2M2-%1098+o`}{Wh^z)57^oi3BKI5xIwjl3oM3nT`{9H0P=6nMoHpxDO7#{QI}`hF2KG!!Ri-L3)W z^NL~PV{uLKx|dDUt$4`dZCFChpx|8jf)C)jQfKBAz>jHbZh5=EXZ>Tj2VR|`k2T*X$J#=eq&Us1WQ63H zI9&eaNpFC{v{6qotNA|NK0yWi6Q~jyKo-VM_g<`^NJ%*nn?VkhdA^Jef${H24rswb z{w$uFk-WFI_AE=JJjjzgv%nWBE!!h3PiItg1r}gA9{Uvq1!>apD%METUHWHoZXNhwhubOoKsaQ0B($p6 z#KRjWCLHrie03wR+o3U})KA>}{E7V(l0z=;?d=VRvfFMlpAw17IxsjGB3NQ+Y5kxl z>^TMhKT^;im=U+-(yqr`WbC-o_9$G;YfWvSd78%m+(zWjhka)%Tahesmv>S|zXYKg8sxxn21V)Yl|NC;mt}>U+3r zU(%hBb)vdzU2VP^&Di3Le)0S#BD-M`%*HBg|6~0s>%(>l0jJqlt0{f2ON6nr0s4|G zCQY%w4`#z!iLmC~p^;`Z{){7CF7HyXc)tLsCJD5PL{YDx{yxO4BTFNi?8yXX1!Nl-WFpE*?o=lQeo^M^hl z@2G3{*3D1kNjBO7ec%5MZQ)PIhPt-IcO{=D!1tEB!8(NnF*so3Yagya zxw7?4rMBe1Q#2&eVWps0Paa6nbKjaa#{)W+X@GV%H@AIGw$vw&S_1z*D1x+UL3S`* z*B88-TDLQ*`q|@tN*c}0O;Hi;Ut)dClYdvQRiSkVwly;l5oyxX2W!3icNoVnV($0X z0}X3ys&IvFZq8Yp)-&^*<{r_Mf1eni;MF1@9vajBa{Xr~)5EGailCD40x!$E^UTXh zW{*}j`M;Ho5A8k#!tF4-=S%Vc>(~0s%_8)$IWYMu&;NfWOU!TPqGa$>amaVz|J~G# zyjk>bR&V~?UXuSawiYiTJQ}gU=Jwt~jj`mvgbo1Kv-kx;YOLSg_s3$|73^wJ7f~>iIM4wuU{=@ zfxbS`q!ceu$l+)78>Bi zw5fPH9N{Ds^{8=dU}y+~wxz|JCaT@?D$}+vh9sxyq?gD(JTjbxx!Ft;TT8G57tZq3 z8R2)ysb$~b^mc8=X@-W1VX1*`nA|f3Pu6{vRT9?e_W=YF86wRgpco z#{Q0dmAW)_ze@UECrITJo0F?c<|*%N?aX|oqsZ+Fx!6D8i6If>HR+dv`uW0oQHk(E z!}6LW3lWMn6{TffWarFG@Xz6bAuprb!2DIxS9R#8?cpVogwDyTQWP^Xs^8eYzJ|fW z!>f;!wbZ0;iwpb?&k=)N=E&|h8n*Q6dgTcF)=W~&(JNGHmQ1fVcI&>0b7e)y#Zp>2&U++2 z-YD5rP_Xs%)M?*aqeLvv>5LZDm>RR7H$7W#N(=vePieN@7v%o*bFNAQde)OY4_U`y z+%zy{QdxJ}I!-9cg7XVdIcQX7>pUUU&)Hbx)^=SSzD+*c5MRy8c9tRz!@=&nEXdUK z{AZAebq?zY`A8BVd6=1Phh{2X`<{1Hwg1VCPcHmn;puuLbW=zqe7lL|D=hpiVBQF5 z6-TAM*L{))^x>Y@Byf34v)%ID8R##Q<6o^)mFJ8tW2$L-kUx3u)}(jT-*t+cH$U#z z!Vu?~{!y#b8AeM;YkZq8aIi{{QFR6v{+AYJLHh8>D)}=Hh7EOz$elelEfgzN1j7!7Q0qdiT9)0P?$KzF+g$7V2Zi4-sD1^J0Fx z=8p<3rxQr3JMDLu+Up=7i#wn%i%i|>>bI?`eQ#eM3ZZ};MI1H`juY-^*D_~&E$Qys zn&tTZd#z9*MvqY&c0Ozpi8o>nW|V(k`zI?N2tx`_Huk;5`NT_xr&o0{EZJ2_zp*o+ z1V0lP-1Mn2wff9Y44ug=@cYKZD+_!r^8EG_0w{2BNLM4m8(c}~wKixttb~MpjShpa zr_T7`5|7vRXcEz66cMD*Ce7*2UU2mGyd4=b-cyQ7ky$} z2Qd`<)Q4|ce`X^V3rF172ZFyaNR_?qizZoUvrYI)$474nl=y30@@fVbOeF1d+gvZ| zNx>8oiQ|jgwUG0Ok}QOl{fb9U>!tPXnuze<|9cvoEm{1>EjS)7htLp%Cx(U?X=y{- zzi6klesDclY#!>H}s0|B`s z73*n^Kch4zGVni!l$Xo36)Fh9`b!c9F&Y6cewZzi_@QlL#oClEOc}=@5QyKYuk%<) ze)gqaAneVPtYOYFkgyRJ>#=v7xb%g|rioUA7ZenP$f%mB;l8v`#2;QV*(bDhpEUQY+L`4= z!wbAL?`1n79lV$8Gd|%)!bF zL91%m4Uy(~-ryG&I_vKPWx|iWg4ngR6`KnBnNH$Q&_W4>rIjc6DX;>73*+n2_m1vO zlfRew2(!ZHq-QA00mGX*p4;hP&A80Ur)wJGb3@k!2!@$LqJ{4@H`LWpPG-Adewn|Neo6=4e4DB#MQ?IVy;kz7vKgMi_d)ba_!keHY__vfB?tCsso+@ZFeSH+p5 zaFF9ObRYlrf@4N*i`vdk`B?DKLTk5j!u_gq>nnQrIzwd!h0SuEe&+A{>>KNz714tT z7A7nQd?k~VR3GE=I*Qn*nDQnEGn7&G#E=oJ={W4jWvzRkYens_n z<};|&?fHs0LxD14F(lu~PRUG1#5vO)={ty9WqzFR#{2YY9mJzEO9m3RMJ<7elAKA zP^81(>cCDSD!fF@q|;rH3gdx#F%b^k94;wAfCw#7k!E;2AeO4UPydzicOMZX{1wYQ z@mEyiDUQCyL4vg1*0DucfXVGO49Huhi(nsi-H+mLQ=5jyJ^R^}`_SFMT>$G~h~*rj z_qQ%4k=&}P>OC~zYJM;*fE$a`d-RHI3YSJ!mF+QfyZ{UuRI5%UU2;%r&5Bns(yaxT zuG;%ZA18C%m5Jh$Uw6ZQ|EfUhZhgE2^i^z{bpf<~U_Vot5XrrP(aG*7>GlZDa?-DH zREO1a$o>AndP#F3{-64yG$YLf_ez}NHTnvHwAk9Up=^s364$2RQ{3 z7llM|vw_#}sxO9PvMUhtoK(@tFzE&`m8GQnIz?IQGJ)Syk`zaW3FgelP?r+M;JTnc ze^m&u-X`(-i6tFyexdMk(;-P>wc7YKWpiWWeB)v)FIR^G1dpV7j)i0Q>oXrf1_YgmQI8Juz=a@ukX zxwyCzg#T?eWx)?e4iL>&iH`y7A8#td1LF}_c;#eQ|L31)3KdV1)=KX{pcGjBhw}ZS zeJoJCHDG4#UnUmFM+r1JRgYx`bS&nrB*moo0@7^+_G|FmLhl!pJFB+po|eEoRGm3I*NySlhk9#UFVo?pCq zMfQptd~!Xy-gMz;#RmB9)}gwu>R!6}&;YU*LiwE5Y39o|Z#p`0)tg>eCD^kAm~9`M zYRnF3fsG$@mH(X?I02_K*)v8OZ3s0PX7YDNHI8Eu);Qw7i@QFgPp8_F{z3=>$;9!hgbnK zof8#I+{mA9ZBmaOKHnM0q0dRV2L(S%`MSB{kO528&=s;E^IL@XSx`v+b z8<4hXk%>WmWntF#grWgT>T1dq^6{~;F0O!&f*mp{Auit3+=vD8?lovGpnt53FC7Mcxj{cbD-r9k(;a_J_&5(1K(QbenB{F0JMeq_0kj&g!P!Ut=B zt?a(_S=RjUF!5$`PpkPx7I3+|c$_}dGc)l!-&RZ*0K^+UL|MQ`|8=G=MptdN!gGTH zssK&Dnc>ur#>3V{z-&^Mdt&xo=0=4+6tLvDx3+|i=Mie9Xv51|2b$o$v+#>&pF-NB z*%bkR)CDC$8%nwFt%_=0pshZ#l-%A0K1*~!q1ePCdI&Udm*%WExO6V9^ww}I(&7{Y zDyp!|jJGfiqNu3`LF%v1IlyhMRGQ(wBBpO?Ii8G5<*>l7so5n_9~?9neS8r@*giz) z;Av|^oFGRWfj)c3?NUYb(pRywZC5iuLm53l+~74jIVPy}a928q;`RVG<2UY$A>c{+&> ziz@gX;k(;T_DB@`xkwNczYwsbNzlnqnZ zKV-6TQ6nE&Fgs_P>4p&9-sC{{<{X?u>RQmcA@ONlgY~}VxL;-Os@mQFdE{F3o zhz>PXZV_iuvK7J0_|pvmYOmocqk_Ctv+m$BZ6Pj_w_VZhgjoO5weF03VrMgB*I*c& zREs+EC}2|rR(Ljgq-8m?Pg*PL%(|8-We#^*4#s1_Ul{=)MI|gnPvdvcC{%8J?Cf+r zT9D3F3YTCvUh|g31;*U2k@_5%FP3>MIvzbetTNt=dHv`~_Zn-0ytNUWB<1E|J)78c zUeTL`bjKN`ZjNkkNYUqr7^B++@qj;QCeq%hPA{lO`C?CFKDSLV8AqZ3JSZ^R=S%*S zz};mv85x<3(8Kh|$dzADsJ7c(?i0dWhKHT^(Ba0Laj9v<#F=CwwvFaPO{>k05jpIT zvDCZk;xRcwA=jw|9kfHWoAoSJXqd+%r@hts+4yuO!Ip>B^!B^Uy!X8!> zfhdPJH$@~WFV~Dv#)){9XU4f9_=tYv)*CFCtNHXQMnf>~S`H`cvh5Q?KV@upm{4_iAw_v*p#%=(#lh@rj74W_96+HL45hO=;R91m|4Gn z#U=uC%IOyAjDNktGDytlm*+*T#e1gM!Hvz$YCy$`g4_3jDw5sDHxo8K z7LP__2@;U0WX#~L9RZ+BZ8JB#(=Y>vefX-4~IbF$MUI)A%55rxh%)od*_AIzm&b+GMH? z4F&Qkybp6zK-(gDu;}ju_~0W8X<6AL!ucq8V8E3nI(^7SmW{r|?VU=PawZqvFL)?W zQ4yls3_UDdYCsA@)K8CKWLa@241rK+kTCsMFH-lTzwx!a%V|GEVj`2qUXU^dGmU41BM&+=KQTX zP28=IH$tfJ|{8v_1bjirdTb_F-^!xTk z=H_l=+ZlB(uaj}7rdC}c%TEXsJ;wUnt{WVv{to3+g<}3nnJjqW_5j^E{xp7g8l&I6 z+_7jz4h=DR`%X1m@e5V-x%(50#XstaaIvZab9N?K2Vq|-K&mY!ECi#HeZWL!qZMDt zDEMUTq{|hLi-3%bkC#Sg$;|KKV~BZ8E#@B}NnZIY?wh4t3if}EJG(?ZAMc|haa#ZL zcJG&>v-x8<15Ib=AZo)qEiLY8n}{6Lee!9+5y3hyvtzc2hf4toa0+c-G)!A9txtT& z%UiEHVC;@vZo+Ymb*(S1moKUQ*&xVxyaH?dR!NE3ai6oLS6m{ya3QPQ>-H*0N2N%a zjum;KRI5Gl*}0)P{$NNhFGL2l{%BzlC^-7p?v*cj7N&`nYFDqD_MGYi>WmGx{`fo{ z*%3SM>OznB&CAj^7;-XI-G~5E9{7+JQKRqdKuWcnBZMy;uia%;9Etzi#!pvhU02uM z*;P-O43`I^uK&UyW@;A9V7Yc@)(`Q9;RjUUvmQniO!w|fYS-VHq^sQnWNHBaOJ#}7Ua>};D)@qhu zb&{Vq&S}F)_Rq2$Yyd6wE`VcDoVUxWfSTbMw%9l~N0jp;A`#f$6$ILU{!5r#n6xD# zW~Gdn2%5*Q@%XQEnw8f9>Ee}N0p>rqAiOC;+Tw|K;|LWuI0SbfjFqjKkbd>H;n%;; zdjbsT^()a(52k10lmYNbQcD1PY@7_IbhpSq%Nxt?8eY0)?#ZXC4odGwE``N>b1q=i zfcf_*#-9q%wS`2b^x~$))lXcN0$MmI4F9%nqFxgHR*BBall|Kx%**40{Bnbqn^JeLm$YhD=;l$h1Z*)oirnx{SOY$7lZ4Lmw9wLM9!e5dHyTh9YOY z@Xo3kU@N18q&|r#CnP6VyG*%0jAV5}=PFUg4l>|WOO!Iv(~~nsbM7FCnwTEXG^WEv z9_~z#eIeSVJQY=xb?s=J?%WNX;pqR@?FLwz8I7==Io-EKT-m%%a&j_xCcos1Im@Jt6g1MconN37Wa)r!}n;R}<@5@4Xjbe!nfPU+*SLmHLLD)xjxYDXx zPu$w}Y#QwRwRFb*c;N}?Au=tE_VA{VHu;06#UI9Pk$*%ys=uWJDZyj`vzHc zAF_D5N3B}Dyu8fvVG&}E|A(SLf4&(rVT&p#jLqMSG&{30Qu5G}l72~1S4*&BOSHAK z>y1cG(D@ag5(5OuYz|7D@^YwRqdu?s+Wkj zL7{YJIE5#EH0twQPL4lz`-|+qFI?Ma+5dB#L))KI1z*Y>_1WFyDKrE*y!Za$H%6mt z(B!HioJqH?=~%}3Bp3{|LdBc;OP^Ch8Vln1rL@_Vc=ZGLVcRA)9bJEh?DVv4_mRtwqJzJ~6<8yj{uA0nSlFSXUr9wdO|_*mj(fto)0myB`Z* z2Krr`jo=RTViebAY`%1^FC)$eEI}Uv}a--#jC^W3rQTXSK(@EZQ|Vz7&F%s`wY->ew^jK6>wCJzL9sa*REz> z?64*PliC0GKp)i1Jggi!-X-EklNP%Db+67T=bIf{7B&AS3T*Yu8(2|*G{wJXD`VfcL+FCKJX&XHV07&_y2mBCn$| zMbILpiq3Xs`0A7W6Eblj8=M%$j>;`YVHEI{5_1vZ?I=B9&W`}so#bs%V%*ydbnsSn zd&4ry09b!dyc^taL3o;3d}`g3VUXL_K^`g%dJMPH55RSEiU4=OR>M!D7UobwdfGC{ z1>I)@Y|F>Df?E-LXTzA(KK@)?;##dRM#KfUdD9?a#BmM^I<)GK?Z*zVKJi8TKyM6K zYuB1?Ng+7y&KBbQ)7XovvN6mK9a?_uBRvb|kaJQHX!@-WdIpjj+TmqS3Am@i0rja< zI;YpyfN%^7gdv%M(ZAzW(OW0g%SOuBt$NgVa^3+HwXBJxA*z~uw&IB0C zIi2_5yXE|8@gMNvZ$KcRH5@cZ?$cUh_bQ~m^@;qeN!g#oVy54X_a;*jxM|vE0!dvM zL?C57^K?XCi}AZzEbwROv#7Hug0k+;UK91`@0xUT&uFjoL+JeqMf0NuSUVGybDgok zEspE1SN@l@W~52GX!xhQ1Y6qq+x_!IMvdd>44`0m9TRwOw;(SD#{vql|GbXzb(_<}u8pK} ze#iT|B*Uz+d+KquyJ7Zi^LkHOzkg3%J=@rq4!7cVFMWR9Uq-Gj5E2Q8J-G7aKuvOh zUomfVcX6?lx(TwhM>?hWR>uCi4R0J+V7sPz^!EW_dmZZBqXvZ*j2I5ZR3|tO~ zvx*pomj5qKYhgqzYy{Uk(`{mcuTe8d?3r@|rrWH!9?+m9&WCm|!Sz$~pv7 zCcujx_;EI84(!-w<%bC-)2AJ90X;^eTBMBOS2PnUTET^`ki z@X~Cvcq3GvGlh?@9?Ac7zG7h}i9&Qc7FfE0e0giK+DU@Y;uxQqad_Tb?7oEgBB)H5 ztZT*VB1g#YE}L>{?ew1AwjXc<6fs?0U-IJ;)jRYNPb!-u%tDcqC?ecc?LM0s3?*M( z*5@6Os_Z4zZt8;-5aoI_`m5kTO1ZK6+rcQ15!kwDj+t3Q7qnDgjH)n9$>UrIUebFK zB(5UL$9=+pqi+`YX|7gTZJ zqQQfLj56o7JGY6pdUCuThpGPD7a#`bN@T=Q&1OQ~o^d9U+O=gl0dB_T{d+IB*h3!r z2lOII(w44r4LCFEj+W{T=5IF0hYTjEarjx>cR?gp<<$`V)dyoKJzH`0Na9R^r@58^ zu5btgpe$k8HyZH!b9X zLMibjOhSM9U&mFlL!-j*-(H)xjely{rV!~PW{hGO`+{EM4d+ewN6}$JOfl{kb2^%nE6n zyD*Li1hKyoTCP{+E)}ACKM;$WbRw{@QL?@_1qa0rE`NZJ&rheo8$lt6@<4}c3T<1` zCMnzj=lWxP=8;s!octR*bq6@`9Cb@VJ#qvf127x&iGr3pB583k=YhE*icLOGS;e}8 zHmD{$sGU8;%J_NFq1(}A_iXH?!EARDy$mca5@f<;OB+T>MCPvNE&E|AA)~e+VSi!7 zwHCpbb3RuJ7Rbupn4sW`o@?)ph1;{*g@UPuI5n$j?GV;jtGxbu0pp($;j6WK3Y&vi z&DTU4kG*dezdCv~+_Jm=ba)mu|DNnU&z?7-{L%9eu~y?^IMhm=VaDfBj%w=n3*)`P zUcCkMBefhMmqYK;n~9lYQ~C7PUuWOtzh|`LAu@M&GvDn?9(=N2N{AABxY|=!Aicay z5V#oC@qDp6h!B85xKux%;8SQn{ZacPcQ(>_iNeXMD;cG>v@7wFZ5AY|A-~9GOj&|( z!zM8|uA^!GLz=OqvdZh7{x;>GBk-+&-=HOT2@^yeLg2Z4U4G_ABbM}x%m+1N7gp@w zE;M(m)s)4Pet1c+X*_<-k$;c>MlMXe z;T)1|q?Zq$;3XTc`I((Eui4XbCjf)ventZ^msvEh!&TQHF<@Sc=A&;<5VrSMz(BEf zPo{$j#{ zryd9N<-g4HXtSuLNP3^E6t~&9@XR~*GV4|N=ORaNs7+J*YNriG(rC0nj4C>5B;4-T z4*~y~-1I0n|LG>=F4L(o;A@ZbMvH$nSsY&$^5D) z`#XXkqwIx3yeDIu#@nvBv-c7-K8cSK$X0qhqMCj*+l+v!!BIWZ1{oD>Z3c23RKY$} ztxl`%WHbe+PwJ05Z+a+mgCf#0E?&SFC)3GREmu3iWdw50YW#_c&)`~i&!JD!-&g^HKT6Nq;V8CKch>f!L; z`Eq}MT7i+&c02Gx*jc&&WWadm+Rjn)o}Au6rv1w1So`A=)*@HQZ|Ph>Ep-0gF(skd zjuR>Hw9i*A^m1^cW6ZICc{%J7DHtCeuTa+iro8RpnUC2@fKZ#1S{6@4xl*V3g;hzuN3bF&8TA(FnQND6c zeYYE#@TwpJGvK>QH0*oD^1d<2UX!{wWE*t!uuG8UR^8cc^o;)F-0zI5LesuqfC6G9 z$bNe$pX4h`M?RZa>_qdVr84eKHZ+|kvYp%bkH71G9v^OmL|KA~k0mwA$%(1@=~8bz zfVc*lz%{>Cky2qdjT$N=9UwC0^7db}!C+;EE>njl(+f(>U?5+AQ;ai4qVz42C02UE zhil`7emcp9y0o88NQZfsCXz-3$iD~KM<+KwX?n3!!HQDnU!yQF58&KN%nMnJ2zs3q zGD1r9QMjF{lVJ1G?JFIsCI@f6-J{KD(UlzJ%ZzIccw>E=z-7!$@u2G6zSpv+{@PT9 znOK>pD}O+Sn98wqK!l=E1{K~O!|qB_jnTj{g$XWxtiNP_hhx_33eGrb8?1W2M|QzbJT^>E+*HIk<2t|YhhHtUjIp+t&Fo0pGBHM!>Lq#hV?8P?nr%4*Ndb{YO@F+61bcxTG?@9Vp z43!nVfm_;fh)TT#bY6A=Ezd`=paVOTM97lAFaP)sQBo{aKwLT~W^RjYJCw1cmX9h^Pb&J-%(Y3Gt@puH=0+MRm@%FM66!w|D+P_+GGAFNh5b9^ z&D-Qn%BYWZ0yTQf7DWOGfA9h4Qw{OP+O(Vl0Wm1apPOKCpRmVIm#Q`Mc^^Ie2bA== zR@1qwyNYsHpkxByA|8U94be1^gbGU=5Mz67j(;H&$kz4e*C58l=gsdWUN$28g(Vz-j= z2{Jdn(tH~5np8L13Xah|4?{sMu+b2qB!dySwP|bG-aAicf2Y@2afU32!lOtRJ}cwe zU>mSQZ%+UI@B_bHXD(YbN4va)#kcue+r4|j`Oj7|iXIA6a)U=qztR|@7wD{V?kZH_ z3`|#&pkg&@2ov=Ue2BR&GZW0P?L37{)Fa1sDfJ1VtP4)OIv%Vqkm3&DV zu_|6>%dUx#+{-CE8m{|}?gsvdN3Lgu%s&j{7?ERjqSrn;y)nJHN0Vx9fs=J(mQFtJ zBLU#IUt`==hBsfOTIhZFSeg2kj-~oC!zH9$-K?;-dEz_ejY){RQ|4pRO*Cz9$2HxM z>he&iY?zNxc#DU>y&1#wtLN-!$8>MK_lPED@kErs=Q|PO52wsSpD)>?8*2?m@uy#{ zi>nRXLJVepRLt5XQ~r~ddl4HpZ5moM9EE8@;v|z0LSNq6tZ0Rwu4|wZfpuYF_YhKX zoS5Cp*nUZO6Qhwo#1NLI&u>VfG_p$YeIN1 zAG1C`d07w;h#2Yz_l*`V(7bmCTKJx7ym`11&wZln-x;{}on(BM7HPwy%y1EVp_F$Y zoona5^|;42u!%dZS6Qu|OUJqse)Um+NVnk&743HS$EF!9xQW=ri2boN*HU*jNX(Q#(%)O?R=NmKY>17Om7t+isRT|?*_WC;3bb3QO+20{6 zL_ka;0R2$!7@pCFIGbBiL zo44BnLl?5KZF>t^40~>&(6ojSNl$X{Hd2A-OpK>o=4*~qWp8BPThm^=N*$Q zB}2rd(ahxeyk7Ae`D-rC#fZB7YvuxU-u%2G^tv3MJploakdSjC9TJ$SE^JdIn5j@z z^-HJv77UaAlBeI9*0`24CJs&7{JWx5EypngB(4-j0_CuyrHTl7?aH+h-bY{tsclT9 zdFs9^#P1tD-e>`M@GcWDYLX;N;+?B*8W!vCPqL*74u|uCinljMU0r8Njt`Jbd{I=+ zsAVYL=xcR9aEUs$p8M<-SLg&vrk8|Z^O`n%p-i{@uJ3?p)=`?p_Bm-)d}l`M=Aeuh zjy@rvX@#KHUo)9pCo$J}nqKieE_|MoJYGT@lV4C69=?Qhf<`&5mjqj&-O??aJRvf|$lQ52wT-P4BgViOsEY;g)i+zcN^P#hsZUpSe0RE-!#BFl5$vRfUIkfXtPhA!8NvkX|)-bDS4@hE~d5mHVl{RtVC zG=l@2sO#_A)})JAO-&;VkTpSkjMriGEJR-z1y@9R9jp%X>@7|jWlDfZDN0^vf*bu? zjn;#9G&eHYrKLQ#IuOmN^5fJXI~y5Y=AUkg*lNj8$|3gBaLgWJM$Ih*$P$B^}pe7Sr!YDZ}njQ*(TxtbB)(1qf!?>k#VYcJD9J)gtQxv9b~OF{t2LF zL(MT>&b8Z=I4r&*m%u{2xC++!vih31v546Ihxof@EFY+*=FNrqE1N$d9_^F~%hVWry z3obkd|0WaxsO^Z;=a>xK=0y@iJ1E7Hn@mY9+}%>H7~-mg)Ru@BB#3uc@}qMRV0avD zI$+`=+!HESL67)TL1}HnfZva_8fPdKYJsQk?p&^UTwGj^iHVg8r}a#HL$9EcV9uly z=4=c7OJ8(+z@nIg>b^GaP;0KbRbgQ%nUDmyzMxwmTfIbOvL<`r+rtzM+(*WHO90Hi zFekn#*zz$U9T-;lz(JN^GVANnp0t#d4AQ?UDGgVZvQgS3d@A}y z%OPjqhM9y0R602WV<~==v~W{vtL8xVwG?TpSqfcM_V$aXoiQ3l)q-Wwoqft*!hL2& zt}U`0e_LTyixkt3%4VjsV=5h=wkbsJ>DOLXtCXH0Q+)~6HBtKUtxT}{?kC^mg83e- zj?6OApL-hr@=^NT;%$uawptY%_23lGSVZVHIgQATL+(j1=Z&db!p{TI3a8c11OB#U zyD-g-|MLYrj}dsmd9_<`6Z+MS7tntIecYe{5?aSgc5`zRD=sH3YxeLFgy7WUiqz7n z^Xd18kawTOKx;*uH^`ZPE*cx>6;qoRAfLn?Mg~VIK=PxG-Qw*SB>efK?5t^Q@;xfo z!IH4NL>X&iCiC@}U^+KRQXqs8TuYDhf!|eId`w{=!Bsl@68u-SFvK2ah+`J&ETeF) z>8$ctXiCBK8b_+tknoA=;`f<-I+DtSR%i=Oeod~(!3~ikvA3BxP})N+C(_9~9v&m- zDY6odWi>D`LPy82yOKTIkD;$xcAs1FZWG%yuSRqrGcD5;vXST{fyO@@Q@rEbcP}b!^Lukb zMXn^lV)Kbfm)mSL%2z`=EkdQ|pd>X^o6Vf}Pp zxr79+B(mH}+wv`x_mFb8Q13I`U**0a!naRlyL{7tqZ>;DMDR_8vw?lNUT$i_t;b37` z>`n-?2X623NzjcQ;L-3|?dLvosn&07WN9_m5%Ib1o;y4~Lgy)@?;Sw;SY$;+o(E%x z*x4tMgyn!Y-gbM#>*6_XbPxCip}aiZq>rWZF^-z!PORr8Tb|umWpnM0r3l`QHMJww zm~t(#H0r#FAP00Bu%+_59Hu)IBVhy`caMU?<^Ds-j~zUbQqh)Im*;m!z(AlBfxpG7 zI;vdtx3>wV2lo5Z+bfB7JZaTx43ck2NXMI;T^^PMHrxCVZXWOYPLLb*whC;h6$rC~ z&cDq@NgGBA-rN6B*%`zhiX>bh8K*Ly?zefkIVmGFz+V%(?FPB)3Hyzgr#1f=HeGGY z1~Umc9m@hZkUCKHTfLg3e`w_C4svpFZYvQ+7@eFPogOcf5t~Jw7lcA|ce5bGWv&*V z8q4E)=^U{Mv|K`TgopHA@S7mZPaDLfbsSApkL88GfLjuhQer9$hXSFV5T3G5b=utT=?pv8muO@Rz-gJ6NdU`y(9XAH6maBhLj`2G0p1|sh zl*jYJ@|h{ttWR@uJmI$6`6MNU@#L{{8GL=j_cIj0jZM2Os_df=3qW-<6-9Rl7jS#o z>Go#M&JS5x+22$j?)t2Lf5nuHAk=jDadX@ng!*18R3;$cj?c-m+4>p9Cyw_2USi#E zdSGDq!}H=Vq-KqLg(Goz=0%?_=~my!yXZe;k@^`a;Ch+i5ri>ul4l2ze}B2XA?F%l zvi66Jm^iJcM;v7+0-xPy2cb5Umz>E3I0XJ>oPia?I)A=ZBmKjVx? zF#yOd;}Q~TA4JK_U|C&+{7w^}WX_w=Z#H#8K|*Riy58e+w0t_8x``wdy<=@rMvY0*|~oUCLKIB`gh@D1+-S-sb= zUB{%E;* z^-o6^zd-BW4~T)0O-g|98;aNgcvJ$%!#Q2opH!n~m*>Z`R%@fFi^DTB1o8dM{X&kH z)}rpR!otb#spJJ6?i=RKdq+W>oz(EJpf#P|wSkPKWAnxC34OP|@2MFce&Nk0V!sY# zH&WQ%`SBh#P%*pL;dp6eWVy@WqlSV`R(saW22lUrr$9){NQ*C|`x{K9ASd@>4nIZn zs{`HUC!g_EAyOI|pjAfKH4nK&gj85$t1;h%lHA`ezU&F-%iYY$-+DSa?HFBs z9W$qH(|gqyEi#C|+I9Qkq*P!B z@(j||Ja_ddyBZhGE1k_p&wLRvD z1@*7}`U1_E7#ixj($d#b1g?!qjt2zHJ7hRuA-RrVXkGi^#>>#cXetEF%2sGl48Jr_I zwzAxGJy)}*qxW9w@Yron$_$KzmY1;M69|b5V!pLN8r8m;O z9wiiN&^0&B@3o%KURF2>=rzvA_uYZqOTm{*7`^vmj+0xOK$*N$J|oUWTWho9$%=?I zwpFOO0t5}8A6%%kH|XHzXd&e8EL6!k;Cwz1;C6GkR2|gTID=&iSZ<*~772r$oxZ5= zXL@_p$~k-fA&J1~E^2tN57(-z*Y==Jj~#+?UxXqcH|K{IlZm&}1?)_Z7KFOB0%oG} zPik*uZnwG-)L~SKJNmVwUUnevXMb@R9|I>@7iadb&C#1L41BwY?{M4 zwA`ws?;4w$Z1}0u0xPp~?~25_wIj-e+5JXwNHcS@Px2zm%bm9+{wDsthRPibNaF*V z9mKB^@xe@^@&#i7BnuX>vRVcDO$-?#1Sru8SpJ^x%k!J@@#Jo3ZjO(MnRZ4DN$o)T zKe;u9dC30q9B;|~(Op2w|GyG!{?mW{zh>(&K!L3FEEZ}M&8A@qXt=qJ4-+PGcuWG5 z4Y*9VegZ`jlh#M%ADSGFM3ag!x`Jv$jU|&3={;+SAS7!i>+6aR@`N5Q^Xw-_^VJ9@q>13`$7-PY5Y;4j+3hB*t<^QEc&hO0mXO&?66b}bZ8*o}UzO&9 zJ$MG~OUG~-A+<*)ts-Tb?qm*_m%7J3t9VAO=S?Rarn?$BVQ=pCoRZ7={B(N<2XY`K z6BWn;;?AZfcFR#X(N3;2cY7Sj*8y4x0hPw8IaN6iSm@RAlqh2;i zSxE_#ghWB3cD*Z5!wZc_I!QTxR1s)`^!LYPm0i9((GizP?e#z($#R3q(D1{K;N7L2 z@1K^ZnFExraq&i;#tBwd_B@6E^ZL|e>gDNnoz<~8hNq(&Y^ed-eED?G%_DIL^QTqL zCwb^`^Dtu0{nM!IoNxN-j=O6jUZvUx5D4zXbg2Tfl`X&Ja6%Cm0vf?PlJthh+0~Gc z5KK(0c#n(l^l~62cYR}$Trwh@kTpe+9H?x|%J_;8vgYjyd@EqT)!!i&bY$lo@!-%C z(wwhwbA3)n!$47|^hb_M{%|2z2510Rxk;^Ba_?6r1`&Gh_wC%+NLflL{Ay{VSUB@k z{anszX0+r0X{gjk!ni&WIz74(LiQD&1)tCNwLjjw0j2%nTq}R<9@n!ay`GMoQp-}9 zTus(5CMqk7XG^eOT;oSmafiKj=OH?He2$kR(Uv)O+V0Ql*%OajQ)2{j5{a^okebDZ ztD}ScA|9KK>A`PaJKK6}sYfa$Q*ec6XJTn%e2)83q82zK)MzloXd- zjshJgrv>C#K(pQg7cv$8Xwg%&BH;FxVD4phRn>g1p>e-!=Iq@k5fKqI0q%OSPoMUC zYIHm`ZLSY_crOBxHiJ-zYxc+5cZQSpwzoOWM!~K>Wx33|f9eXZteBYW70;H{w%^(h zL55T21+%nlOa|JX5O#Zima|t9rA894e&FZ7T)KO6=^*H}4sG)DY_snz2^+_UEDLwh z&N3}Gr^5w99S(#3wfDV|2jhYtXlY&cRCUJOz8?eC_goe$N=F!At>qm7XRmrDt=&bJ z^D(~7U~r~f4;w8JSnoYCbBXlYH~Q(KR#}ATowYVq~EietLf?e z_E02hPH%|xxEcYPl%E16X-B)7EuLL@VnLC==sju+fwELvl)Tmd`2r@1j^lN_HeIw+ zx}+(xCEl*K8lC)H@9O3WmXgYsPpw;}q-WQ3&)<#sL_L}+R6bF_=V(lV^&V)urSUeU z4;qw@g+uJR4kU&pPE4>ew_i6Uz%SOCJq!*pOREiVt$KLf_E$=At+2dwG;Wv*wylW4 zqvDO1lL#maSf8E(nd&XqgEB`8bx+sLCr-x;4i^SZ$FMdd8Grupoit=Sl@qr=o<9I( z*g>fHk6T!g!J&_j?#s*1%`=}1mB+=r?p@sPFB8M-)I_~r+VpZ0tH<|(AOhTu!*~Qd zo;sCufE^gG;|>&%DK9z)faOWn_o&k{6z%t_%1wNeT*ks^>z5dqP=W3f+qnUkEH}kg zZK0tGYAjx9PoQ|qtrE8r#~bTxo8$A@oQQnkf!wM-Vwc;w34Nib^J3}k^T|#KzktA= zJC~|zQVeklA7r<3Q|RgOO)C4-sL9lEg9T)CStgDt`%Hh`Zp=QU=402z>FH|^E)Krd z>f6>@uNU^c-e9LyPA&J-33AEFqcEAD*{osk=1or>A|lWo(C#v zHGP_(PD@Gouu`&FTXY1a1qN$-Z1?o{_X`0QTKzGJhS$SstCNZFhpXdtKZFQk&fjSj za}9=ekb-_b%XT^NG$0Z##`rB*@Y){tlM@=|An9(so|R@q$*T1jU4dyrUYQLI&Z|Ct zVnJ`qnhGM8BinQ^jus|~f-V2A^4=?|sV$24Mo~l&vCyQdC`gwY1R*LY5Gm3-97Rg# zNG~CB5K-xZ^dh|%C4?SS1f-YHOK4I;3nc^yC3oTZzHy)K%N=*zm%CrcAbV#dd#%0Z zTyy^CZ_-S%)~dQU{YqVKH+Qrn;1EQnmC1v~XSaYZC+oQYyc$5k(1>$gIe!{=^RKnK zO5XVhaogTn3J5N>0#OV=M$jNL=S>IpJpmEMSn5>uCcT~xrSwSwpc2pNjXB~V79{5Qkhcs zVs73`^OMqI=0?g**EhNXpx*je4_SH)4BwD}B_)8)CJ{=4DDifI!j*6 zCmmbI+vV!)o7P%dTH2q{NT7BFd>a}#)MBbU>asz2QG7fkH9al*5mGdP@*KE{@Y46; zp-}%{W6PpKk_i}flnk#^d1E@XWXmIBS&8Es8#An_Eva9f3JE!{+VaCqosCHJorHPC2~Jr3(4IZUOq|jR(iA!&4~F=c-ld zxsS+7^$qp5dU5(a|B4jZ$sWJ!XA?)Lu0ibmN|wDJ)-EJgEdG*InSZs|nM`@lI=zBx znwy(#mQ?9e{-POzIqzQUIKI)~w>wUFVwuq3VdV%`7l5{`sm}jJo0=M<2H7tuSF%VMbdwESVtOrSorusa)~_-{}%Jf_Xm z<4UX1D)uz8P0`BA%Ivj-9G^_SFXWvE-Pb#{&KPp$a_2W~$8bR#Ya6}Ee_l4Uh?e&b z(UKy@opPY@4_>>J8;;(yAiEb9%<=cac&sdR%}cA|M++O207875+xN}Wv%aj3l5B8o zreQT0?05s2u9hynhUCT==gruw@>A0G#mhzarW zRkd5&glZa<4Pzbeiims_&8*4HZtp7Gwm{64+^?{10A`p7=aTnKiBp81hx{T<4|+00 zKg}|_hcqwalD7p6dUS7|&%SI#V}|5ac0rx%1xJ;IbEkrWf=Fvn?|q)(x)6I0*#mz> zO{GGMcyVMEM7YGDfknRPWh1&A-NoaURHkq{6Qur<8?$Egt2=hV^vt$J*Me@6|Z4SslavvtN}DkB-uPJTo%d z{azu|!+3+W9|32C5p1noL?j!{>JxCs-;OWBVlPE!FeF*L}>!);n6q1tE}l z$*9Okufb&E!^2-~A77geqHdTE>1EX)9)@*xhHwSM%4{vx*SiTs4mZ>{yf1m9ntrKE zB0bT8lWhXD27@aF!ZYOPIzFdjW-7zPo?i7p5_=P7Mv{DQF z`FK;&frVZ_P5Dci380~HP1b&}Qqx{sSEfi#Qu(=Ovp#dBZV#UYfqJshQa|N6z!oDrIT%t1Nkauj%Qz`1uQd8Bg3@ z9l-=4D$$NjA9nIbYIFaw4^p=0oXz}|oh^Op6h?t^C(iTAnypVo3)G+z8itcE8BhOY z=hzKfF`lgTtQ5`ci)WU%TQP3>kvCmE{rw?%xx{Q&)8GGr{Hu6aj$>2PuaIob60>m^ z+33W?{l#S|uyx2vOu?zaRIi-L@Y|)JM|-BqwsQyuoS$NJY!PK%krdpIG%{@!3VP8QMlES40**ui=?GJc34nRU2@%#wPTqcCxgoFNU52e z;AfaCX6%XjiZM~uY?Mo*Cp$YYZ@2WdXwt;$qBt;cvP8mEJm!-m3Au&8y17^rEZPsV z9Aq?W{cj8plH8k?7W9ONV`z@IudGc}%H6%U5vK|{duDHKq~>6Y$tguR1Mlui}auL3}zQ?0^F)=aCT(H^M*|92T z<)XVFo^f5we`KUDYt>6t{n@ac2SmO8Q9bht5ijx8byKtv6~#s9S5JL_&yB?SL4<@* zEl?2daH%~_FPxW}vYQ5`z202Q@2ZTxLK@94Am*p|3crMB09M@YH*;`i8-qN5A?@|(x3 zIZtLt^@};)^MXDkeBLcN&o6`n9Zd!4awKIy*FLP~oBUy9MgwBO*;xkQ*T@`4NTV#7H7b&Hb9o z(41WqfY(fyupveBD0j5OiiA0P%cc%}7`ULWKlQRQ_|E>~26k23_ZyGQy}LXTp`iHn zuC4e5Dvc0*6&!++6~a}I97ByScBd*j$iMGB9gN7YZBmXD+21PShiKNWbj#UZltGYI zZJuFVUOlh0(=Rq0EHR1KgdRuDRVO4QXhJF4^~)c4d)|AlS>q79DWervuU<7c+#I%~ zTcAUyC@FRj1ois`eo++s2@!HOtYIx)I0WIVGFQNHjg9SSd3HAOfE$Ucb4G?E*JFC( zI@5f5g4`B+{a^2G^no3`c?uGA{K)CRV0^Z+>j+62g3iQV3X&UXM_~KYp7!sY5Oaqg zxBuQ=79D2q@S5*s3Rbg250a~tr#%(yV=1XzF=k$_}K) zRN9g(nKa}sJNb@2S-+GV(@lAa9|lgk^=6$jR(TwQSZrjM?v2%6dx3Xz-Js72Y#-}; zc0l{p?B(CpCgH|H1PE~c#|ubLrvMN#%C}}OwF+Mp{V9nwTU7R8bf=xg2qi!@{qf7*g36KCg*~^7n1pq9@skH3>2U_hjeuC`M@W1$moAh44`-bP zToqq6McG^8VZc@;V3eT3lxAFOCsX~Hp5rJ6(fj38a+On~|CR7q;%?NBpruFzZ%F`h z_bBq0XNeFZHIf_p*SMjSUheBET8+qUKnl=^Ngq;Ga%sVY=}#JYl(~IH>ju#emN90&Y@OL0e!KH`A}e^ zj`p%)2F~d!rBGP_B5KaayX1{+c5#f9IQwS(-=99pd;Im}&|yE0;8qdz27 z91LdllCQC`i6e2;AS&44a+CM2jLX39Xt1@i=}s$IyVBm?ZfD}|!&dBfRLJ3@h_v`> zx0zXGs};+$?Ls(4I#Z=Vdy>6fsvTa4h;~(L6MLWV8D<=&K&n0M#$mGDjk33*F1=OuA9a~?+Pt< z1qJw3>%5~S`Hl{D{uby2l|k^PTwrSI??n4^Jw8{QJKOS}K>H7cy>`wmt0mdooo?n1uB8c~|=X5(jUj zIHZg!i}?lLKKS_d?+Mcrw)bwS=e3=_*>vy0Nm1#^p%L!Z{ZPdrSc?(Vf`H>Y+=xdj zCZe%7HZY3${QZcgH~-(oZ*3XBZE(P~)vwb&G1X~2DRl)$cv$WuV=f-`5j8)YZC*Sn{fq^zn2OQNTDG-04L|yaf0JaT(pf;s^~s5@|9Zny zcB-UkKT0I)eTP2{?ClvkTAh|;dE>vY`Q|C8tNwe{|5laq-z)!5l|27H z|6zt;9=Mc%bPzr3y|*unJFC=Qy#l?hdcS;$5;60QS}ArthFKniNo-qlq7j-N+a{4x zWp^|_WWSga7ZSt|R+K%RgRxDPRN-dU)~<~fYyzz7cuGfe zp6!ZR^6u^JF)^`lK|Q^D(l|-yc6aC!+_ACXClZOMh6B$QrS!77NVM~m>+(R|Qa>JR z!=~O6so};=51evnnJva|-^4>0xbH-(R5?wuvU|N)F2RnY z!|w|HwrC6hsf6M^+8IY4kdj(M-W4cZp})pgS5pgYaQ*&NeZA=c4e-ijh3d`p^;$)3vf%!T8+ak5N?T(Ua_QpHJ179;n3%YgO> zIIUJ<$xIjKEhFXY*3$=hl*7)SzvQ<#X;6p(jX*HFoz>2}>uc^6<7a3z)7d(0Xt9b5`J#{hMm z&2N4Qzjj##GCi@6O@@@@1-iAz!V^cY1M%#sQ-RNNA6Y-jq+5K&sB{QJJx;85I(TUP+W&&^Fy8qf##bxPTI1(FQYqE=T z97kt`W^3auBcaOMGqoPboLpW84w(TU69Qca)OhRgxh9Vv|GdPlQ0~iq1NRpG@~%KFhWRT8a%5Q3D?!I5nL+>Z@S+O}PrFDOBA>pi?3& zc(!UPAbaPIuY{oBQ8NqfBAtD5Z0tj4ws7)Pz`-Na3g@XuA_DO}$fMs5(*4P!Mn;nS zecn1z=?C3%W$$N%^$@;*XRoLOMPC47!z zbWYcR!2mYPrU8kP27(gQcdJHKiBGIC%K@mv_S|Gf;g!#^x~6vw&CNBC&#c7Bau14A z_46vL=!Eb7X2IUDlDrNdy?5_kfgqB}kxuS<$#xq(%@;2S z(wi6Pw(ZBQ1t5@++(!py0Ge-QAzb!4Ih@9xr+=1$Hsbqi)L4n<#IyHRC==cQ(Ta!& zHO<8W@?$F-8@+~6-g`@%t?7BL25Lb=tvww%IRif1Tn(2IaZWw^Zh1St@?ded2{wOv ziYg^u>DIiE>EL~i*1kk0$eAAs=~hN*b0WyM5z$(j4>tXE^Gan?*Jqnic9AV6LV8@I zx1!sZDOr{O+!3{}e{Kp^`?Ke4v{6H;y z;U|)vR~OM|H)_3Nr?)`oUm0F(lRP;opYHu~Ch8{7l0;W~dv-?mD`~x&-8)|KZLbzZ zQ}Devzgf2Cx6dC|O^KLc4oy<^LExd9#lp7(158!QWuQ@bvkmJ}p}TMa&A=@h-*Mzm z-H#PD$MDkop;lYIUER#pE?CCqH9jt7mH9^n$gV)>(091B1V!^XE=_yK{N{@p5`KF6 zp0T6bpGcShr9Fo|ij8?_C?JOUcAXbt@JrtF&X3=D+5^7j4l-!#!qlP(+6xY*e2t@hgad7#nKFEzxt zEu!nzlh)SObncs!d(ffCK@7Nv&~Pqke*kX+BLe)f>3WeAC|`V=bIrUb1j3NS3&hhCSZ;gMK|Dx$B;Lz$e48J`e zqo~-VConkuqX}fD0}e-KYAN|YzK9A4EDbfp7ZtI8mBOb;ZU5*_t?R?{C|~ki8H-W4 zgpg!Yx9Le6MmrwR$Op6aL9cKyYI`7zuh!d>)wsE#WHtPDPU1b;h5G7KKqtC{^`b)L zy5oZw6kZR9je0=69moAY!#ZG6Z0bsNB<%u8@o#RxkgOOcE*3|7bi~oa3sfa+x^n_+ z%<4!4%zW`2G7vCo3aU>CH1>)0C6bJax1&Us9!JvUh0zpocGy7qNR3 zPog_pBmmaQ;}Frj30cP(w=%4qshx3qBs4&A(Rm8P|3y97Cfz4-sQnV>1af^Mv2fRz zLRbcEy#UFYg^8)u?7+s~U#$V@K8_X!`VXHMGwh-m;_Z@$@2VW*lj79oY+t^dXxHmF zX=c(P?3K`&TVbQ|A+=KGn9+`r?W*;R&~-H1m%Ds&E02jatz3R_)_KL9Ve^gt2G(ns z;9t#8D3!kR`C4MQ|(48+@NkGMJA`TihYv0v6Z18?kM4{+WHU-|iZlxVJaL z>uP|-<~C2{1H@wLu876)WQ#gwBH*CfThW+`>QvAPJMdznn!x3mi5jd)mBS6-0~DXg z&49<)4YXMsr(2DLc1l$(Ki#Iui8Kv)<-Mi1Kt-E&B<_3Q{{%(vU3{IZk~J@)h8NCrH}vl G_kRFzc_;Az literal 0 HcmV?d00001 diff --git a/frontend/scripts/build-og-tts.mjs b/frontend/scripts/build-og-tts.mjs new file mode 100644 index 00000000..0153b97e --- /dev/null +++ b/frontend/scripts/build-og-tts.mjs @@ -0,0 +1,250 @@ +// Copyright 2026 Rob Macrae. All rights reserved. +// SPDX-License-Identifier: LicenseRef-Proprietary + +// Builds the social preview card for the open-weight TTS benchmark: +// public/benchmarks/og-open-weight-tts.png, 1200x630. +// +// A script rather than a hand-made PNG because the y-axis is live. Reader +// preference comes from blind rankings that keep arriving, so the cloud moves +// and the card goes stale on its own; the two existing cards were drawn once and +// cannot be re-cut without redrawing them. +// +// The x-axis is LOGARITHMIC, which the on-page chart's is not, and that is the +// one deliberate difference between them. The page fixes x at 6.8s because that +// covers the engines that keep up with real time. Ten of the thirty rated +// engines are slower than that -- Bark, Zonos and CSM by a wide margin, up to +// 51s a phrase -- so on a 6.8s axis they are drawn off-canvas rather than +// clipped or dropped. A card is a single glance with no filter controls and no +// hover, so it has to show every point it counts: log x fits 0.12s and 51s in +// one frame, and the flatness of the cloud across two and a half decades is the +// finding. +// +// Usage: +// node scripts/build-og-tts.mjs # live scores from the API +// node scripts/build-og-tts.mjs --scores f.json # a saved /tts/scores response +// node scripts/build-og-tts.mjs --html-only # write the HTML, skip the shot + +import fs from "node:fs"; +import path from "node:path"; +import { fileURLToPath } from "node:url"; + +const HERE = path.dirname(fileURLToPath(import.meta.url)); +const ROOT = path.join(HERE, ".."); +const RUN = path.join(ROOT, "src/data/benchmarks/open-weight-tts/2026-08.json"); +const OUT_PNG = path.join(ROOT, "public/benchmarks/og-open-weight-tts.png"); +const OUT_HTML = path.join(HERE, ".og-tts.html"); +const SCORES_URL = process.env.TTS_SCORES_URL || "https://api.orcabot.com/tts/scores"; + +const argv = process.argv.slice(2); +const arg = (name) => { + const i = argv.indexOf(name); + return i === -1 ? null : argv[i + 1]; +}; + +// Card geometry and palette, both sampled from the existing cards so a new one +// sits beside them rather than beneath them. +const CARD = { w: 1200, h: 630 }; +const BG = "#0b1a2e"; +const SURFACE = "#101e33"; +const GRID = "#1e3354"; +const AXIS = "#2a4570"; +const INK = { primary: "#e8edf5", secondary: "#c3cee0", muted: "#94a3c0" }; + +/** Fixed order, never cycled -- the same three the page uses. */ +const CLASSES = [ + { id: "det-ff", label: "Deterministic feed-forward", color: "#3987e5" }, + { id: "stoch-ff", label: "Stochastic feed-forward", color: "#5ecfb0" }, + { id: "ar-lm", label: "Autoregressive LM", color: "#d95926" }, +]; + +const run = JSON.parse(fs.readFileSync(RUN, "utf8")); +const INDEX = Object.fromEntries(run.columns.map((c, i) => [c, i])); +function col(name) { + const i = INDEX[name]; + if (i === undefined) { + throw new Error(`no "${name}" column in run ${run.run}: have ${run.columns.join(", ")}`); + } + return i; +} + +async function loadRatings() { + const file = arg("--scores"); + const body = file + ? JSON.parse(fs.readFileSync(file, "utf8")) + : await fetch(SCORES_URL).then((r) => { + if (!r.ok) throw new Error(`${SCORES_URL} -> HTTP ${r.status}`); + return r.json(); + }); + const raw = body.scores ?? {}; + const out = new Map(); + const entries = Array.isArray(raw) ? raw.map((s) => [s.config, s]) : Object.entries(raw); + for (const [config, v] of entries) { + const rating = typeof v === "number" ? v : v?.rating; + if (rating != null) out.set(config, Number(rating)); + } + return { ratings: out, minBallots: body.minBallots ?? null }; +} + +const { ratings, minBallots } = await loadRatings(); + +const points = run.rows + .map((r) => ({ + config: r.config, + name: r.display, + cls: r.cells[col("Class")].sort, + synth: Number(r.cells[col("x̄ synth")].sort), + disk: Number(r.cells[col("Disk")].sort), + rating: ratings.get(r.config), + })) + .filter((p) => p.rating != null && Number.isFinite(p.synth) && p.synth > 0) + .map((p) => ({ ...p, color: CLASSES.find((c) => c.id === p.cls)?.color ?? INK.muted })); + +if (points.length < 6) { + console.error(`only ${points.length} rated engines; the page hides the chart below 6, so no card`); + process.exit(1); +} + +// Plot box inside the panel. The header (title, subtitle, legend) takes the top +// 128px and the footer the bottom 32, so the SVG has to fit in what is left -- +// getting this wrong pushes the x-axis title off the card and prints the footer +// on top of the plot. +const SVG_TOP = 128; +const SVG_H = CARD.h - SVG_TOP - 32; +const PANEL_H = SVG_H - 42; +const P = { x: 74, y: 26, w: CARD.w - 74 - 52, h: PANEL_H - 100 }; +// X_MAX sits well beyond the slowest engine (51s) so its marker has room to draw +// inside the plot rather than half over the border. +const X_MIN = 0.09, X_MAX = 80; +const lx = (v) => Math.log10(v); +const sx = (v) => P.x + ((lx(v) - lx(X_MIN)) / (lx(X_MAX) - lx(X_MIN))) * P.w; +// Ratings run 0-100 but the axis goes to 105, so the best-rated engine sits a +// little below the frame instead of against it. Without that headroom its label +// has nowhere to go but down, into the densest part of the cloud. +const Y_MAX = 105; +const sy = (v) => P.y + P.h - (v / Y_MAX) * P.h; + +const DISK = { + min: Math.min(...points.map((p) => p.disk)), + max: Math.max(...points.map((p) => p.disk)), +}; +/** Area-proportional, floored so the smallest engine is still visible. */ +const radius = (d) => 5 + 7 * Math.sqrt((d - DISK.min) / (DISK.max - DISK.min || 1)); + +const X_TICKS = [0.1, 0.3, 1, 3, 10, 30]; +const Y_TICKS = [0, 25, 50, 75, 100]; +const fmtX = (t) => (t < 1 ? `${t}s` : `${t}s`); + +// Pearson r, quoted in the subtitle. Computed rather than typed, so a re-cut +// after more ballots cannot leave a stale number sitting next to fresh points. +const xs = points.map((p) => Math.log10(p.synth)); +const ys = points.map((p) => p.rating); +const mean = (a) => a.reduce((s, v) => s + v, 0) / a.length; +const mx = mean(xs), my = mean(ys); +const cov = xs.reduce((s, x, i) => s + (x - mx) * (ys[i] - my), 0); +const sd = (a, m) => Math.sqrt(a.reduce((s, v) => s + (v - m) ** 2, 0)); +const r = cov / (sd(xs, mx) * sd(ys, my)); + +// Label only the corners of the story: best, worst, cheapest, slowest. +const byRating = [...points].sort((a, b) => b.rating - a.rating); +const labelled = new Set( + [ + byRating[0], + byRating[byRating.length - 1], + [...points].sort((a, b) => a.synth - b.synth)[0], + [...points].sort((a, b) => b.synth - a.synth)[0], + ].map((p) => p.config) +); + +const sub = points.length; +const fast = points.filter((p) => p.synth <= 1); +// "1-94" reads as a subtraction at a glance, so the range is spelled out. +const fastSpread = `${Math.round(Math.min(...fast.map((p) => p.rating)))} to ${Math.round( + Math.max(...fast.map((p) => p.rating)) +)}`; + +const svg = ` + + + ${Y_TICKS.map((t) => ``).join("")} + ${X_TICKS.map((t) => ``).join("")} + + + ${Y_TICKS.map((t) => `${t}`).join("")} + ${X_TICKS.map((t) => `${fmtX(t)}`).join("")} + Compute per phrase, log scale (seconds, Apple M2) + Reader preference + ${/* Markers first, labels after, in two passes. In one pass a marker drawn + later paints over an earlier point's label, which is how "NeuTTS-2E Q8" + rendered as "euTTS-2E Q8" with a circle sitting on the missing letters. */ ""} + ${points + .map((p) => { + const rr = radius(p.disk); + return ` + + + `; + }) + .join("")} + ${points + .filter((p) => labelled.has(p.config)) + .map((p) => { + const rr = radius(p.disk); + // Always above the marker: the y headroom guarantees room, so there is no + // ceiling case to special-case. Only the right border flips the anchor. + const right = sx(p.synth) > P.x + P.w * 0.72; + return `${p.name}`; + }) + .join("")} +`; + +const html = ` + + +

Open-weight TTS: reader preference vs compute per phrase

+

${sub} engines ranked blind. Compute barely predicts preference (r = ${r.toFixed(2)}); the sub-second engines alone span ${fastSpread}.

+
+ ${CLASSES.map((c) => `${c.label}`).join("")} +
+
${svg}
+
orcabot.com/benchmarks
+`; + +fs.writeFileSync(OUT_HTML, html); +console.log(`[og-tts] ${sub} engines, r=${r.toFixed(3)}, minBallots=${minBallots}`); +console.log(`[og-tts] widest x: ${Math.min(...points.map((p) => p.synth))}s .. ${Math.max(...points.map((p) => p.synth))}s`); + +if (argv.includes("--html-only")) { + console.log(`[og-tts] wrote ${OUT_HTML}`); + process.exit(0); +} + +const { chromium } = await import("playwright"); +const browser = await chromium.launch(); +// 1x, not retina: generateMetadata declares og:image:width=1200/height=630, and +// the two existing cards are exactly that. A 2x file would contradict the +// declaration it ships with. +const page = await browser.newPage({ + viewport: { width: CARD.w, height: CARD.h }, + deviceScaleFactor: 1, +}); +await page.goto("file://" + OUT_HTML, { waitUntil: "networkidle" }); +await page.screenshot({ path: OUT_PNG }); +await browser.close(); +fs.unlinkSync(OUT_HTML); + +const { size } = fs.statSync(OUT_PNG); +console.log(`[og-tts] wrote ${path.relative(ROOT, OUT_PNG)} (${Math.round(size / 1024)} KB)`); diff --git a/frontend/src/data/benchmarks-posts.json b/frontend/src/data/benchmarks-posts.json index 79692bde..4548e5c2 100644 --- a/frontend/src/data/benchmarks-posts.json +++ b/frontend/src/data/benchmarks-posts.json @@ -7,7 +7,7 @@ "author": "Rob Macrae", "coverImage": null, "coverVideo": null, - "ogImage": null, + "ogImage": "/benchmarks/og-open-weight-tts.png", "toc": false, "headings": [ { From 7e694c7bab3f0f7c372f73fd12a5b28e17c3a05c Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Fri, 21 Aug 2026 15:43:08 +0100 Subject: [PATCH 03/34] Import the export that adds Orpheus 3B Q8 Thirty-seven configurations now, up from thirty-six. Q8 is the only new row the page shows: the export also carries the three NeuTTS-2E CPU builds, which DROP_ROWS already excludes because they are the same weights on a slower path and say nothing about the model. Nothing else moved. Every cell of every pre-existing row is byte-identical to the last import, so this is an addition rather than a re-measurement. Q8 needed the three registries Q4 already had -- display name, model URL, and the licence proof. Both quantizations are the same weights off the same model card, so they cite it rather than the repository. Two counts in the prose follow from the new row and are updated with it: the medium.en gap is twenty-two of thirty-seven, and thirteen configurations rather than twelve now sit above two times real time. The cutoff sentence still holds unchanged -- OmniVoice at 1.51x is still the slowest that keeps up and Chatterbox Q4 at 2.03x still the fastest that does not. The ballot pool is synced to thirty-seven so Q8 can be ranked. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- controlplane/src/tts/handler.ts | 12 +- .../content/benchmarks/open-weight-tts.md | 4 +- frontend/public/benchmarks/tts/orpheus-q8.mp3 | Bin 0 -> 22627 bytes frontend/scripts/import-tts-export.mjs | 4 + frontend/src/data/benchmarks-posts.json | 2 +- .../benchmarks/open-weight-tts/2026-08.json | 103 +++++++++++++++++- 6 files changed, 115 insertions(+), 10 deletions(-) create mode 100644 frontend/public/benchmarks/tts/orpheus-q8.mp3 diff --git a/controlplane/src/tts/handler.ts b/controlplane/src/tts/handler.ts index f00ed8b2..6a813438 100644 --- a/controlplane/src/tts/handler.ts +++ b/controlplane/src/tts/handler.ts @@ -23,12 +23,12 @@ const RUN = '2026-08'; const CONFIGS = [ 'supertonic', 'piper', 'chatterbox-turbo', 'kittentts-micro', 'kittentts-nano-int8', 'qwen3-tts-vd', 'kittentts-mini', 'dots-tts', - 'qwen3-tts', 'kittentts-nano', 'chatterbox', 'f5-tts', 'kokoro', 'xtts', - 'orpheus-q4', 'nt-2e-q8-metal', 'nt-2e-fp32-mps', 'chatterbox-q8', - 'vibevoice', 'nt-2e-q4-metal', 'cosyvoice3-rl', 'chatterbox-q4', - 'styletts2', 'melotts', 'omnivoice', 'fastpitch', 'cosyvoice3', - 'bananamind-tts', 'tada-3b', 'zonos', 'csm', 'tada-1b', 'speecht5', - 'mms-tts', 'vibevoice-1.5b', 'bark', + 'qwen3-tts', 'kittentts-nano', 'chatterbox', 'f5-tts', 'kokoro', + 'orpheus-q8', 'xtts', 'orpheus-q4', 'nt-2e-q8-metal', 'nt-2e-fp32-mps', + 'chatterbox-q8', 'vibevoice', 'nt-2e-q4-metal', 'cosyvoice3-rl', + 'chatterbox-q4', 'styletts2', 'melotts', 'omnivoice', 'fastpitch', + 'cosyvoice3', 'bananamind-tts', 'tada-3b', 'zonos', 'csm', 'tada-1b', + 'speecht5', 'mms-tts', 'vibevoice-1.5b', 'bark', ] as const; /** diff --git a/frontend/content/benchmarks/open-weight-tts.md b/frontend/content/benchmarks/open-weight-tts.md index 628200db..7e077941 100644 --- a/frontend/content/benchmarks/open-weight-tts.md +++ b/frontend/content/benchmarks/open-weight-tts.md @@ -48,7 +48,7 @@ face value. **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it compares directly and cannot be gamed by padding. - **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger - `medium.en` was run too, but it did not finish for twenty-one of the thirty-six + `medium.en` was run too, but it did not finish for twenty-two of the thirty-seven configurations, so `base.en` is the one recogniser that covers the whole table. Being the weaker model it hallucinates words onto trailing silence, which understates good engines more than bad ones — so the true spread between engines is **wider** than this column @@ -72,7 +72,7 @@ face value. **Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family. -**Nothing is filtered out.** Every configuration compared is listed, including the twelve +**Nothing is filtered out.** Every configuration compared is listed, including the thirteen that cannot keep up with their own speech, because "how far off is it" is a real question and a table that quietly omits the answer cannot be checked. The **Real time only** control above the table hides anything above 2x for readers who only care about what can be spoken diff --git a/frontend/public/benchmarks/tts/orpheus-q8.mp3 b/frontend/public/benchmarks/tts/orpheus-q8.mp3 new file mode 100644 index 0000000000000000000000000000000000000000..d5d00ed5cfed8ec4dfe8a1d538583c8f2ecd94c4 GIT binary patch literal 22627 zcmZ76b8sYW)F|)?H@0otwr$(CosDhVHa6aDY$qEV8*l8)-MruZ?jQH5nwsvZ>3)85 z>h#p<(~q&fa0Krz%($f03VGjU+nL66rn>)Ax)U2#siHzONtR0Cwja`Xcy&c@F z%w4U0%*}{Qyorn*h=iqy+|0e)=-nLYU7gI$O|6LR9nH+`7>K0Zi2fbw;OItVOypv2 z>gZx-?O;jdXz_2?%H96ok^jDKY-{eq08mp^75_Ia!oO+dj6Ezkm>D=3n3VBz?02l3y_4FJ&W2S7r>z#}4~W8vTt5Rp?-)6z4sad7eS3yMle$;c@v zscUNM8W@>bSlik=x_Wr|_yvT7MMTBMC#I%n<>VEVmRDBSHng<;=|3u z{^yiJyLWfqLH}a=?}UQS|E3_(wuX)^hC_&^zU|gSzNe;XHFI;jx#7ZPO0&_BvTUN4 zNg|C>p5u0i>KGEDvstv=Sg5?k`|@*=LeGdNFT%06%;$PI=*Zw< z;os8N8x*jk_-1ODD)@?5Nv>0@`T98lo4N>XU`^X!2lG1Vepn>T8X^Xqza`ggn|_|j z`=fR5+q7GTy&dj6FT+S!gJa>P{wW?Jq)6QViPoBu`r~ptr;{#kV%Wga6P6XL@CHy$ z4mat3agBJy?NS5%;kevKsv@>J@NOY?VZrX@*dd)!Kf73w%Gv~y62_^*D?bf~Q0-0C z8zuVX*C#s|;p$%te-BgW7b7XhJKEFA>H zd_^vDmKrmmN0ZqfQu*^s9YRu}-fUo(h=Zo{_fTAdoXRkj$zz!e+y38DO3n&(^{_DR zTfcWg7VsfBmeQMpgL+-9C9UqTbse}w-j;8b?(?Jdv)QV$2! zf@#n$bO1PU6G>>B8UIF-RtaZ)@TxPsti4-VII z(3t9Km8ACYL(#%xCV4L7Vt(`T$U7CDZ18HPia|6aWB00+xMIj*G)eGdhr<#Fk|;v$ z&YdvTv12&wLzhXoiIRZtN0XJ`tL$G6TGf}4_4u;xyR!pZ9@y#X6@kDqc}1hHhy@-KfeiAy?v(1L{eCZv`&=;2gZMaSrLT&N_mpS&G;iBz+z%KNkXRFr=uNTKv zQ7w&iHcxGM5wwK(pvb|rKB5%QjjZ4(84zj|sHH+h`_6C32)(ytS*ScAIDd@OB|6c* z{F9U=#cY5DIc4ZIUEx~-?ZRk@bVKwE@t66Hi)vr$Cb8QF#_UsWDp(pRlNdi?GK@c= z$q~77Kh&n_HyCoviYZcwldFnh>B2oV*tuz!^{_2QP zoH$Spjfr>5T`Tm=49Lzn{5p&I`o*JW$kydfAu}?Ad0+k=8r5|sz`C5gR_YdpX2lJL z+~GJ+?PpD|`uft5t4fE#_uRL#n>veFgT_{ucuR6B3a3ul)RJ$(;~G2k65?Z@BMg${_lCu1iPH350G zb5)uOS*ZtZVt$UrdwHToDwlbZ=-7$;@Z9rpOFFflKZ}+KLNY3)eoeNd5R*R?l-{K< zvzcWgAIwp7?k`oSSlKJT{C{W;!kKnA;Iz}g3N<8@YA0QDb=*lc68h*H2Zn4JF)5r? zK;5EjvE)>-(Aa2`YfJT|n#7A^Q)mPEYS?&LMKkQk%omxp2nt`BEet}Vp5GJC$?E z-}Rd!I@EZnXYO8C)&wP|xB)^;a*#^o%R!3_kpnCPzNDre>_;qE^_J6i+mCFXv0wgc zmLVQ0z%dFMV>VY6Ts|s?s%ZW}CZr=Rtf~TQB+X0*xkXvr<|yZkkp|b zQWX1*p6!icb$H-u3tfI0F;(w9Xs?5;Df|UsZL|Qt{J-f{*XaO@Q}`%8u?vzIkCNBI z3lwQhfr_ske>UuP#Kt$0;RB*vfbp?9@XI{eh>&Fd^i3N{Zj_p+NLe4#ZG*#J_WTHQ zYH~Y`+>VJ0hNrvOr>%MiKYom5)K*xV%q}{<#Ye2C-(@Hs0jHv-y)`|zFR|7t)XCz2 zx13?>cORG236e7ry*Ul(fZcg^O^CuHS_+C=oN+0%S|#;EhL2VhhbrqSm#$Xj+&I zu>4a{+oo2;-G7S;r&3qPv016(h0EWXXO%3gv~(>n#$X?<=FX9|Sf%f)(Ne`epbBez z>jYPxf$4=sL8nX^Ue>CS&y)nam03vUSh9M_Db<{AuX&&#OJi&j8e&2AwIi*nFF%%b zAlN@RKP0TYRg^(4r=<Zwka|`?1`Qnp&H?;`c;F^E@fwKp7FMXn0Bu=5tFicuy6F1of z?SCIvW>NDR8Y!$|caRpLJ%Bo8U_Dk@#h*oz;Hp*}bqe{L`{nVDrl87&3?5hz+1gDm zeNJIfcJRQ<_w|kzKnf!hnaOn^KX$`d7a%^6hO?XmWTkmeP zC!Sd`7zEe;R2P#nay*Z8FJ!_(qmbJ(A+;~&Df#kG(YAsz0XE?DW%!HS{NUY=l7c+n zn4%TuxUi(ykwjXOg|&|pxZ!(Q&(36TXH*_f_x`y^yPYK*89mY~`yysBe7CF|6K)Ar2TF$swC}v{umpsB<=asx~vC1y0+q zav5=#izmEuG28aONxWvU%U}LIa$XrO zp?~#{+P5gK!;gJayq(n-AJ#$k6NjPOUvlPd@ZKZ7>x-pb$|S7(6@Jp# zVKQXv2#6zAMkYw%dMqM)LlZ~t?5GoN>n#)Dl9Zr!hX;<@_e6@f(}%@%!&djMhk~F zQ&zwKBZe0>HZ^o#d>?&6b9QN6tzFA4$`Jiu{RU(QN1+02pnSZj$Wl-l`DlI!;l{LI z@2Io|{NC~ipIE@>$f$XIE_v zPMe{b2&&Z7cMu^-)D3-?#R}(0sP%xLFtQO_>iQm7E(=tfy>2<=4f~ylPe!vu18fq9 zbb@eoCd_Xtp;#s(5Vv|1L7CK&2RDC%R0ztv3=l%zaV>iF3Y&4zHMoR}W(uq|L>Z$* zKC96OhbB$cYaqYkzb$bPzy;VQ_}DeVffQpgvZvaiA}WY9X@WD8ZWRr>Q=9p_U7~_H zu!sDe^hxoyKzX#B{V?jT66^$#X#}`_g@WSsC1eMi21wE zD~PWsLlhxT0 z3$lFuj$1h&gKg&|If3P9v&A724Rz;R4AqQbqHMhH&QI+k zL&D3rN4;=1c36KyPMt%1N7h*A+5|5cdl^md9}^QSEBkV zJ->8X2EX^od@O=4GNo$+Ac%Ph#=Ad2ubwuTAAh9R+h`m(%KBU9ruEnosO=*?7+?hc zp6Ljh9)LVPP%@G30pw>M+V`u?y`&A2⪅)ruHLmsLUnP#&_j%1vzFKtrMePQ}6@ zjMz!bq0v z2NJ`&qV(|R$YQEsvvf)|naFVQsB62OVhkfTJ{Lwl8;un&x{PY#+&oHfZKI_X{kCl8 zeudAZUoe|b275@b6d@TP&t^#BQE@m__fea0B@sPuqj`@O1^2$P)_mXb6`=gg3*p88=@z3#n|yF0cO6w*pgt7=)t)DMYINLM08&qxa5qw!$5@ zQJlJ4EEjx?@S8rbmjvyCiIVV=-p{>l0kcvZFPw=!{#85yYALwH)q@?BmeIE!9@6f9 z^zXbLo$qDeR3lV_sBB6N+z#r=pM0x#zWkVUd@Su6S_SymS>anr3a07-7SqiF&Ae9E zGMwkYX7qy7W`^wQnz~4hSpDEaDI>Wm+Ew%kcg7kL?U>2Ovu?xuj4(R>yX|ffsACIdMrA|HYn4c(yF;F9%C(}Z@_L?if zZf+>}Aiy)>J?Pi+-m*q{7+@HfNum5+lI3VR^4*dc_f>@9LuW81Ll;xnB1dZRq}NYG zK=$>dD;;=WLz*vdMy+5SSp&6BnxL^l3-lu>#z9pk-|DOKN~Ib z0k=?RUoW{V?~WybTrbtV2DoxtC-cgHU76#F60S55GLYlU-vz5lpo6V<- z-!wF>>-F=rVmTcyb_VmlOijbea#nxbj1RdUEX|y31IqH$MPQ_%g<&J;J+lF`ByhQi zb<{AZ(M;LRBBVLsObf^qPyw{mIu8*COKIrQCsy@cRVS)@KR!sUu0Cj9l79c3#{(R6 z3xmJaTR8sdX!r)s8HR(27pdg|{FOl)Pf*R~B;-;@4{ZCmc+imW#Bd(bMJ;RBRR8EQ ze3VQd{xAOi~$Bw^MP1^00cQ3W*+A> zfK6t|Iz|;8RfNnxKgMpj{9AQ)Ws*0_k50jaj#vJW0AKY)@G4SSXbMNxMefLqoguD_ z#Xn4*|2iXgOJdTkE^dhTAQBjn7DxA*S*c4JTW!~*uTlOy&B_^UDZ1Q(A>`*ens^8!?Ld;ja-4wKyJ(xZTG`R*bwCl|98D7X&u58yZ5|0HH7daDL$~N;xqy zXTE+4kX=sodC)ej=&FFcOa?5b<=tP8HrZ2`WpJU>a-ywP@*r@W$=en%7?CJF;{xs| zJ|wFD)ejfc8$5n98=7=HA4i{u1%$x7z;i@H>=-#9I1yP>3bz3!?O^OH%CP`LJx`9! zuQC#(TCdxr>jplW;}!oQVKKSly*(ofZptfT5Lq6QZr1gwj-S_$M-467-Fg!HWTv@S zKy%-WDRG%B)qKywVEVDPK$(8P$O1jcFNC5HN>f=3xGj}JEFtz&p;}Ui4!JoACT1#X z4=%e0+;Na7jnORg$K250XW;b|rr!M#lwBi%_foAccWPRiC7X~UD?sPWJ{fCbjq zNr+T^`70$4d7?fiobj@j2n^V`IAb!^s17YayA_!%AezIa3RgmqoT-pUf@vo%@o86O z8H3B|l!q@{9=2DBmn&;j@R67HjJC;l$H`!8Y0ynfP{K4lQ(Rz_J@*mqjjWRYIK&pM z8b|$Yv@HS*KX)x`dk)`OyKrcNNICv==|%%Ga}VzeKsJl>Ix8;;7uLm`_UUY( zvW(iDAsa}JWnzd2qLk?vN$Ct_La33}z<&&?PeQnz*A(q<3x4@|*;9Fn^}M#Ld1g2HC@&h)92qfA2gqIOPi$b(!;*STEGBr^BbsXP#m=v*8LxtuH~z zr*F4t0UZ>mQ_4efj-X*gG>U06;%_%8&Da^FChq9G(@c8!Nfczp%90&~G0MDR!zb1- z++lfPu%V}A+0mG_aSO$xBh8F0{o#>5?#*rWJg`^QXDv(FSo!eKnT|2k?W$#Ai4WmS$KY`ycw#UhSh!Of@#hLmSNe0~Z79n1oqS56Uh5TiX0 z&QLV6J)Q~kJJV`ub&25NRfSy3J*%xI2IZZKNtD1hpRY(BzBvpczir zm03Gxo5qvRsz;a@tvHhfSYsJU=|Xq7kr#kF;;+dZS!>#^SY76qKIFKg(H&T`JRg_o z%>+^roC=;ByN5?w1@ljeNG{g&%dbS{8{dK>mBoi(p<2v> zj5edtT6B=W%H`Kd1G*$hBYnxD4KOS*~S< z#r95IUa5JTx2d8}mPR%TW zCaKs{S|CiVtli5p?oJ4;d}tGk7}5+2&!+sQD2WL}%LVTn1Zb>~y$p5$PQb{EDh^LC zVQ0^`v==YniF+VFqh(yqNIr)KZ3-z=PR^|{A4+s*ZGRiR`epqg;sW~mSR;4-X6iJ1` zmp_ftRip-QaiS-(SV?o?FT9d9DBNQ*i=oRDTXj*73=$akgRXB$a?#UK)KP|IKuf1+31z3z7<&ECCej{%?8XFFw5(mFw?c^)6b7MT1=vf<&;u>)WJYVzWI+^FzZnJrfQ^XYkqJ> z#rL+39IdtVQqO%Gp1A!!@WQ*GI=Q{J>O8e9xdSWAYgUcFzp6CiNsX?$Th zT+ns)&XW-HBIgMQ@EgL#Hke5f7g_9wE(}>@m zSq3eAF=C4(4wEJLX?@RHQnA&gXeI6YuvPiDt@^V5ji5a%jN77gq>sD;w@t5iYDZl$ zbM<-VgqHIx5!3g3T5Khq(r}%fz>csHNC+BX%%<* z)x_n-Z{ID$&nBM-;3an>(@ECqP;M*%FSDll*^ShfAINxa;52kF^&akX>x&AD00J;V zt%4c97c72>xIW?-m6h=f~MnwSh;}WY8035*p+#ppTD12)a|SN=x|ofg@yX~P=md8LxM?_zARy>c~c`&sST-cw@p=} zwI}NO0t%C0x6hLv5c~;4Jq_(U&@y2^NY`^Q6XS%U`c`YJ{ShtWSKEIfWIZm)pLlJt zU!xRgZylAf1Gm`!>;K#eUcmSrE)w+0cR1eMeNs!MN_b}ofD(h0nE^B$7ZlMEL02;! zCSl%Wo0sWG5Yji(w13+4aeG4oe{|KZoV-#)SR6qFXv)vh6 zZ%C0=$NA7fr&T%)t^ilXSls@`JBO=~$Pe#rjp@fdIK^_P7FMJ(_dQzMpIO>R+c;0n zA2W0kH^oH|*ldq15>+{$+x= z`|)Cfi_;6I!B=i81razB_m+&+Cnwm9dOhquJ$KTC!^WX=0W?0n2Hxe;&6(|l5Mcb5 z(d8XpDZ-%2T!pFg*wYD;%_ec@1iu>y_XLvRfe$RI-9t}k1>fdGo*#_WBg4ay;1V5- zv*W?KN}FYICCe&v1##<(E8j9bXj{`c3f`i+VnSlKd@UFQ(Z*b+vjZ{dEW>^hw7QCF z($PZSMgGLCpkYfSJRM6Q?BA#x*rXkJEI%zf{4f7?$xUHez;&lf@9}KN$x0R!b;lPO z&$bGFPLn3>mfozMq*`W%gS=mBbG+KM?{CG>d1(jj(s)IIXb^Noyu3=XW#gnNOKH8S zfbg4m1NucKb8%z3_TbH8Kv5w|M#Az$Mr?mLF+X$#@-x>4)_S!L`|pni3$$E6$eW=3 z-9JW~uX*5o7i-1zU;BGMCWlgn0|G^0g`U*W10wW?1OiX5C^fEY(V%IpB#JhX{dZ9d zzwL^~u}b3>K3DpG&riLmRXNHJec^{3p}idi%}?_Su9S(HB5T&*+>12_$3V(5({k;A z1x=)>!_zR0Fq0ECh44u>X~trybpU*unDPZTFpy|E3#FDMvYvZB^=smp*=X~wZf`F} z+n?a??}G~A3rh|{v%T$Wr#S(-ckB_T>T&j#7~XD{eb>Cy(X40fkdmEoyW=25o#zy01nrCXlSN>P~+WApSZ9+YY6mZ z5rS~*YxO-JD)P`g@c>D72h>dk3;P%I$8At6Um&QT+5 zNxMzbrj$6v=@W;ALDG|sw#I_kB}q6h7X!3LafcI%kX=9uRis*G6ShO8w9u%7swjeL z4P!)wxoTSl9e+YwwT_jNFGebpsWhlNn^TFr^@OX^#=B{1mXNktX^GVxqP9R-GTUU) zQ=lOb^Tx?N-oQ4@ED4x(cX8f`<92#v=;xs65trOHAypl-XYAS+wD*_ zpDEygtVT{)UYzx7Eus4V8*%{XAjqBZjzgjFCT>d*9`WEr7Fx(pnSJSJz5r5Pwsb|x zw9jE<;!b@&djzFOy+P2r;kYmV5-FHYn8E@C8&b&cG%U82W(I7BSE?-sUh$fJuFCJO z5{;ezF0a9nh$uB`Md9W@->pfXOM#Ch=C|B%{X@K1QsJt3-`A#+eFP7A#%-_^bkrHP z>@H}O*6M9n-f}1J0=AkkjGAk{eV-Pocg~5eh5*_KE3np6f~m41%bIA*Up+0M(?xRWn`0Pq7t>b`yv&6Cnh-s@A3*# zV(P*VFQHqICkn8+jGjNZ#)IcHO&X24u$buk`+Jr8Q}5u*4`c*uc4F9=$40zkZh*G4 z=8_#BOKd47stip-4$to5;N{BE#=cccvs`e=^YwEKko9|@uokVONniDIcLPCcgcuDY zN^A@gvL#g}QyeAfWe~nHH$kRoS;>h-1Bx zWJIW>S3Z-TvG^<;TqGz1$b7%x29!)g3w-TXXG|K5mgie=h zXzE+Z?{6}f_{KjwL9I&uTYvIMo73#Zu`bJ%wsQS475MYb)1MNLv*7&fkNG!!l3+pt zjwxbrF^QojG0rMGW}wCEtY#;71F0rFgE%)sQoD_2w+ga0#b*0?nNvh1@B_1iJ1M3` z;GC@f%<4vjAHVMY!SKOB0+woZt~!nT`i`a_0TOa3b3z^BO~M_V2^onRe##@1T?t?Y zc>BR5Wi=A?jrlq_76qOFkG9f#Q63A1QZ(0mc2*Ukr8AI0h$d^3W@*Da5aZfLW;*{X z{!cVHU3Me;vV2Dy*_S$%kB#AdyZ#CiyYAxBW4Cg|@|EZmpk9wQQQG?sb@bY!u8!^P z+3xJK1U})%)6Noiv&u8s%-XCdHux6`2tynWwAQ%Sq=wS8T}jHp;V7|6a)A!9^(&IP zy84OKAxut$vW&nKp`5YHWRP$;vEV58dvYitcnq|27@d2T>gSKF-w?osFKC0NaEq{Vppe;&6f-gKpumS%e1iKMHIXOH zT6V1#?*BEdw4`Jli1VPoMS%aB&v+a~U^;MYm*UH5w|L+!c-Gv8uDkqH0FXfI=VDBX?8_nzU5kC2*OiVoM zVYw@MW!R00VFVGIrbSdnz~YiU9lfq+|Is$mNxHF0n_aE>1UT;nu2LG|&XEFq6RFEn zm^DKLNUCZc+A`{(#Wo$S3cX)y9m~e&B;sO)NnB3cKllU_m(4e4Fe#3~g)O8}&cAEK z!0wISu!SHGLnXx}#Y8)ACE{au zkdpPojAQ#TFkC|~D=R1v2%OFP9r$!Lon{Y-RQu@cd=8x?je@f=e8R&dH!zIeEAjTL zWh}9{ttKrtNiCt^r%4El4X^6*26^-5FSmEhGcvxOulRpwZ|RFl9MAuVFWO~Vgq#I8Hxa{BjN-V@^1A5|S*zKj{DRpHnJX*9_3hCNxi;tsS>rQ?>zTU?U|2S6Qu*5%$AN$XnHfYahNQn%c&D4 zZ`vM}KIkl<($hwlDc^qi4Jc@%i?zHK-V5A0UOPH3mb6JEys}Wp!-Q4F4^{NjP>>-8 z$%<1MlB@YC%CKd!dpZU^SVn8|Wr-n_*v8seQTk%*=#T4(sgXzt06U*&7E3sFD2R&?O4|NT;MyYDetM$);Z9zXgjUwv$QqTvI4&Q zGLkf&DdqAL7j#t8B+HZA;H}zwY-%7-$nlyO)YD5KQiK|V3sArcA6AA2-9^WnfBC)0 zLc*P%Cz0r@$it8L;jx%7QN3xAlu6;y z_{t{AAf-OUlB8T=rVm6Q)HIQBeqo3Rt)LHsn*fv`Zw4WEA$5>1VMkywxaEwEjd`B- z?jBRw4g$7j5kQJq3Rx^upiIUqth=uAuda<38(sW{IPiq4vhod~gadJL+vV$KEe2sv zUPO%fffkxNZN(S-MU3c^E~y z%-f>B{xX53Z#&legff6FIt3O0B-#Ipb?xJ)~_)*HKVcV;HB}IXbA~5x5o-7 z3e#f?Ixg$sZ&MDwdm-E5iwvGp7#D6LlPqO&T47AAW^kmWT(u)eH<&X9=?c%f5u?FZ(6w}!I zepiKeZZhSTHE`m3e)+RGOBnq7w!=414-K8AgF_5}Pm`1jVn)gllA2~BvD3hO{jn-@ zd5XYK%gSQY&`@)GQVKk zq8CEPw;_r#A9nANXocOLJs+f{u4KV~#_DqhmNxs_r=s zt#AqtB3IzD_PrdeET}@bu+BmC#vG-ziBugzBbcfDuu~&M+Jp>Wlnj2?Zsf~fL%#0d zN4GJtgapR|i_D1bORC5U?p2zKCRRH3p3>qxDTDerqA^aV#Dydg=YDoKh}!z0Xif%! znIv!MN=)eO&kg+nim^XFyz}06T97X#Zt!q>3G5Yg?o?hM>91=DM4iqdu z=LI86FV6edc@@+J;wJzd@A>#mxQu~oe}o|L{TG!$zro|zw=RG|$F}}v>rWd{bp-*N z2BnW6vP6>dn%h$y_o-lAoiN`~#df!A4uATuei$JGqX;9IkXWmh-kzk$?oPqzM}c4r zRUjPyeMC8O=F&1_LgX}+DupBYBJlcRc;1jQ*R5?a!7Yr2D+3kUdhBIacS64bw5h(` z*m zMBAOm@(2T`5$}go%lRGyAsrN#7=nw=FRI;|AA&IZPhZ0R^Vjz}o`=h&M z>yFL-x9R|h=;l}aPw7l%?8x?~?F56bO{hQnAfqRV{2`_S34!mx_e~WzI#J7nk3Jyq zgY9jei?6#P$wMAeR|o5_vB@JN$O^k+jAaNX($qdqO6x1j>ESOTcTj?f#^Bh%utlS z5l5^@%bivHIZN&C@;^r1-sij0j_`Ytjga0_^|RkTWCb0kIaaUr`5e|9zWis@tRqnf zi%6|IhVQ_O79W>G&lhd!f>|AM@o7<4b$uC27mYRDRTI*j%{GqT+9+8}Y9iziMnp?J zW|!wSt=V?&gokk2dqJp8XrNF*s;eU0(K3h~QkBWSLMe41$ylswLm-ifWgw6v<^Is@ z-jPDbLwSK3fd~_;z>~@X1cdtWRh}|QgK`}9LTW+k60;!3j$mR_?a@f^N=kylRzq8a zOHjnnMAg14hHh*8h7SpoWWb_&*xvXB15Nbhf0kq833{AwerM=LcAU%1s``y#9IdrN zTtOVf%dp5EUVe-pE{P~XxS}vYJWe}u-r8OeAPC8oQh!+ATSZ6zT;&ZeM&5n`1vp=; z;q;N5aOqEpVV4DehL&{^_Yky$7e7$`lGjky!$;-`yocBDmA^E z$`h=-qTzZ!VEyE|;8dsjx~!_KgCDKP!fbt4)Y`;XpxlWq06Wh4Bfw}JT^jAmWn9%;dIQXOp6j9;q&f<3Ei#Wk;;TS*B-~0+=N31Ea5MlJm!2(? zUDyic_~d~`=n6xsT%8cR3PVw_aocqJIin_j?_NOplhR*W1erv-oHii%cpBqMTF(XB z>E(W+tS45M4jwRsfq7%8)dY44?%8(il$S{_c>XDs3CbF&9S& zQ+{rT=1?L8Y;Z5qFvQz4T^ik{l)|0aPyhDb)4z^RPdYA>8OO68xxGXFia$HKZ*;wu zly)_{vdY71&+Uf#s9UW+Vv?B=D=NdMlAIqSbz6XK=kJ^ui+kuB7&O_y*&A0F#VzY! zpK>U94Xi@l7Bx{QQuOG_5z}F0DYHd0Hmb`HD!q|~EgxFeI@=x|b2)lHxF+Z~;wb6c zrFG%)7;8C1>$lWQ;`yD6&My1amTfyNBKJM-@tAY-H9Ubo1Cydv+DKo16|w`Cb{(U|dT^z9xQ5Svb@?NAU7uT2 z6gX-jhz|KT$|f3HXXI20Q3{zBit@DjA&u;3j7;sGX!`MY^%We*4muQatuWW3D~U2I z2!pgdBfNBVJf)kx85op@obs$llzj8QVj|4an29B~Au1BV8Rj~3Vebxsvw_vBey$Ux zxv^4>tGJL9o!ns1^$I>qCNN&=p1_?`zS!v|ae+MZ!)*)O^Uv44Y@2cPf7vI1m_z1) z&%@Mj2Or${<#(on6-ZZ`Q0m#Ur9iMvx}PEx=%=$oBgA7@*{!Le(pImD`K=;cRYlXu zMU$4N@m{sSG#7*NAX5!sg~Y1`+fk((L&_W06r@73y4M+S?k$@vdMk%@@O{X8&ySHP_nGz zeaXxD6>|9%;R|C!I@Rh7Uf-KROZpE(N11XfJ4ifR28H(BPyT9HHf~FNtnnKo|zx);Cs?`B{izA!p?|8xxJN^^kody>icVgxm zsGaR7QtG4(b7MR3ZNqoKp?Qdq< zf9wB;$i8oT$Zbq)+RWW%==*B~C_k|iTVYGYc`PfHX5sU)Mg<>=2l1N0F>Gc!=`qCh zDQw>taW9F+!|9M%MP1$Chr<53C4IIq1qycN5t_eW|J1g|5fXA4v;+zS?q%OTfc1`# z?goswD>|p2Y;1J_XcB|WUz>%&ZY=7Sy`qV@NW;GO&|6n{Bu155LQbT8>p4PFL%;uF z8LErI7o98?Gyw$zlIV_guE;W^#&wA(i4F3pju(pOKGvW075`1rz4ZXG35hMy`4wJh z?cxYBDeZ5<&}qv`s(P*4-^i?aSxx*&qCGObP9GS4bkA(c>`hFhCPC9KKquL zhvoqbOV)CJg%+0WFV{j_u-MA0fxYxo9HPxXT;fENuq|8Q%P~@Qwzt9-R|B1CGbpgK-n5 z4PX9Rrn{N)NAJwu75wzB)clW$5{GL`W-0qK2Ed=Qh^OpwcWEo{a4rD2$`6rdX|_)g z0eEs9ObWZl={xZmvQaWyWoN0z5T&vlZ7K@28RLr6?F@mKi4Fsab$1D*A4} zI!Fzx`KQgSpF5o}_%Hqs3v|fg$?i#Bbx}2|9WjtXd9lbZe-0U~L=A!uL&l9ncPn#*YGT9K1b?s8%tGtYe% zT<1Qy934s&3OsVZl!n=me;9a@vYQYk92`K+-AE)+Eqc3s&@>_o=W5d@g*xFLEe@T@ z!DdLSnSZs6x5*O*9SbArqnJk2PTx<{CkO=J2q6BTkslBU=Jd*T;%Ip~UUDf3-P_Wq zPf@}%j9qR?Wnj}1f%Ho9U;g+0^A{gVO-@bPXa2r9x3BHF2n|v=RE+mTQ|mWO$o0vK8@A*|c$#%n8Rk z`_l31g#0oi72Pd>!bf&kNVcS~dAG(wablk=4S^!{V>xzVNIjEA2rO*S{3r7N>E%45 znrgQ^PHzbiAaqay0YapAkX`}_p?5@j@6wd2^p14tCG?0$Zz>|aOGl|vRX|jV0tyWH z&U$CvnLBsQhr8x{I&GcxJNxYCJmtUlE)a5k{8%3vIlw`02UqA7lCVisP>QTsHe3*G z>cZ1&iGRKE7LalD7JfghgJXk-gsz6d|AWJy{I6B+7l>q)K=e{cKq_bu<`hdv28hD5 zv)6l~Sw3N&k41usQU)77C#m0DqQX}eNMx#_M`c3#o^V#X)Cfr%iESmvt8JB=d{vt1 z6K9xJcs0%6{?5bF%&U1k^a8q(#gwYLgO5YK2byHqYVVB?e+tN9Jja{mKefj#5)VbI z^g)DflBX$ikPP2onzPv{Sj5A|H`)yA%|#Hgzjwa78Rd#1xvW|X>WRw+v4N6Iqne|4 zh^YW6Wyu*-T-W|{cx(O4FLFvv)X)avA6W(kHtD29&e#Kz_0>R6NxMF0R3O0AR2g&C z8X49a#^afLR!<3-4Rb6x{k{sQ*nEmpTekOk3Vx87F;mSsn>ySvYjIRsayTIeu6%FV zF|+RCaEmw+ZqOVWD12dq96@@TgH;FVl_c7r$(TdI8<71K8Mf! z+sZsuOKcsFoa2lSCpw$0xJ&&JJwbUGIeMa0TBax25vGIzLgrX0!P&jn{6Bb^5pD!i z8ce@NFbReCgaS~IZ#6G@l;f@!4J`PPCF?4vEMorU z8o4GaA2pen+P^z1?Eh_JHk$HT;N^)1I|p#t$=~pvf)diQUh|*uHtbswOas8CL$O_L zD0rRRT@HlvBZ3Smru2=&7#tC6rmOJS;z)c#>8(~@ZzBD6H->U2MJ3d%65T6HB5*|K z7XvPz;{MhWu~OIhj*@5E8dtG2{x-SOIoQgxcUvOa9G#>EJ98QrbKjhICWO(_Y{T~y zVV^+3s$P6ESm$g0!5fFzSn(c$WC|q&^T0Z1&UC@&Amh;GMmbeL;d72X z%iX{rez5Ih#O-~i0|j;Oh+IKAjD3;sNAHZ>9`Y5M1Qwgia#X6<_>ERha}BZD9I0(j z7phPLCj=>4yDj3w6poN<%1IqeGN;QfY3B_7zZ;%ufxjcUm=UZ4YQSsya#z8i1R7~o z`aJvMBq0MfsDAT&p2=q!2 z>(}A^e4dqZ9t8vmAv%s;t$}A4{8l8#Q|{Bu-xUblc`5k}e6^JSUHpC@QlM2^TRD`@ zX!fTaf04wpNhEt|M~M(p_fx!AoBv2(|6M!15>dI<0bE!u zE-XK5DalP49jAV4&$dZLZw)6)9gyZ3LKfs>Qd)N+CiVLcGnW@bT6~3cUADFSwf(oa zm>s<6h6!~v3)5x<=W-8PH74K3l)7ZSgcz_+qrMBy`T1#B3?^mtQ9j?Hy4<<#;OoMZ z%MY#p{mMb_7-l?H46p%dF?e#5@=6M|d3;&vtCCY9*=hpUb|y# z&E67L4?bl#KbLk9GRl@@pthuntSTLIQt$MHad(4!qYtlHawB(JqKlC>=kC$5KYJK{EQ~r??N)0h zNGOqf7Tiu>plK&lYg+JhZ^7wssi4f*%yj&*F{vF;J$IQ3B8Aa;Io=|rFZVRh_FF@lXAUA0JF8TYRpWx5 za`a4g*Zdotd02D0>E3J|;pJET-_Y~yHRt@5iw)~h+-Ap|o=!5#LfGCKJC*u0ofZ;^ zVkL0jVnj^sA>8tgP?`Yyc?<0vtI%N#e z!+`FT7%Mjip@Rj44Qeow1|hH9UT<7oFR_VkrZk*ISNdE=9k$#&{krnx!)B*yO0IYR zWvd`o4z}`(H|X;8`TL#ufX5Sv3Q4j3mXxn&7)7S6Sb%gCMD$2zfis?nh{8=_&<~i; z;?mqV5C$N_IZecIiL>Z$_tX*U7mjo06TsIY{V=Re_`%p@n4vVbT)nh{3+Ag*4!+yZ z@t6PK>NlXf;_qlm8Fci0&)!(7A2LPQ`LW3&1S(aYUV5FoE&olc{yxZ0Gm;d{OOX7@-HaBI}RqY2&+z#W( z?zT;!`^0O?7RQ#sge|Zq)8e7})S#;B@~{!9da^R|SJN3WqSXKuZ$ zzY2P8erK zmT11Ie&^vLaqqdI!nY&&t5r*kk*h4}%J5-EUY#C`CRrf**y6b2CWVDRd_)5wn=evPhFEer8jnmNrHY*V_Fa z9cYor5V6bay(<_@DG)`;a7K(b3EkU~s_Nn~5lRq2Lk#9_m=VcK(6DFUGp<}sXke^= zS-ZB2rEtMXeyJPh5~Y?Aj=4)VGnkHa74F+VyxpzT^8xA*(KGRT9iWxp3n_Z%DhS8u z1yI}k!m{bbRFWW4Zq_vVg%jZp?Z}+KzDqq%ix|+pPU$6Vf14s4l?kz?c?n70T$Ijc* zCEwwh=;#PAC;m#(7)rk(Lw%1!OTO^a4uJh#py%eJ$N&ICV>v6Z z*cc^#6c$9J@4gb(S(qYE%Wtd(*`jYz@4HFB5WPh!=;92D_MoMsNGxss`S2e%eqf!XMgqiG%x#P>3{VEOtm>GR=6foy5zb~Bq>BB3W$_8 zHr*&?zT+$DClryVmD!ut4wYHC(NtG-X{UCU~xfP8JKC%~d7M%gB z*sqgd^3fkZnUY7M^$9GqfXLofFnLfU7#IT_Ez>X zxSwhQfl(O?n<}o?^hMgGzH#Ep#Zkv-7pkR8%2oUS(hTc>QOoX+-y-ydf=*%8nIv_> z#&r;~JX`Ijru8N#NIpmGUaI-1!oir5o2SMOeEG*#PlmgrJ5G_iG}^7py07VRSweo5 z8Cx$OPb$E>7)igW8o_7XnC-CWPsTAsHJnV~JE^R#m>R(+iEcw59rH*L=<#)H5Tx?{ z`Tt)E_~HrnjsRta)oVPnV{;LlU0#AMqBI(o#}d?_Wc>Tsito>6={Sg{;JrQ z9XJ&Hq8%+_IZGXL||$> zj1O0wM$@8~CY4c)CM^U6zve#?NVwp|8K)-5E@caI8mQ^?m*_?bnqv~V`M2>i$1y)?SkWV|vYS;G`>gXnpfCV`0wn6*@cvdjp!Y3mb( zw0TK9P}MD{q?V8;EjuMXw1OauhhAsS)_NC_CU&}%er^8^uAo3k_HMLdocqoYQXkS# zHkOj*yJ((YnCa9l$bu$|#a_QfHQ1_tTeZI6<`_AWWP zlR$7vIrZA!SA&;UZ?o-4jKur7U6R(NF4y}ZrlI7M877{q z9Blykv6Ol`wfIO+W$2p!xA<>33~z*5pGC%~f{l?F8eJ5K&f|90!7`YR(9?(!rU`Iu zfd<6O3|WxKAVZ_CsA_f0PS`!*t}8=`xrpXWyf;W z9go#Gt+nacQx~lR%s^QH0gz3FMZ6|#(vZvRgxqTG$SZ~#Mp_ftm%06R{@V;3s7L~Q zkmIE8Rh$A%8u3|sv=?2YaEF>oSExR`yPs766D;lS#z#7Cp32%F;s|Z#@6=z)x|~D#vnqeWo1ph z&)?Z~vvNbcjBab(#*7&ZY*>mmp<)vwJ5PNV_?Zi?Mn>#Y_ypPU!bs4%zel7ml-ocg zQ@>_zR1MBF@+=+*T8Xf*l(>hJl>Kg=BeI^ZsQGvGhrf2yeX!18no*c_F$aEJBTgOXp2k=?E~c=JMKq);WV*V4{JD zrORKU&V7a9+o@1zRdbIk9dCQ6TVb?q>3wOTzz=`V%TqmPZHn>v_SaiXFx_o)a}Q1V z3oX$`e%6B8L+^?tFhQ`V|AVp1nS9fG?H6PEjXz6K@D1Q@z9asGXX|8GzF57|^psQ>@~ literal 0 HcmV?d00001 diff --git a/frontend/scripts/import-tts-export.mjs b/frontend/scripts/import-tts-export.mjs index 8b9f726d..d7f9e696 100644 --- a/frontend/scripts/import-tts-export.mjs +++ b/frontend/scripts/import-tts-export.mjs @@ -116,7 +116,9 @@ const LICENCE_PROOF_URLS = { // weights carry, which is what this column states. Supertonic is the standing // reminder that the two can differ. "qwen3-tts-vd": "https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign", + // Both quantizations are the same weights off the same card, so they cite it. "orpheus-q4": "https://huggingface.co/canopylabs/orpheus-3b-0.1-ft", + "orpheus-q8": "https://huggingface.co/canopylabs/orpheus-3b-0.1-ft", "speecht5": "https://huggingface.co/microsoft/speecht5_tts/blob/main/README.md", "styletts2": "https://github.com/yl4579/StyleTTS2/blob/main/LICENSE", "dots-tts": "https://github.com/studio-dots-ai/dots.tts/blob/main/LICENSE", @@ -165,6 +167,7 @@ const MODEL_URLS = { // family repository the base model uses. "qwen3-tts-vd": "https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign", "orpheus-q4": "https://github.com/canopyai/Orpheus-TTS", + "orpheus-q8": "https://github.com/canopyai/Orpheus-TTS", "omnivoice": "https://github.com/k2-fsa/OmniVoice", "bananamind-tts": "https://huggingface.co/Banaxi-Tech/BananaMind-TTS-V2", "dots-tts": "https://github.com/studio-dots-ai/dots.tts", @@ -267,6 +270,7 @@ const DISPLAY = { "qwen3-tts": "Qwen3-TTS", "qwen3-tts-vd": "Qwen3-TTS VoiceDesign", "orpheus-q4": "Orpheus 3B Q4", + "orpheus-q8": "Orpheus 3B Q8", "speecht5": "SpeechT5", "styletts2": "StyleTTS2", "vibevoice": "VibeVoice", diff --git a/frontend/src/data/benchmarks-posts.json b/frontend/src/data/benchmarks-posts.json index 4548e5c2..813e7458 100644 --- a/frontend/src/data/benchmarks-posts.json +++ b/frontend/src/data/benchmarks-posts.json @@ -31,7 +31,7 @@ "slug": "limitations" } ], - "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nEvery row can be played, so you can hear what a word error rate actually sounds like.\n\n**You can run this yourself.** The harness, the corpus and the scoring are open, and\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this\npage from nothing — there are no stored results in it. A single engine takes minutes\n(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it\nspent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template\ndoes the whole thing in one terminal and shows the report next to it.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for twenty-one of the thirty-six\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones — so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the twelve\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2x for readers who only care about what can be spoken\nlive, and sorting by RTF or compute per phrase draws a line at the cutoff so the two groups\nare visible at once.\n\nThe line falls at 2x exactly. Nothing sits awkwardly against it: the slowest engine that\nkeeps up is OmniVoice at 1.51x, and the fastest that does not is Chatterbox Q4 at 2.03x.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes, feed-forward over one.\nDifferences under roughly two points are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" + "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nEvery row can be played, so you can hear what a word error rate actually sounds like.\n\n**You can run this yourself.** The harness, the corpus and the scoring are open, and\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this\npage from nothing — there are no stored results in it. A single engine takes minutes\n(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it\nspent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template\ndoes the whole thing in one terminal and shows the report next to it.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for twenty-two of the thirty-seven\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones — so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the thirteen\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2x for readers who only care about what can be spoken\nlive, and sorting by RTF or compute per phrase draws a line at the cutoff so the two groups\nare visible at once.\n\nThe line falls at 2x exactly. Nothing sits awkwardly against it: the slowest engine that\nkeeps up is OmniVoice at 1.51x, and the fastest that does not is Chatterbox Q4 at 2.03x.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes, feed-forward over one.\nDifferences under roughly two points are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" }, { "slug": "agent-skills", diff --git a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json index 9ddf6fa4..e6d17950 100644 --- a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json +++ b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json @@ -1,7 +1,7 @@ { "benchmark": "open-weight-tts", "run": "2026-08", - "label": "2026-08-19", + "label": "2026-08-21", "machine": "Apple M2, macOS", "corpus": "84 phrases", "rtfCutoff": 2, @@ -1335,6 +1335,107 @@ } ] }, + { + "config": "orpheus-q8", + "display": "Orpheus 3B Q8", + "group": "", + "sample": "orpheus-q8.mp3?v=1b15da0d", + "cells": [ + { + "v": "Orpheus 3B Q8", + "sort": "orpheus-q8", + "tone": "", + "align": "left", + "restricted": false, + "href": "https://github.com/canopyai/Orpheus-TTS" + }, + { + "v": "ar-lm", + "sort": "ar-lm", + "tone": "", + "align": "left", + "restricted": false + }, + { + "v": "Apache-2.0", + "sort": "Apache-2.0", + "tone": "", + "align": "left", + "restricted": false, + "href": "https://huggingface.co/canopylabs/orpheus-3b-0.1-ft" + }, + { + "v": "3B", + "sort": "3000.0", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "2025-03", + "sort": "202503", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "3.7 GB", + "sort": "3788.8", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "3.7 GB", + "sort": "3788.8", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "4.0 GB", + "sort": "4096.0", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "8%", + "sort": "0.0770", + "tone": "good", + "align": "", + "restricted": false + }, + { + "v": "3.81", + "sort": "3.81", + "tone": "good", + "align": "", + "restricted": false + }, + { + "v": "8.62x", + "sort": "8.6208", + "tone": "bad", + "align": "", + "restricted": false + }, + { + "v": "43.80s", + "sort": "43.8036", + "tone": "bad", + "align": "", + "restricted": false + }, + { + "v": "The same 3B at q8. Only three quants are published and none reach real time.", + "sort": "The same 3B at q8. Only three quants are", + "tone": "", + "align": "", + "restricted": false + } + ] + }, { "config": "xtts", "display": "XTTS", From 5d173a067e33d74e9e02be38b65493a39fbb0c7b Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Fri, 21 Aug 2026 21:34:27 +0100 Subject: [PATCH 04/34] Take the corrected notes from the re-cut export Four notes change and nothing else does: the run is identical, only the prose in the Notes column moved. The counts in the article still hold at thirty-seven configurations and thirteen above two times real time, so no prose follows. Three of the four were stale superlatives or stale figures, which is the failure the benchmark's own README warns about -- a note claiming "slowest here" is a claim about the whole table at the moment it was written, and adding a row falsifies it silently: dots-tts said 11x real time, measures 14.22x, and is now correctly named as the slowest rather than left implying Orpheus was zonos said 8x, measures 11.07x orpheus-q4 claimed to be the slowest at 8.1x while ranking fourth; now states its size and speed and claims nothing about the table orpheus-q8 replaces a wrong count of published quantizations with a comparison against q4 that the two rows actually support Each was checked against the columns rather than taken on trust: zonos really is second on PESQ behind piper (4.02 to 4.10), dots-tts really is the slowest, and q8's "half a point more accurate" is 7.70% against q4's 8.23%. A fifth note was refused. The export reverted omnivoice to "no upstream model card found, so licence is unverified", which is the claim CELL_OVERRIDES exists to hold down -- the licence was since established as Apache-2.0 from k2-fsa's own Space, and the export still prints "?" in that column. The override kept both the licence and the note. Worth fixing upstream in build_benchmark_page.py so the next export stops trying. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- .../data/benchmarks/open-weight-tts/2026-08.json | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json index e6d17950..64b6be52 100644 --- a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json +++ b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json @@ -822,8 +822,8 @@ "restricted": false }, { - "v": "4.4 GB f16 and 11x real time, for 6% WER - mid-pack accuracy.", - "sort": "4.4 GB f16 and 11x real time, for 6% WER", + "v": "4.4 GB f16 and 14.2x real time, the slowest here, for 6% WER - mid-pack accuracy.", + "sort": "4.4 GB f16 and 14.2x real time, the slow", "tone": "", "align": "", "restricted": false @@ -1428,8 +1428,8 @@ "restricted": false }, { - "v": "The same 3B at q8. Only three quants are published and none reach real time.", - "sort": "The same 3B at q8. Only three quants are", + "v": "The same 3B at q8: half a point more accurate than q4 and slower, which is the ordering a sound conversion gives.", + "sort": "The same 3B at q8: half a point more acc", "tone": "", "align": "", "restricted": false @@ -1630,8 +1630,8 @@ "restricted": false }, { - "v": "Llama-3.2-3B finetune over SNAC: the largest LM backbone here, and the slowest at 8.1x real time.", - "sort": "Llama-3.2-3B finetune over SNAC: the lar", + "v": "Llama-3.2-3B finetune over SNAC, 2.4 GB at q4_k and 8.1x real time.", + "sort": "Llama-3.2-3B finetune over SNAC, 2.4 GB ", "tone": "", "align": "", "restricted": false @@ -3145,8 +3145,8 @@ "restricted": false }, { - "v": "Second-best PESQ here after piper, but 8x real time.", - "sort": "Second-best PESQ here after piper, but 8", + "v": "Second-best PESQ here after piper, but 11x real time.", + "sort": "Second-best PESQ here after piper, but 1", "tone": "", "align": "", "restricted": false From 3d796d4bc0808c05f63d2d9639e9de61068fce62 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Fri, 21 Aug 2026 21:56:42 +0100 Subject: [PATCH 05/34] Make the TTS benchmark template actually run, and prove it The benchmark submodule now ships its own harness, so tests-repo is gone from the boot command. Running it from an empty workspace then turned up three more things, none of which reading the code would have found -- and two of which fail green. TALKITO_AUTO_APPROVE_DOWNLOADS=1 talkito declines to fetch model weights when it cannot ask a human, and a boot command has no TTY. Without it every phrase fails with "Piper model unavailable", and the sweep still exits 0 having published a table reading 0/84 phrases, WER 0.0%, RTF 0.00x. A green run that measured nothing is the worst outcome available here. ffmpeg make_samples.py shells out to it for the wav-to-mp3 step and the sandbox image has no ffmpeg. Missing, the numbers are all correct and the page has nothing to play. TTS_SAMPLE_DIR build_benchmark_page.py and build_export.py default it to a hardcoded /Users/robertmacrae/... path, so anywhere else they find no clips at all. make_samples.py resolves the same variable with expanduser and writes into an mp3 subdirectory, so the two want different values and it is set on the benchmark.py line rather than once at the top. Validated from an empty workspace: clone, install, weight download, 84-phrase sweep, Whisper scoring, page build, serving. About three and a half minutes. piper 82/84 phrases WER 5.0% RTF 0.07x The published Apple M2 figures are 4% and 0.04x. Real-time factor is wall-clock, so a throttled container reading higher is expected rather than a discrepancy. The report reaches the browser block at 51 KB with the clip embedded. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- benchmarks/README.md | 8 ++-- benchmarks/talkito-tts/README.md | 60 +++++++++++++++++++--------- benchmarks/talkito-tts/template.json | 4 +- 3 files changed, 48 insertions(+), 24 deletions(-) diff --git a/benchmarks/README.md b/benchmarks/README.md index 4d81dce0..55bd0e55 100644 --- a/benchmarks/README.md +++ b/benchmarks/README.md @@ -10,7 +10,7 @@ results surface in a browser/note block. | Benchmark | Dir | Runnable? | Execution model | |-----------|-----|-----------|-----------------| | slop-code-bench | [`slopcodebench/`](slopcodebench/) | ✅ yes | **Host** (no Docker) via the fork's tmux executor | -| Open-weight TTS | [`talkito-tts/`](talkito-tts/) | ⚠️ blocked upstream | **Host**, one PTY; no agent, no keys — see its README | +| Open-weight TTS | [`talkito-tts/`](talkito-tts/) | ✅ yes | **Host**, one PTY; no agent, no keys, ~3.5 min for piper | | SWE-bench (Lite/Verified/Full) | [`swebench/`](swebench/) | ✅ yes | Inference on host + **Docker-per-task** eval (Modal or DinD) | | SWE-bench Pro (public 731) | [`swebench-pro/`](swebench-pro/) | ✅ yes | Inference on host + **Docker-per-task** eval (`--use_local_docker` or Modal) | | Terminal-Bench 2.1 | [`terminal-bench/`](terminal-bench/) | ✅ yes | **Harbor** harness; Docker-per-task (or `--env daytona`/Modal remote) | @@ -80,8 +80,10 @@ instead — direct INSERT is the local/curated path. ## Status -All templates here are **prototypes**. slop-code-bench is validated end-to-end -(see its README). The Family-B templates are grounded in each harness's real +All templates here are **prototypes**. slop-code-bench and the open-weight TTS +benchmark are validated end-to-end (see their READMEs); the TTS one is the +cheapest to re-verify, since a piper run is about three and a half minutes and +needs no key. The Family-B templates are grounded in each harness's real commands but are **not** validated end-to-end from here, because they need a Docker-capable backend (Modal token or DinD sandbox) + provider keys that aren't available in this environment. Each dir's README states exactly what's verified diff --git a/benchmarks/talkito-tts/README.md b/benchmarks/talkito-tts/README.md index 9be77896..af5341ac 100644 --- a/benchmarks/talkito-tts/README.md +++ b/benchmarks/talkito-tts/README.md @@ -32,25 +32,42 @@ that refreshes every ten seconds. Without that, the browser block would show a connection error for the whole run and only become useful at the very end. When the report is built it is copied over the placeholder and the next refresh picks it up. -## Two things the sandbox forces +## Four things the sandbox forces + +All four were found by running it, not by reading it. Three of them fail *quietly* -- +the run still exits 0 and still publishes a page. **HTTPS clones, not submodules.** talkito's `.gitmodules` points at -`git@github.com:robdmac/…` for both submodules. The sandbox has no SSH key, so -`git clone --recurse-submodules` cannot work. The boot command clones the three public +`git@github.com:robdmac/...`, and the sandbox has no SSH key, so +`git clone --recurse-submodules` cannot work. The boot command clones the two public repos over HTTPS instead and arranges them as /workspace/talkito/ talkito /workspace/talkito/benchmark/ talkito-benchmark - /workspace/talkito/tests-repo/ talkito-tests -which is the "submodule position" `benchmark/tools/paths.py` resolves via `../..`. -The sandbox image installs the talkito *package* but keeps no checkout -(`sandbox/docker/Dockerfile` clones with `--no-recurse-submodules` and deletes the -tree afterwards), so all three have to be fetched at run time. +The harness now ships inside talkito-benchmark, so `tests-repo` is no longer needed. +`tools/paths.py` still wants a talkito **checkout** rather than the installed package +-- it looks for `talkito/tts.py` -- and `benchmark/` inside `talkito/` is the layout it +resolves via `../..`. **`--only`, never the picker.** `tools/benchmark.py` with no arguments opens an -interactive checklist. It reads from stdin and exits on EOF, so it cannot be driven -from a boot command. `--only` names the models directly and skips it. +interactive checklist that reads stdin and exits on EOF, so it cannot be driven from a +boot command. + +**`TALKITO_AUTO_APPROVE_DOWNLOADS=1`.** talkito refuses to fetch model weights when it +cannot ask a human, and a boot command has no TTY. Without it every phrase fails with +`Piper model unavailable` -- and the sweep still reports `rc=0`, having published a +table reading `0/84 phrases, WER 0.0%, RTF 0.00x`. A green run measuring nothing is +the worst failure here, so this variable is not optional. + +**`TTS_SAMPLE_DIR`, set per step.** `build_benchmark_page.py` and `build_export.py` +default it to a hardcoded `/Users/robertmacrae/.cache/...` path, so anywhere else they +find no clips and build a table with nothing to play. `make_samples.py` resolves the +same variable with `expanduser` and writes its mp3s into an `mp3` subdirectory, so the +two only agree when it is set explicitly -- and they want *different* values, which is +why the boot command sets it on the `benchmark.py` line rather than once at the top. +ffmpeg is installed for the same reason: `make_samples.py` shells out to it for the +wav-to-mp3 step and the sandbox image ships without it. ## Running more than piper @@ -71,17 +88,22 @@ easy to make by accident. ## Status -**Prototype, not yet validated end to end.** The template's own mechanics — clone -layout, dependency install, report serving, browser wiring — are testable and tested. -The benchmark run itself is currently blocked upstream: -`benchmark/tools/paths.py` requires `tests-repo/test_tts_asr_roundtrip.py`, and that -file is not in the public `talkito-tests` repo. Until it is published, a clean clone -fails at import with +**Validated end to end in the Docker sandbox**, from an empty workspace: clone, +dependency install, weight download, 84-phrase sweep, Whisper scoring, page build and +serving. The run takes about three and a half minutes for piper and finishes + + piper 82/84 phrases WER 5.0% RTF 0.07x + +against 4% and 0.04x for the published Apple M2 figures -- close, and real-time factor +is wall-clock so a throttled container is expected to read higher. - RuntimeError: cannot find talkito. Set TALKITO_ROOT to a checkout containing - tests-repo/test_tts_asr_roundtrip.py +The report reaches the browser block at 51 KB with the listening clip embedded. -This affects anyone trying to run the benchmark, not just this template. +One cosmetic gap remains: `build_export.py` reports `0 audio samples` because the +single `TTS_SAMPLE_DIR` it shares with the page builder points at the mp3 +subdirectory. That affects only the counter and the zip's `samples/` folder, not the +page, which is what the browser block shows. The real fix is upstream -- neither +builder should carry a hardcoded home directory. ## Seeding diff --git a/benchmarks/talkito-tts/template.json b/benchmarks/talkito-tts/template.json index 239bf006..2982799c 100644 --- a/benchmarks/talkito-tts/template.json +++ b/benchmarks/talkito-tts/template.json @@ -7,7 +7,7 @@ { "placeholderId": "runner", "type": "terminal", - "content": "{\"name\": \"TTS benchmark\", \"agentic\": false, \"bootCommand\": \"MODELS=piper\\nPORT=8099\\nROOT=/workspace/talkito\\nBENCH=$ROOT/benchmark\\nSERVE=/workspace/tts-report\\necho \\\"== Orcabot: talkito TTS benchmark ($MODELS) ==\\\"\\n\\n# Serve a placeholder first, so the browser block has something at a stable URL from\\n# the first second rather than an error until the run finishes.\\nmkdir -p \\\"$SERVE\\\"\\n[ -f \\\"$SERVE/index.html\\\" ] || printf '%s' 'TTS benchmark

Benchmark running...

This page reloads every 10 seconds. The report replaces it when the run finishes.

' > \\\"$SERVE/index.html\\\"\\ncurl -sf -o /dev/null --max-time 2 \\\"http://127.0.0.1:$PORT/\\\" || ( cd \\\"$SERVE\\\" && nohup python3 -m http.server \\\"$PORT\\\" >/workspace/.tts-serve.log 2>&1 & )\\n\\n# talkito's .gitmodules uses git@github.com: URLs and the sandbox has no SSH key, so\\n# --recurse-submodules fails. Clone the three public repos over HTTPS instead.\\n# benchmark/ living inside talkito/ is exactly the layout tools/paths.py resolves.\\n[ -d \\\"$ROOT/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito.git \\\"$ROOT\\\"\\n[ -d \\\"$BENCH/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito-benchmark.git \\\"$BENCH\\\"\\n[ -d \\\"$ROOT/tests-repo/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito-tests.git \\\"$ROOT/tests-repo\\\"\\n\\n# piper needs no torch at all, and faster-whisper scores on CPU through CTranslate2,\\n# so the whole piper arm installs in seconds. Heavier models would need much more.\\npython3 -c 'import piper' 2>/dev/null || pip3 install --break-system-packages -q piper-tts\\npython3 -c 'import faster_whisper' 2>/dev/null || pip3 install --break-system-packages -q faster-whisper\\n\\ncd \\\"$BENCH\\\" || exec bash\\n# The listening clip has to exist before the page is built, or the table is numbers\\n# with nothing to play. benchmark.py builds the page at the end, so samples come first.\\npython3 tools/make_samples.py $(echo \\\"$MODELS\\\" | tr , ' ')\\n# --only skips the interactive picker, which needs a TTY it does not have here.\\npython3 tools/benchmark.py --only \\\"$MODELS\\\"\\nrc=$?\\nif [ -f tts-benchmark.html ]; then\\n cp -f tts-benchmark.html \\\"$SERVE/index.html\\\"\\n echo \\\"report published -> http://localhost:$PORT/\\\"\\nelse\\n echo \\\"no tts-benchmark.html was produced; measurements (if any) are under $BENCH/tools/\\\"\\nfi\\necho \\\"[tts benchmark done rc=$rc]\\\"\\nexec bash\", \"subagentIds\": [], \"skillIds\": [], \"mcpToolIds\": []}", + "content": "{\"name\": \"TTS benchmark\", \"agentic\": false, \"bootCommand\": \"MODELS=piper\\nPORT=8099\\n# talkito refuses to fetch model weights when it cannot ask a human, and a boot\\n# command has no TTY. Without this every phrase fails with \\\"Piper model\\n# unavailable\\\" while the run still exits 0 and publishes an empty table.\\nexport TALKITO_AUTO_APPROVE_DOWNLOADS=1\\nROOT=/workspace/talkito\\nBENCH=$ROOT/benchmark\\nSERVE=/workspace/tts-report\\necho \\\"== Orcabot: talkito TTS benchmark ($MODELS) ==\\\"\\n\\n# Serve a placeholder first, so the browser block has something at a stable URL from\\n# the first second rather than an error until the run finishes.\\nmkdir -p \\\"$SERVE\\\"\\n[ -f \\\"$SERVE/index.html\\\" ] || printf '%s' 'TTS benchmark

Benchmark running...

This page reloads every 10 seconds. The report replaces it when the run finishes.

' > \\\"$SERVE/index.html\\\"\\ncurl -sf -o /dev/null --max-time 2 \\\"http://127.0.0.1:$PORT/\\\" || ( cd \\\"$SERVE\\\" && nohup python3 -m http.server \\\"$PORT\\\" >/workspace/.tts-serve.log 2>&1 & )\\n\\n# talkito's .gitmodules uses git@github.com: URLs and the sandbox has no SSH key,\\n# so --recurse-submodules fails; both repos are cloned over HTTPS instead.\\n# The harness now ships inside talkito-benchmark, so tests-repo is not needed --\\n# but tools/paths.py still wants a talkito CHECKOUT (it looks for talkito/tts.py,\\n# not the installed package), and benchmark/ inside talkito/ is the layout it\\n# resolves via \\\"../..\\\".\\n[ -d \\\"$ROOT/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito.git \\\"$ROOT\\\"\\n[ -d \\\"$BENCH/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito-benchmark.git \\\"$BENCH\\\"\\n\\n# make_samples.py shells out to ffmpeg to turn the listening clip into an mp3, and\\n# the sandbox image ships without it. Missing, the sweep still scores normally and\\n# only the play button is lost, so this is a soft dependency -- but a table of\\n# numbers with nothing to hear is most of the point of the page.\\ncommand -v ffmpeg >/dev/null 2>&1 || { apt-get update -qq && apt-get install -y -qq --no-install-recommends ffmpeg; }\\n\\n# piper needs no torch at all, and faster-whisper scores on CPU through CTranslate2,\\n# so the whole piper arm installs in seconds. Heavier models would need much more.\\npython3 -c 'import piper' 2>/dev/null || pip3 install --break-system-packages -q piper-tts\\npython3 -c 'import faster_whisper' 2>/dev/null || pip3 install --break-system-packages -q faster-whisper\\n\\ncd \\\"$BENCH\\\" || exec bash\\n# The listening clip has to exist before the page is built, or the table is numbers\\n# with nothing to play. benchmark.py builds the page at the end, so samples come first.\\npython3 tools/make_samples.py $(echo \\\"$MODELS\\\" | tr , ' ')\\n# build_benchmark_page.py defaults TTS_SAMPLE_DIR to a hardcoded\\n# /Users/robertmacrae/... path, so off that one machine it finds no clips and\\n# builds a table with nothing to play. make_samples.py resolves the same variable\\n# with expanduser and writes its mp3s to a \\\"mp3\\\" subdirectory, so the two only\\n# agree when the variable is set explicitly -- and they want different values,\\n# which is why this is set here rather than once at the top.\\n# --only skips the interactive picker, which needs a TTY it does not have here.\\nTTS_SAMPLE_DIR=$HOME/.cache/talkito/benchmark-samples/mp3 python3 tools/benchmark.py --only \\\"$MODELS\\\"\\nrc=$?\\nif [ -f tts-benchmark.html ]; then\\n cp -f tts-benchmark.html \\\"$SERVE/index.html\\\"\\n echo \\\"report published -> http://localhost:$PORT/\\\"\\nelse\\n echo \\\"no tts-benchmark.html was produced; measurements (if any) are under $BENCH/tools/\\\"\\nfi\\necho \\\"[tts benchmark done rc=$rc]\\\"\\nexec bash\", \"subagentIds\": [], \"skillIds\": [], \"mcpToolIds\": []}", "position": { "x": 0, "y": 0 @@ -37,5 +37,5 @@ "y": 0, "zoom": 1 }, - "setupGuide": "# talkito TTS benchmark\n\nThis template runs the open-weight TTS benchmark from\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) end to end inside\none terminal, and shows the report it produces in the browser block beside it.\n\nUnlike the other benchmark templates there is **no orchestrator**. Nothing here needs\nOrcabot chat, an agent, or an API key: the terminal's boot command is the whole run,\nand the browser block is pointed at a local web server the same command starts. Start\nthe dashboard and it goes.\n\n## What the terminal does\n\n1. Serves a placeholder page on `localhost:8099`, so the browser block has a stable\n URL immediately instead of an error until the run finishes.\n2. Clones `talkito`, `talkito-benchmark` and `talkito-tests` over **HTTPS**. talkito's\n `.gitmodules` uses `git@github.com:` URLs, which cannot work in the sandbox (no SSH\n key), so `--recurse-submodules` is deliberately not used. Cloning `benchmark/` into\n `talkito/` reproduces the layout `tools/paths.py` expects.\n3. Installs `piper-tts` and `faster-whisper`. Piper uses no torch at all and\n faster-whisper scores on CPU via CTranslate2, so this arm is small and quick.\n4. Synthesizes the one listening clip per model (`make_samples.py`) **before** the\n sweep, because `benchmark.py` builds the page at the end and a sample produced\n afterwards would not be in it.\n5. Runs `python3 tools/benchmark.py --only piper`, which sweeps the 84-phrase corpus,\n transcribes each clip back with Whisper, scores word error rate, then rebuilds the\n page and the export zip.\n6. Copies `tts-benchmark.html` over the placeholder, so the browser block shows the\n report on its next refresh.\n\n## Running more models\n\n`MODELS` on the first line of the boot command is the only thing to change. It is\npassed to `--only`, so it is comma-separated:\n\n MODELS=piper # the default: minutes\n MODELS=piper,kokoro # kokoro pulls torch, so expect a long first run\n MODELS=piper,kittentts\n\nThere are 73 providers. A **full sweep is around 11 hours** and much longer on a\nfirst run, because every engine downloads its own weights -- which is why this\ntemplate ships pinned to piper rather than to everything.\n\n## Reading the result\n\nThe report is the same page the published benchmark is built from: one row per\nconfiguration, with word error rate, real-time factor, compute per phrase, PESQ and a\nplay button per row. `tts-benchmark-export.zip` beside it is the shareable archive\n(page + measurements + audio).\n\nMeasurements are saved per unit as they complete, so an interrupted run keeps its\nfinished work and re-running resumes rather than restarting. `--keep` adds to existing\nmeasurements instead of replacing them.\n\n## If the run stops early\n\nThe terminal drops to a shell either way and prints `[tts benchmark done rc=N]`.\n`rc` is `benchmark.py`'s exit code -- non-zero means at least one provider failed, and\nthe per-provider output above it says which. The page and the zip can always be\nrebuilt from stored measurements alone:\n\n cd /workspace/talkito/benchmark\n python3 tools/build_benchmark_page.py tts-benchmark.html\n" + "setupGuide": "# talkito TTS benchmark\n\nThis template runs the open-weight TTS benchmark from\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) end to end inside\none terminal, and shows the report it produces in the browser block beside it.\n\nUnlike the other benchmark templates there is **no orchestrator**. Nothing here needs\nOrcabot chat, an agent, or an API key: the terminal's boot command is the whole run,\nand the browser block is pointed at a local web server the same command starts. Start\nthe dashboard and it goes.\n\n## What the terminal does\n\n1. Serves a placeholder page on `localhost:8099`, so the browser block has a stable\n URL immediately instead of an error until the run finishes.\n2. Clones `talkito` and `talkito-benchmark` over **HTTPS**. talkito's `.gitmodules`\n uses `git@github.com:` URLs, which cannot work in the sandbox (no SSH key), so\n `--recurse-submodules` is deliberately not used. The harness ships inside\n talkito-benchmark now, so `tests-repo` is not needed -- but `tools/paths.py` still\n wants a talkito *checkout* rather than the installed package (it looks for\n `talkito/tts.py`), and `benchmark/` inside `talkito/` is the layout it resolves.\n3. Installs `piper-tts` and `faster-whisper`. Piper uses no torch at all and\n faster-whisper scores on CPU via CTranslate2, so this arm is small and quick.\n4. Synthesizes the one listening clip per model (`make_samples.py`) **before** the\n sweep, because `benchmark.py` builds the page at the end and a sample produced\n afterwards would not be in it.\n5. Runs `python3 tools/benchmark.py --only piper`, which sweeps the 84-phrase corpus,\n transcribes each clip back with Whisper, scores word error rate, then rebuilds the\n page and the export zip.\n6. Copies `tts-benchmark.html` over the placeholder, so the browser block shows the\n report on its next refresh.\n\n## Running more models\n\n`MODELS` on the first line of the boot command is the only thing to change. It is\npassed to `--only`, so it is comma-separated:\n\n MODELS=piper # the default: minutes\n MODELS=piper,kokoro # kokoro pulls torch, so expect a long first run\n MODELS=piper,kittentts\n\nThere are 73 providers. A **full sweep is around 11 hours** and much longer on a\nfirst run, because every engine downloads its own weights -- which is why this\ntemplate ships pinned to piper rather than to everything.\n\n## Reading the result\n\nThe report is the same page the published benchmark is built from: one row per\nconfiguration, with word error rate, real-time factor, compute per phrase, PESQ and a\nplay button per row. `tts-benchmark-export.zip` beside it is the shareable archive\n(page + measurements + audio).\n\nMeasurements are saved per unit as they complete, so an interrupted run keeps its\nfinished work and re-running resumes rather than restarting. `--keep` adds to existing\nmeasurements instead of replacing them.\n\n## If the run stops early\n\nThe terminal drops to a shell either way and prints `[tts benchmark done rc=N]`.\n`rc` is `benchmark.py`'s exit code -- non-zero means at least one provider failed, and\nthe per-provider output above it says which. The page and the zip can always be\nrebuilt from stored measurements alone:\n\n cd /workspace/talkito/benchmark\n python3 tools/build_benchmark_page.py tts-benchmark.html\n" } From d6dbf809840568fa22e7fc73a99fca0f02357156 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Fri, 21 Aug 2026 22:04:16 +0100 Subject: [PATCH 06/34] Drop the sample-directory workaround, now that upstream resolves it build_benchmark_page.py and build_export.py no longer default TTS_SAMPLE_DIR to an absolute home directory, so the template no longer has to force one. It now sets nothing, which is what the three scripts want: they need different directories -- make_samples.py writes mp3s into an "mp3" subdirectory the page builder reads and the export builder does not -- and each defaults correctly on its own. Forcing a single value was what made the export report "0 audio samples". The workaround was right for the code it was written against and wrong the moment that code was fixed, which is the usual life of a workaround. Re-verified from an empty workspace: piper 82/84 phrases WER 5.3% RTF 0.08x 1 configurations, 1 audio samples The page reaches the browser block at 50 KB with the clip embedded, and the export zip now carries samples/piper.mp3 rather than nothing. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- benchmarks/talkito-tts/README.md | 33 ++++++++++++---------------- benchmarks/talkito-tts/template.json | 2 +- 2 files changed, 15 insertions(+), 20 deletions(-) diff --git a/benchmarks/talkito-tts/README.md b/benchmarks/talkito-tts/README.md index af5341ac..dffd9925 100644 --- a/benchmarks/talkito-tts/README.md +++ b/benchmarks/talkito-tts/README.md @@ -34,7 +34,7 @@ report is built it is copied over the placeholder and the next refresh picks it ## Four things the sandbox forces -All four were found by running it, not by reading it. Three of them fail *quietly* -- +All four were found by running it, not by reading it. Two of them fail *quietly* -- the run still exits 0 and still publishes a page. **HTTPS clones, not submodules.** talkito's `.gitmodules` points at @@ -45,7 +45,7 @@ repos over HTTPS instead and arranges them as /workspace/talkito/ talkito /workspace/talkito/benchmark/ talkito-benchmark -The harness now ships inside talkito-benchmark, so `tests-repo` is no longer needed. +The harness ships inside talkito-benchmark, so `tests-repo` is not needed. `tools/paths.py` still wants a talkito **checkout** rather than the installed package -- it looks for `talkito/tts.py` -- and `benchmark/` inside `talkito/` is the layout it resolves via `../..`. @@ -58,16 +58,16 @@ boot command. cannot ask a human, and a boot command has no TTY. Without it every phrase fails with `Piper model unavailable` -- and the sweep still reports `rc=0`, having published a table reading `0/84 phrases, WER 0.0%, RTF 0.00x`. A green run measuring nothing is -the worst failure here, so this variable is not optional. +the worst failure available here, so this variable is not optional. -**`TTS_SAMPLE_DIR`, set per step.** `build_benchmark_page.py` and `build_export.py` -default it to a hardcoded `/Users/robertmacrae/.cache/...` path, so anywhere else they -find no clips and build a table with nothing to play. `make_samples.py` resolves the -same variable with `expanduser` and writes its mp3s into an `mp3` subdirectory, so the -two only agree when it is set explicitly -- and they want *different* values, which is -why the boot command sets it on the `benchmark.py` line rather than once at the top. -ffmpeg is installed for the same reason: `make_samples.py` shells out to it for the -wav-to-mp3 step and the sandbox image ships without it. +**ffmpeg.** `make_samples.py` shells out to it for the wav-to-mp3 step and the sandbox +image ships without it. Missing, every number is still correct and only the play +button is lost -- which is most of the point of the page. + +**`TTS_SAMPLE_DIR` is deliberately not set.** The three scripts want different +directories: `make_samples.py` writes mp3s into an `mp3` subdirectory that the page +builder reads and the export builder does not. Each now defaults correctly on its own, +so setting one value for all of them is a way to break the export, not fix it. ## Running more than piper @@ -92,18 +92,13 @@ easy to make by accident. dependency install, weight download, 84-phrase sweep, Whisper scoring, page build and serving. The run takes about three and a half minutes for piper and finishes - piper 82/84 phrases WER 5.0% RTF 0.07x + piper 82/84 phrases WER 5.3% RTF 0.08x against 4% and 0.04x for the published Apple M2 figures -- close, and real-time factor is wall-clock so a throttled container is expected to read higher. -The report reaches the browser block at 51 KB with the listening clip embedded. - -One cosmetic gap remains: `build_export.py` reports `0 audio samples` because the -single `TTS_SAMPLE_DIR` it shares with the page builder points at the mp3 -subdirectory. That affects only the counter and the zip's `samples/` folder, not the -page, which is what the browser block shows. The real fix is upstream -- neither -builder should carry a hardcoded home directory. +The report reaches the browser block at 50 KB with the listening clip embedded, and +the export zip carries `samples/piper.mp3` alongside it. ## Seeding diff --git a/benchmarks/talkito-tts/template.json b/benchmarks/talkito-tts/template.json index 2982799c..143ac3be 100644 --- a/benchmarks/talkito-tts/template.json +++ b/benchmarks/talkito-tts/template.json @@ -7,7 +7,7 @@ { "placeholderId": "runner", "type": "terminal", - "content": "{\"name\": \"TTS benchmark\", \"agentic\": false, \"bootCommand\": \"MODELS=piper\\nPORT=8099\\n# talkito refuses to fetch model weights when it cannot ask a human, and a boot\\n# command has no TTY. Without this every phrase fails with \\\"Piper model\\n# unavailable\\\" while the run still exits 0 and publishes an empty table.\\nexport TALKITO_AUTO_APPROVE_DOWNLOADS=1\\nROOT=/workspace/talkito\\nBENCH=$ROOT/benchmark\\nSERVE=/workspace/tts-report\\necho \\\"== Orcabot: talkito TTS benchmark ($MODELS) ==\\\"\\n\\n# Serve a placeholder first, so the browser block has something at a stable URL from\\n# the first second rather than an error until the run finishes.\\nmkdir -p \\\"$SERVE\\\"\\n[ -f \\\"$SERVE/index.html\\\" ] || printf '%s' 'TTS benchmark

Benchmark running...

This page reloads every 10 seconds. The report replaces it when the run finishes.

' > \\\"$SERVE/index.html\\\"\\ncurl -sf -o /dev/null --max-time 2 \\\"http://127.0.0.1:$PORT/\\\" || ( cd \\\"$SERVE\\\" && nohup python3 -m http.server \\\"$PORT\\\" >/workspace/.tts-serve.log 2>&1 & )\\n\\n# talkito's .gitmodules uses git@github.com: URLs and the sandbox has no SSH key,\\n# so --recurse-submodules fails; both repos are cloned over HTTPS instead.\\n# The harness now ships inside talkito-benchmark, so tests-repo is not needed --\\n# but tools/paths.py still wants a talkito CHECKOUT (it looks for talkito/tts.py,\\n# not the installed package), and benchmark/ inside talkito/ is the layout it\\n# resolves via \\\"../..\\\".\\n[ -d \\\"$ROOT/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito.git \\\"$ROOT\\\"\\n[ -d \\\"$BENCH/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito-benchmark.git \\\"$BENCH\\\"\\n\\n# make_samples.py shells out to ffmpeg to turn the listening clip into an mp3, and\\n# the sandbox image ships without it. Missing, the sweep still scores normally and\\n# only the play button is lost, so this is a soft dependency -- but a table of\\n# numbers with nothing to hear is most of the point of the page.\\ncommand -v ffmpeg >/dev/null 2>&1 || { apt-get update -qq && apt-get install -y -qq --no-install-recommends ffmpeg; }\\n\\n# piper needs no torch at all, and faster-whisper scores on CPU through CTranslate2,\\n# so the whole piper arm installs in seconds. Heavier models would need much more.\\npython3 -c 'import piper' 2>/dev/null || pip3 install --break-system-packages -q piper-tts\\npython3 -c 'import faster_whisper' 2>/dev/null || pip3 install --break-system-packages -q faster-whisper\\n\\ncd \\\"$BENCH\\\" || exec bash\\n# The listening clip has to exist before the page is built, or the table is numbers\\n# with nothing to play. benchmark.py builds the page at the end, so samples come first.\\npython3 tools/make_samples.py $(echo \\\"$MODELS\\\" | tr , ' ')\\n# build_benchmark_page.py defaults TTS_SAMPLE_DIR to a hardcoded\\n# /Users/robertmacrae/... path, so off that one machine it finds no clips and\\n# builds a table with nothing to play. make_samples.py resolves the same variable\\n# with expanduser and writes its mp3s to a \\\"mp3\\\" subdirectory, so the two only\\n# agree when the variable is set explicitly -- and they want different values,\\n# which is why this is set here rather than once at the top.\\n# --only skips the interactive picker, which needs a TTY it does not have here.\\nTTS_SAMPLE_DIR=$HOME/.cache/talkito/benchmark-samples/mp3 python3 tools/benchmark.py --only \\\"$MODELS\\\"\\nrc=$?\\nif [ -f tts-benchmark.html ]; then\\n cp -f tts-benchmark.html \\\"$SERVE/index.html\\\"\\n echo \\\"report published -> http://localhost:$PORT/\\\"\\nelse\\n echo \\\"no tts-benchmark.html was produced; measurements (if any) are under $BENCH/tools/\\\"\\nfi\\necho \\\"[tts benchmark done rc=$rc]\\\"\\nexec bash\", \"subagentIds\": [], \"skillIds\": [], \"mcpToolIds\": []}", + "content": "{\"name\": \"TTS benchmark\", \"agentic\": false, \"bootCommand\": \"MODELS=piper\\nPORT=8099\\n# talkito refuses to fetch model weights when it cannot ask a human, and a boot\\n# command has no TTY. Without this every phrase fails with \\\"Piper model\\n# unavailable\\\" while the run still exits 0 and publishes an empty table.\\nexport TALKITO_AUTO_APPROVE_DOWNLOADS=1\\nROOT=/workspace/talkito\\nBENCH=$ROOT/benchmark\\nSERVE=/workspace/tts-report\\necho \\\"== Orcabot: talkito TTS benchmark ($MODELS) ==\\\"\\n\\n# Serve a placeholder first, so the browser block has something at a stable URL from\\n# the first second rather than an error until the run finishes.\\nmkdir -p \\\"$SERVE\\\"\\n[ -f \\\"$SERVE/index.html\\\" ] || printf '%s' 'TTS benchmark

Benchmark running...

This page reloads every 10 seconds. The report replaces it when the run finishes.

' > \\\"$SERVE/index.html\\\"\\ncurl -sf -o /dev/null --max-time 2 \\\"http://127.0.0.1:$PORT/\\\" || ( cd \\\"$SERVE\\\" && nohup python3 -m http.server \\\"$PORT\\\" >/workspace/.tts-serve.log 2>&1 & )\\n\\n# talkito's .gitmodules uses git@github.com: URLs and the sandbox has no SSH key,\\n# so --recurse-submodules fails; both repos are cloned over HTTPS instead.\\n# The harness now ships inside talkito-benchmark, so tests-repo is not needed --\\n# but tools/paths.py still wants a talkito CHECKOUT (it looks for talkito/tts.py,\\n# not the installed package), and benchmark/ inside talkito/ is the layout it\\n# resolves via \\\"../..\\\".\\n[ -d \\\"$ROOT/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito.git \\\"$ROOT\\\"\\n[ -d \\\"$BENCH/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito-benchmark.git \\\"$BENCH\\\"\\n\\n# make_samples.py shells out to ffmpeg to turn the listening clip into an mp3, and\\n# the sandbox image ships without it. Missing, the sweep still scores normally and\\n# only the play button is lost, so this is a soft dependency -- but a table of\\n# numbers with nothing to hear is most of the point of the page.\\ncommand -v ffmpeg >/dev/null 2>&1 || { apt-get update -qq && apt-get install -y -qq --no-install-recommends ffmpeg; }\\n\\n# piper needs no torch at all, and faster-whisper scores on CPU through CTranslate2,\\n# so the whole piper arm installs in seconds. Heavier models would need much more.\\npython3 -c 'import piper' 2>/dev/null || pip3 install --break-system-packages -q piper-tts\\npython3 -c 'import faster_whisper' 2>/dev/null || pip3 install --break-system-packages -q faster-whisper\\n\\ncd \\\"$BENCH\\\" || exec bash\\n# The listening clip has to exist before the page is built, or the table is numbers\\n# with nothing to play. benchmark.py builds the page at the end, so samples come first.\\npython3 tools/make_samples.py $(echo \\\"$MODELS\\\" | tr , ' ')\\n# TTS_SAMPLE_DIR is deliberately NOT set. The three scripts want different\\n# directories -- make_samples.py writes mp3s into a \\\"mp3\\\" subdirectory that the\\n# page builder reads and the export builder does not -- and each now defaults\\n# correctly on its own. Setting one value for all of them is what made the export\\n# report \\\"0 audio samples\\\".\\n# --only skips the interactive picker, which needs a TTY it does not have here.\\npython3 tools/benchmark.py --only \\\"$MODELS\\\"\\nrc=$?\\nif [ -f tts-benchmark.html ]; then\\n cp -f tts-benchmark.html \\\"$SERVE/index.html\\\"\\n echo \\\"report published -> http://localhost:$PORT/\\\"\\nelse\\n echo \\\"no tts-benchmark.html was produced; measurements (if any) are under $BENCH/tools/\\\"\\nfi\\necho \\\"[tts benchmark done rc=$rc]\\\"\\nexec bash\", \"subagentIds\": [], \"skillIds\": [], \"mcpToolIds\": []}", "position": { "x": 0, "y": 0 From e1b86c7afb64f22ddf3beb67722f5a7f1b8e28fd Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Fri, 21 Aug 2026 22:13:41 +0100 Subject: [PATCH 07/34] Bias ballot selection toward the clips ranked least Selection was uniform over whatever the voter had not heard, and nothing looked at how often a configuration had been ranked overall. Across 43 real ballots that left appearances spread 2 to 9 around a mean of 4.7, standard deviation 2.0 -- almost exactly the sqrt(4.7) a Poisson spread predicts, which is the signature of no balancing at all. Six configurations sat under the display cutoff and showed no human score, while others were measured twice over. Selection is now weighted by 1/(count+1) via Efraimidis-Spirakis, which samples without replacement in one pass -- necessary here because a ballot must never offer the same clip twice. The anchor draw follows the same rule. Deliberately a bias and not an ordering. Always serving the k rarest as a set would leave them only ever ranked against each other, and Bradley-Terry cannot place a disconnected clique on the same scale as the rest of the table. The simulation says that worry is not just avoided but inverted -- at today's volume the weighted selector reaches a connected comparison graph every time where uniform manages it three runs in four, and it produces slightly more distinct pairs, because spreading coverage beats concentrating it: ballots selector min max stdev under-3 connected 43 uniform 0.9 10.5 2.11 5.1 75% 43 weighted 2.2 9.8 1.56 1.4 100% 100 uniform 5.0 23.9 3.80 0.1 100% 100 weighted 6.9 21.3 2.82 0.0 100% The uniform arm reproduces the live numbers closely, which is the reason to believe the weighted column. 1/(count+1) self-attenuates: it biases by over 3x while 2 against 9 decides whether a row is visible, and fades to near-uniform once everything sits in the twenties and the deficits stop mattering. Counts come from an aggregate over counted ballots rather than by reading every row, since this runs on each issue rather than only when the table renders. Abandoned ballots are excluded -- they told us nothing about the clips they offered, and counting them as coverage would starve exactly those clips. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- controlplane/src/tts/handler.test.ts | 100 +++++++++++++++++++++++++++ controlplane/src/tts/handler.ts | 68 ++++++++++++++++-- 2 files changed, 163 insertions(+), 5 deletions(-) create mode 100644 controlplane/src/tts/handler.test.ts diff --git a/controlplane/src/tts/handler.test.ts b/controlplane/src/tts/handler.test.ts new file mode 100644 index 00000000..d3516525 --- /dev/null +++ b/controlplane/src/tts/handler.test.ts @@ -0,0 +1,100 @@ +// Copyright 2026 Rob Macrae. All rights reserved. +// SPDX-License-Identifier: LicenseRef-Proprietary + +// Ballot selection is weighted toward the configurations ranked least, because +// uniform sampling left the counts Poisson-spread: at 43 real ballots that was 2 +// to 9 appearances around a mean of 4.7, with six configurations stranded below +// the display cutoff while others were measured twice over. +// +// The properties worth pinning are the ones a plausible rewrite would break. +// Duplicates are the sharpest: a ballot showing the same clip twice asks the +// reader to rank it against itself. The bias is the point of the function. And +// the bias must stay a bias rather than becoming an ordering, because always +// serving the k rarest as a set would leave them only ever ranked against each +// other, and Bradley-Terry cannot place a disconnected clique on the same scale +// as the rest of the table. + +import { describe, expect, it } from 'vitest'; +import { pickLeastRanked } from './handler'; + +const counts = (o: Record) => new Map(Object.entries(o)); + +describe('pickLeastRanked', () => { + it('returns exactly k items', () => { + const pool = ['a', 'b', 'c', 'd', 'e']; + for (const k of [1, 2, 3, 4]) { + expect(pickLeastRanked(pool, k, counts({})).length).toBe(k); + } + }); + + it('never repeats a configuration within one ballot', () => { + // Sampling without replacement, not k independent draws. With a heavy skew + // an implementation that drew independently would return the rare item + // several times, and the ballot would ask for it to be ranked against itself. + const pool = ['rare', 'common1', 'common2', 'common3']; + const c = counts({ rare: 0, common1: 40, common2: 40, common3: 40 }); + for (let i = 0; i < 400; i++) { + const got = pickLeastRanked(pool, 3, c); + expect(new Set(got).size).toBe(got.length); + } + }); + + it('returns the whole pool when k covers it, without dropping any', () => { + const pool = ['a', 'b', 'c']; + const got = pickLeastRanked(pool, 3, counts({ a: 9, b: 0, c: 4 })); + expect([...got].sort()).toEqual(['a', 'b', 'c']); + expect(pickLeastRanked(pool, 9, counts({})).length).toBe(3); + }); + + it('picks the least-ranked far more often than the most-ranked', () => { + const pool = ['rare', 'mid', 'common']; + const c = counts({ rare: 0, mid: 4, common: 12 }); + const seen: Record = { rare: 0, mid: 0, common: 0 }; + for (let i = 0; i < 3000; i++) { + for (const p of pickLeastRanked(pool, 1, c)) seen[p]++; + } + expect(seen.rare).toBeGreaterThan(seen.mid); + expect(seen.mid).toBeGreaterThan(seen.common); + // Comfortably wide of the 3000/3 a uniform pick would give each. + expect(seen.rare).toBeGreaterThan(1500); + }); + + it('still reaches a well-ranked configuration sometimes', () => { + // A strict least-first ordering would return `rare` every time and starve + // the rest, which disconnects the comparison graph. The weighting has to + // leave every pairing reachable. + const pool = ['rare', 'common']; + const c = counts({ rare: 0, common: 8 }); + let common = 0; + for (let i = 0; i < 3000; i++) { + if (pickLeastRanked(pool, 1, c)[0] === 'common') common++; + } + expect(common).toBeGreaterThan(0); + }); + + it('treats an unseen configuration as the rarest', () => { + // Absent from the map means never ranked, which is a zero rather than a gap. + // Defaulting it the other way would starve exactly the new arrivals a fresh + // export adds. + const pool = ['brand-new', 'established']; + const c = counts({ established: 6 }); // brand-new absent entirely + let fresh = 0; + for (let i = 0; i < 2000; i++) { + if (pickLeastRanked(pool, 1, c)[0] === 'brand-new') fresh++; + } + expect(fresh).toBeGreaterThan(1400); + }); + + it('is near-uniform once counts are high and close together', () => { + // 1/(count+1) self-attenuates: the bias matters when 2 against 9 decides + // visibility, and should fade when everything is well measured. + const pool = ['a', 'b']; + const c = counts({ a: 60, b: 64 }); + let a = 0; + for (let i = 0; i < 4000; i++) { + if (pickLeastRanked(pool, 1, c)[0] === 'a') a++; + } + expect(a).toBeGreaterThan(1900); + expect(a).toBeLessThan(2600); + }); +}); diff --git a/controlplane/src/tts/handler.ts b/controlplane/src/tts/handler.ts index 6a813438..abf483e3 100644 --- a/controlplane/src/tts/handler.ts +++ b/controlplane/src/tts/handler.ts @@ -1,10 +1,10 @@ // Copyright 2026 Rob Macrae. All rights reserved. // SPDX-License-Identifier: LicenseRef-Proprietary -// REVISION: tts-ballots-v2-full-config-set +// REVISION: tts-ballots-v3-balanced-selection import type { Env } from '../types'; -const MODULE_REVISION = 'tts-ballots-v2-full-config-set'; +const MODULE_REVISION = 'tts-ballots-v3-balanced-selection'; console.log(`[tts] REVISION: ${MODULE_REVISION} loaded at ${new Date().toISOString()}`); /** The run these ballots score. Bumped when a new TTS run replaces the samples. */ @@ -79,6 +79,57 @@ function shuffle(a: T[]): T[] { return out; } +/** How often each configuration has already been ranked, across counted ballots. + * + * Aggregated in SQL rather than by reading every ballot into the worker, because + * unlike the score computation this runs on every ballot issued, not only when + * the table is rendered. + * + * Counted ballots only, matching what the scores are built from: an issued and + * abandoned ballot has told us nothing about the clips it offered, so treating + * it as coverage would starve exactly the configurations that were offered and + * never ranked. */ +async function appearanceCounts(env: Env): Promise> { + const { results } = await env.DB.prepare( + `SELECT je.value AS config, COUNT(*) AS n + FROM tts_ballots, json_each(tts_ballots.ranking_json) je + WHERE run = ? AND status = 'counted' + GROUP BY je.value` + ).bind(RUN).all<{ config: string; n: number }>(); + return new Map((results ?? []).map((r) => [String(r.config), Number(r.n)])); +} + +/** Weighted sample without replacement, favouring the configurations ranked least. + * + * Efraimidis-Spirakis: give each item the key u^(1/w) for u uniform on (0,1) and + * take the k largest. One pass, no rejection loop, and a correct weighted sample + * without replacement - which matters because a ballot must never offer the same + * clip twice. + * + * The weight is 1/(count+1), not a strict least-ranked-first ordering, and the + * randomness is the point rather than a compromise. Always serving the k rarest + * together would make the same under-sampled configurations appear as a set every + * time, and Bradley-Terry needs the comparison graph connected: a clique that is + * only ever ranked against itself cannot be placed on the same scale as the rest, + * however many ballots it collects. Weighting pulls the counts level while every + * pairing stays reachable. + * + * 1/(count+1) also self-attenuates. Early on, when 2 against 9 is the difference + * between invisible and well-measured, it biases by more than 3x; once everything + * sits in the twenties the ratio is near 1 and selection is effectively uniform + * again, which is what you want when the deficits no longer matter. */ +export function pickLeastRanked(pool: string[], k: number, counts: Map): string[] { + if (k >= pool.length) return shuffle(pool); + return pool + .map((config) => ({ + config, + key: Math.random() ** ((counts.get(config) ?? 0) + 1), + })) + .sort((a, b) => b.key - a.key) + .slice(0, k) + .map((x) => x.config); +} + /** Coarse voter fingerprint for rate limiting and dedup. Hashed, never stored raw. */ async function voterHash(request: Request): Promise { const ip = request.headers.get('cf-connecting-ip') ?? request.headers.get('x-forwarded-for') ?? ''; @@ -123,17 +174,24 @@ export async function issueBallot(env: Env, request: Request): Promise return Response.json({ exhausted: 'clips', submitted, max: MAX_BALLOTS_PER_VOTER }); } + // Selection is biased toward whatever has been ranked least. Uniform sampling + // leaves the counts Poisson-spread - at 43 ballots that was 2 to 9 appearances + // against a mean of 4.7, with six configurations stranded under the cutoff and + // therefore invisible, while others were measured twice over. + const counts = await appearanceCounts(env); + // The anchor has to be one they have not heard either, or the attention check - // is just a memory test. + // is just a memory test. Which of the two, though, follows the same rule as + // everything else. const freshAnchors = ANCHORS.filter((a) => fresh.includes(a)); const anchor = freshAnchors.length && Math.random() < ANCHOR_RATE - ? freshAnchors[Math.floor(Math.random() * freshAnchors.length)] + ? pickLeastRanked(freshAnchors, 1, counts)[0] : null; const pool = fresh.filter((c) => c !== anchor); const want = Math.min(ITEMS_PER_BALLOT, fresh.length) - (anchor ? 1 : 0); - const picked = shuffle(pool).slice(0, want); + const picked = pickLeastRanked(pool, want, counts); const items = shuffle(anchor ? [...picked, anchor] : picked); const id = crypto.randomUUID(); From 00b337ac2106452dc438a56220e368ec3052f871 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Fri, 21 Aug 2026 22:20:02 +0100 Subject: [PATCH 08/34] Colour word error on what the cell shows, and stop green at 6% Three rows read "8%" and only one of them was green. The export tones word error on the unrounded rate while the column prints whole percents, so its cutoff at 0.08 fell inside a displayed band: Orpheus Q8 at 7.70% was under it, XTTS at 8.03% and Orpheus Q4 at 8.23% were over. Identical text in different colours reads as a rendering fault rather than a distinction, and no reader can recover the third decimal that would justify it. The threshold now applies to the number actually on the page, so two cells that read the same are guaranteed to be coloured the same whatever the export's own rounding does next. Green is six percent and below, which moves the boundary off the crowded 8% band and onto one where the displayed value changes anyway. Kokoro at 7% and Orpheus Q8 at 8% lose their green; nothing else moves. warn and bad are left where the export put them, marking the 24%-and-worse rows: those are not in question, and re-deriving them here would mean a second threshold to keep in step with the first. Verified from the rendered page rather than the data alone -- reading computed styles back out of the table, every distinct displayed value now maps to exactly one colour. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- frontend/scripts/import-tts-export.mjs | 37 +++++++++++++++++-- .../benchmarks/open-weight-tts/2026-08.json | 4 +- 2 files changed, 36 insertions(+), 5 deletions(-) diff --git a/frontend/scripts/import-tts-export.mjs b/frontend/scripts/import-tts-export.mjs index d7f9e696..041e58b0 100644 --- a/frontend/scripts/import-tts-export.mjs +++ b/frontend/scripts/import-tts-export.mjs @@ -45,6 +45,36 @@ const OUT_AUDIO = "public/benchmarks/tts"; * the methodology instead. */ const DROP_COLUMNS = new Set(["Frame rate", "Passed", "Libs", "Lead-in", "WER med", "Avg audio"]); +/** Columns the page re-colours itself, keyed by the export's own header name, and + * the displayed value at or below which a cell is green. + * + * The export tones word error on the unrounded rate while the column prints whole + * percents, so its threshold lands inside a displayed band: three rows all read + * "8%" and only the first was green, because 7.70 was under the cutoff and 8.03 + * and 8.23 were over it. Two cells showing the same number in different colours + * reads as a rendering fault, not as a distinction, and no reader can recover the + * third decimal that justifies it. + * + * So the threshold is applied to what is actually on the page. Identical text is + * then guaranteed identical colour, whatever the export's own rounding does next. + * + * warn and bad are left exactly where the export put them. They mark the 24%-and- + * worse rows, which are not in question, and re-deriving them here would mean + * owning a second threshold to keep in step with the first. */ +const GREEN_AT_OR_BELOW = { "WER base": 6 }; + +/** The tone a cell should carry: the export's, unless this column is re-coloured. */ +function retone(header, cell) { + const limit = GREEN_AT_OR_BELOW[header]; + if (limit === undefined) return cell.tone; + if (cell.tone === "warn" || cell.tone === "bad") return cell.tone; + const shown = Number(String(cell.v).replace("%", "").trim()); + // A cell that is not a number at all - "·" for unmeasured - keeps whatever it + // had rather than being silently promoted to green by a NaN comparison. + if (!Number.isFinite(shown)) return cell.tone; + return shown <= limit ? "good" : ""; +} + /** Configurations the comparison does not show. The NeuTTS-2E CPU builds are * the same weights as the rows that remain, run on a slower path: they tripled * the size of the NeuTTS band while saying nothing about the model, only about @@ -417,11 +447,12 @@ const rows = rawRows.map(([, rowAttrs, inner]) => { const withOverride = override ? { ...cell, v: override.v, sort: override.sort ?? override.v } : cell; + // After the override, so a corrected value is coloured on what it now says. + const toned = { ...withOverride, tone: retone(headers[i], withOverride) }; if (headers[i] === "Licence" && LICENCE_PROOF_URLS[config]) { - return { ...withOverride, href: LICENCE_PROOF_URLS[config] }; + return { ...toned, href: LICENCE_PROOF_URLS[config] }; } - if (override) return withOverride; - return cell; + return toned; }) .filter((_, i) => keep[i]); cells[0] = { ...cells[0], v: display, ...(MODEL_URLS[config] ? { href: MODEL_URLS[config] } : {}) }; diff --git a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json index 64b6be52..07333e56 100644 --- a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json +++ b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json @@ -1301,7 +1301,7 @@ { "v": "7%", "sort": "0.0684", - "tone": "good", + "tone": "", "align": "", "restricted": false }, @@ -1402,7 +1402,7 @@ { "v": "8%", "sort": "0.0770", - "tone": "good", + "tone": "", "align": "", "restricted": false }, From 087dce739b1be490c14f53639cc183c66f9fc9b1 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Fri, 21 Aug 2026 22:46:40 +0100 Subject: [PATCH 09/34] Run the picker instead of forcing piper, and stop chat duplicating the blocks Two faults, both mine. The benchmark was pinned to --only piper because I concluded the picker "reads stdin and exits on EOF, so it cannot be driven from a boot command". That is true of the docker exec I tested through and false of the thing it stands in for: a terminal block is a PTY. Driving benchmark.py under a real pty shows the picker renders and takes input exactly as it does on a laptop -- and it is now an arrow-key TUI, which is not the shape of something meant to be bypassed. The reader chooses; the boot command no longer chooses for them. The duplicate terminal and browser were the setupGuide. A setupGuide is a prompt addressed to Orcabot chat -- slopcodebench's opens "you are Orcabot chat" and tells it in as many words not to call create_browser -- and mine was written as documentation for a human, describing a dashboard that clones a repo and shows a report. Chat read that as a brief and built both blocks, on top of the two the template had already placed, leaving a second benchmark writing the same measurement files as the first. The guide now opens by saying the dashboard is already complete and forbids create_terminal and create_browser by name. Interactive selection moves two things that were previously fixed. Samples can no longer be synthesized before the sweep, because which engines to sample is only known once the reader has picked, so the selection is read back out of tools/sweep_rerun.json afterwards and the page and export are rebuilt. And the page has to be seeded from tools/page_template.html before that rebuild: build_benchmark_page.py substitutes into an existing page rather than writing one, and benchmark.py only seeds it on a run it completes -- so quitting the picker with prior measurements on disk died on FileNotFoundError with every clip already synthesized. That path is now tested, because it is the one that found the bug. Also corrects the seeding docs. There is no code seed for cloud: seedStarterTemplates returns early unless SURFACE_TOKEN is set and only the desktop app sets it, so every production row is a direct insert -- which is why the five templates in the prod grid are all hand-inserted under one author. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- benchmarks/README.md | 23 +++++-- benchmarks/talkito-tts/README.md | 91 +++++++++++++++++----------- benchmarks/talkito-tts/template.json | 4 +- 3 files changed, 77 insertions(+), 41 deletions(-) diff --git a/benchmarks/README.md b/benchmarks/README.md index 55bd0e55..c01b27c4 100644 --- a/benchmarks/README.md +++ b/benchmarks/README.md @@ -10,7 +10,7 @@ results surface in a browser/note block. | Benchmark | Dir | Runnable? | Execution model | |-----------|-----|-----------|-----------------| | slop-code-bench | [`slopcodebench/`](slopcodebench/) | ✅ yes | **Host** (no Docker) via the fork's tmux executor | -| Open-weight TTS | [`talkito-tts/`](talkito-tts/) | ✅ yes | **Host**, one PTY; no agent, no keys, ~3.5 min for piper | +| Open-weight TTS | [`talkito-tts/`](talkito-tts/) | ✅ yes | **Host**, one PTY; no agent, no keys; reader picks engines interactively | | SWE-bench (Lite/Verified/Full) | [`swebench/`](swebench/) | ✅ yes | Inference on host + **Docker-per-task** eval (Modal or DinD) | | SWE-bench Pro (public 731) | [`swebench-pro/`](swebench-pro/) | ✅ yes | Inference on host + **Docker-per-task** eval (`--use_local_docker` or Modal) | | Terminal-Bench 2.1 | [`terminal-bench/`](terminal-bench/) | ✅ yes | **Harbor** harness; Docker-per-task (or `--env daytona`/Modal remote) | @@ -74,9 +74,24 @@ benchmarks/bin/seed-benchmark-templates swebench # just one, by dir name It upserts each `benchmarks/*/template.json` (delete-by-name then insert, `status='approved'`) against the desktop d1-shim (`D1_URL`, default `http://127.0.0.1:9001`). `slopcodebench/bin/scb-insert-template` is a thin -wrapper that seeds only the SlopCodeBench template. For a **remote/cloud** control -plane, export from an existing dashboard via the templates API (`POST /templates`) -instead — direct INSERT is the local/curated path. +wrapper that seeds only the SlopCodeBench template. + +For a **remote/cloud** control plane there is no code seed at all: +`templates.seedStarterTemplates` returns early unless `SURFACE_TOKEN` is set, and only +the desktop app sets it, so `starter-templates.json` never reaches production. Every +production row is a direct insert. Use the script's `--sql` mode: + +``` +AUTHOR_ID=
AUTHOR_NAME=Orcabot \ + benchmarks/bin/seed-benchmark-templates --sql talkito-tts > /tmp/seed.sql +cd controlplane && npx wrangler d1 execute orcabot-db-production --remote --file /tmp/seed.sql +``` + +`AUTHOR_ID` is not optional. It defaults to `dev-desktop`, which exists on the desktop +shim but not in a cloud or wrangler-dev database, and `dashboard_templates.author_id` +is a foreign key into `users` — the insert then fails with `FOREIGN KEY constraint +failed`, quietly enough to look like success. Grep the output for `ERROR` rather than +trusting a clean exit. ## Status diff --git a/benchmarks/talkito-tts/README.md b/benchmarks/talkito-tts/README.md index dffd9925..18186c1b 100644 --- a/benchmarks/talkito-tts/README.md +++ b/benchmarks/talkito-tts/README.md @@ -7,6 +7,10 @@ runnable version of the published [Testing open-weight Text To Speech models](https://orcabot.com/benchmarks/open-weight-tts) table: same harness, same 84-phrase corpus, same page. +The reader chooses which engines to measure. The terminal opens the benchmark's own +checklist — arrows move, space toggles, `n` clears, Enter runs — and everything after +that is automatic. + ## Why this one has no orchestrator Every other template here drives a benchmark through an agent, because those @@ -18,9 +22,16 @@ parts are a terminal and a browser block. That also makes it the cheapest template here to run: no LLM calls, no API key, no tokens. It costs CPU time and nothing else. +**The setup guide's job is to stop Orcabot chat from helping.** A `setupGuide` is a +prompt addressed to chat, not documentation for a human. An earlier version of this +one described the dashboard instead of instructing chat, so chat obligingly called +`create_terminal` and `create_browser` — and the reader got two of each, with a second +benchmark writing the same measurement files as the first. The guide now opens by +saying the dashboard is already complete and forbids both calls by name. + ## Architecture - terminal block ──runs──> tools/benchmark.py --only piper + terminal block ──runs──> tools/benchmark.py (interactive picker) │ │ │ └──> tts-benchmark.html └──serves──> python3 -m http.server 8099 <──reads──┘ @@ -32,10 +43,10 @@ that refreshes every ten seconds. Without that, the browser block would show a connection error for the whole run and only become useful at the very end. When the report is built it is copied over the placeholder and the next refresh picks it up. -## Four things the sandbox forces +## What the sandbox forces -All four were found by running it, not by reading it. Two of them fail *quietly* -- -the run still exits 0 and still publishes a page. +All of these were found by running it, not by reading it, and two of them fail +*quietly* — the run still exits 0 and still publishes a page. **HTTPS clones, not submodules.** talkito's `.gitmodules` points at `git@github.com:robdmac/...`, and the sandbox has no SSH key, so @@ -47,62 +58,72 @@ repos over HTTPS instead and arranges them as The harness ships inside talkito-benchmark, so `tests-repo` is not needed. `tools/paths.py` still wants a talkito **checkout** rather than the installed package --- it looks for `talkito/tts.py` -- and `benchmark/` inside `talkito/` is the layout it +— it looks for `talkito/tts.py` — and `benchmark/` inside `talkito/` is the layout it resolves via `../..`. -**`--only`, never the picker.** `tools/benchmark.py` with no arguments opens an -interactive checklist that reads stdin and exits on EOF, so it cannot be driven from a -boot command. - **`TALKITO_AUTO_APPROVE_DOWNLOADS=1`.** talkito refuses to fetch model weights when it -cannot ask a human, and a boot command has no TTY. Without it every phrase fails with -`Piper model unavailable` -- and the sweep still reports `rc=0`, having published a -table reading `0/84 phrases, WER 0.0%, RTF 0.00x`. A green run measuring nothing is -the worst failure available here, so this variable is not optional. +cannot ask a human. The picker is interactive, so it *could* ask — but the prompt would +land in the middle of a sweep the reader has walked away from, so it is answered in +advance. Without it every phrase fails with `Piper model unavailable` and the sweep +still reports `rc=0`, having published a table reading `0/84 phrases, WER 0.0%, +RTF 0.00x`. A green run measuring nothing is the worst failure available here. **ffmpeg.** `make_samples.py` shells out to it for the wav-to-mp3 step and the sandbox image ships without it. Missing, every number is still correct and only the play -button is lost -- which is most of the point of the page. +button is lost — which is most of the point of the page. + +**Samples come after the run, not before.** `benchmark.py` sweeps and builds the page +but never synthesizes the listening clips the page embeds, and in interactive mode +*which* engines to sample is only known once the reader has chosen. So the boot command +reads the selection back out of `tools/sweep_rerun.json` after the run, synthesizes +clips for exactly those, and rebuilds the page and the export. + +**The page has to exist before it can be rebuilt.** `build_benchmark_page.py` +substitutes into an existing page rather than writing one, and `benchmark.py` only +seeds that page on a run it actually completes. Quit the picker with prior +measurements already on disk and the rebuild dies on `FileNotFoundError` with every +clip already synthesized, so the boot command seeds from `tools/page_template.html` +first. **`TTS_SAMPLE_DIR` is deliberately not set.** The three scripts want different directories: `make_samples.py` writes mp3s into an `mp3` subdirectory that the page -builder reads and the export builder does not. Each now defaults correctly on its own, -so setting one value for all of them is a way to break the export, not fix it. - -## Running more than piper +builder reads and the export builder does not. Each defaults correctly on its own, so +setting one value for all of them is a way to break the export, not fix it. -`MODELS` is the first line of the boot command and the only thing to change: +## Choosing engines - MODELS=piper # the default - MODELS=piper,kokoro - MODELS=piper,kittentts +Everything is ticked when the picker opens, and a **full sweep is around eleven +hours** — longer on a first run, because every engine downloads its own weights. Press +`n` to clear, then space to pick a few. -piper is the default because it is the cheapest complete arm in the table: it uses no -torch at all, and scoring runs through faster-whisper on CPU via CTranslate2, so the -install is seconds rather than gigabytes. Most other engines pull torch and download -their own weights on first use. +piper is the cheapest complete arm in the table: no torch at all, and scoring runs +through faster-whisper on CPU via CTranslate2, so the install is seconds rather than +gigabytes and the run is about three and a half minutes. Most other engines pull torch. -There are 73 providers. A **full sweep is around 11 hours**, longer on a first run. -Pointing this template at all of them is not a mistake the boot command should make -easy to make by accident. +The picker shows what each engine cost on previous runs, and `?` where it has never +been measured. ## Status **Validated end to end in the Docker sandbox**, from an empty workspace: clone, dependency install, weight download, 84-phrase sweep, Whisper scoring, page build and -serving. The run takes about three and a half minutes for piper and finishes +serving. A piper run finishes piper 82/84 phrases WER 5.3% RTF 0.08x -against 4% and 0.04x for the published Apple M2 figures -- close, and real-time factor +against 4% and 0.04x for the published Apple M2 figures — close, and real-time factor is wall-clock so a throttled container is expected to read higher. -The report reaches the browser block at 50 KB with the listening clip embedded, and -the export zip carries `samples/piper.mp3` alongside it. +The picker was driven under a real PTY to confirm it renders and takes input the way a +terminal block presents it, and the quit-without-running path was tested separately +because that is the one that exposed the missing-page bug above. The report reaches the +browser block at about 50 KB with the listening clip embedded, and the export zip +carries `samples/piper.mp3` alongside it. ## Seeding benchmarks/bin/seed-benchmark-templates talkito-tts -See [`../README.md`](../README.md) for the general path and the remote/cloud -alternative. +A cloud control plane needs a direct insert instead: `seedStarterTemplates` returns +early unless `SURFACE_TOKEN` is set, and only the desktop app sets it, so the code seed +never runs in production. See [`../README.md`](../README.md). diff --git a/benchmarks/talkito-tts/template.json b/benchmarks/talkito-tts/template.json index 143ac3be..fe298246 100644 --- a/benchmarks/talkito-tts/template.json +++ b/benchmarks/talkito-tts/template.json @@ -7,7 +7,7 @@ { "placeholderId": "runner", "type": "terminal", - "content": "{\"name\": \"TTS benchmark\", \"agentic\": false, \"bootCommand\": \"MODELS=piper\\nPORT=8099\\n# talkito refuses to fetch model weights when it cannot ask a human, and a boot\\n# command has no TTY. Without this every phrase fails with \\\"Piper model\\n# unavailable\\\" while the run still exits 0 and publishes an empty table.\\nexport TALKITO_AUTO_APPROVE_DOWNLOADS=1\\nROOT=/workspace/talkito\\nBENCH=$ROOT/benchmark\\nSERVE=/workspace/tts-report\\necho \\\"== Orcabot: talkito TTS benchmark ($MODELS) ==\\\"\\n\\n# Serve a placeholder first, so the browser block has something at a stable URL from\\n# the first second rather than an error until the run finishes.\\nmkdir -p \\\"$SERVE\\\"\\n[ -f \\\"$SERVE/index.html\\\" ] || printf '%s' 'TTS benchmark

Benchmark running...

This page reloads every 10 seconds. The report replaces it when the run finishes.

' > \\\"$SERVE/index.html\\\"\\ncurl -sf -o /dev/null --max-time 2 \\\"http://127.0.0.1:$PORT/\\\" || ( cd \\\"$SERVE\\\" && nohup python3 -m http.server \\\"$PORT\\\" >/workspace/.tts-serve.log 2>&1 & )\\n\\n# talkito's .gitmodules uses git@github.com: URLs and the sandbox has no SSH key,\\n# so --recurse-submodules fails; both repos are cloned over HTTPS instead.\\n# The harness now ships inside talkito-benchmark, so tests-repo is not needed --\\n# but tools/paths.py still wants a talkito CHECKOUT (it looks for talkito/tts.py,\\n# not the installed package), and benchmark/ inside talkito/ is the layout it\\n# resolves via \\\"../..\\\".\\n[ -d \\\"$ROOT/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito.git \\\"$ROOT\\\"\\n[ -d \\\"$BENCH/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito-benchmark.git \\\"$BENCH\\\"\\n\\n# make_samples.py shells out to ffmpeg to turn the listening clip into an mp3, and\\n# the sandbox image ships without it. Missing, the sweep still scores normally and\\n# only the play button is lost, so this is a soft dependency -- but a table of\\n# numbers with nothing to hear is most of the point of the page.\\ncommand -v ffmpeg >/dev/null 2>&1 || { apt-get update -qq && apt-get install -y -qq --no-install-recommends ffmpeg; }\\n\\n# piper needs no torch at all, and faster-whisper scores on CPU through CTranslate2,\\n# so the whole piper arm installs in seconds. Heavier models would need much more.\\npython3 -c 'import piper' 2>/dev/null || pip3 install --break-system-packages -q piper-tts\\npython3 -c 'import faster_whisper' 2>/dev/null || pip3 install --break-system-packages -q faster-whisper\\n\\ncd \\\"$BENCH\\\" || exec bash\\n# The listening clip has to exist before the page is built, or the table is numbers\\n# with nothing to play. benchmark.py builds the page at the end, so samples come first.\\npython3 tools/make_samples.py $(echo \\\"$MODELS\\\" | tr , ' ')\\n# TTS_SAMPLE_DIR is deliberately NOT set. The three scripts want different\\n# directories -- make_samples.py writes mp3s into a \\\"mp3\\\" subdirectory that the\\n# page builder reads and the export builder does not -- and each now defaults\\n# correctly on its own. Setting one value for all of them is what made the export\\n# report \\\"0 audio samples\\\".\\n# --only skips the interactive picker, which needs a TTY it does not have here.\\npython3 tools/benchmark.py --only \\\"$MODELS\\\"\\nrc=$?\\nif [ -f tts-benchmark.html ]; then\\n cp -f tts-benchmark.html \\\"$SERVE/index.html\\\"\\n echo \\\"report published -> http://localhost:$PORT/\\\"\\nelse\\n echo \\\"no tts-benchmark.html was produced; measurements (if any) are under $BENCH/tools/\\\"\\nfi\\necho \\\"[tts benchmark done rc=$rc]\\\"\\nexec bash\", \"subagentIds\": [], \"skillIds\": [], \"mcpToolIds\": []}", + "content": "{\"name\": \"TTS benchmark\", \"agentic\": false, \"bootCommand\": \"PORT=8099\\n# talkito refuses to fetch model weights when it cannot ask a human. The picker\\n# below is interactive, but the download prompt would land in the middle of a\\n# sweep the reader has walked away from, so it is answered in advance.\\nexport TALKITO_AUTO_APPROVE_DOWNLOADS=1\\nROOT=/workspace/talkito\\nBENCH=$ROOT/benchmark\\nSERVE=/workspace/tts-report\\necho \\\"== Orcabot: talkito TTS benchmark ==\\\"\\n\\n# Serve a placeholder first, so the browser block has something at a stable URL from\\n# the first second rather than an error until the run finishes.\\nmkdir -p \\\"$SERVE\\\"\\n[ -f \\\"$SERVE/index.html\\\" ] || printf '%s' 'TTS benchmark

Waiting for a run...

Pick engines in the terminal. This page reloads every 10 seconds and the report replaces it when the run finishes.

' > \\\"$SERVE/index.html\\\"\\ncurl -sf -o /dev/null --max-time 2 \\\"http://127.0.0.1:$PORT/\\\" || ( cd \\\"$SERVE\\\" && nohup python3 -m http.server \\\"$PORT\\\" >/workspace/.tts-serve.log 2>&1 & )\\n\\n# talkito's .gitmodules uses git@github.com: URLs and the sandbox has no SSH key,\\n# so --recurse-submodules fails; both repos are cloned over HTTPS instead.\\n# The harness ships inside talkito-benchmark, so tests-repo is not needed -- but\\n# tools/paths.py still wants a talkito CHECKOUT (it looks for talkito/tts.py, not\\n# the installed package), and benchmark/ inside talkito/ is the layout it resolves.\\n[ -d \\\"$ROOT/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito.git \\\"$ROOT\\\"\\n[ -d \\\"$BENCH/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito-benchmark.git \\\"$BENCH\\\"\\n\\n# make_samples.py shells out to ffmpeg for the listening clip, and the sandbox image\\n# ships without it. Missing, the sweep still scores normally and only the play button\\n# is lost -- but that is most of the point of the page.\\ncommand -v ffmpeg >/dev/null 2>&1 || { apt-get update -qq && apt-get install -y -qq --no-install-recommends ffmpeg; }\\n\\n# piper needs no torch at all, and faster-whisper scores on CPU through CTranslate2,\\n# so the cheapest arm installs in seconds. Heavier engines pull their own weights on\\n# first use and the picker warns what each has cost before.\\npython3 -c 'import piper' 2>/dev/null || pip3 install --break-system-packages -q piper-tts\\npython3 -c 'import faster_whisper' 2>/dev/null || pip3 install --break-system-packages -q faster-whisper\\n\\ncd \\\"$BENCH\\\" || exec bash\\ncat <<'BANNER'\\n\\n Everything is selected by default and a full sweep is around eleven hours.\\n Press n to clear, then space to pick a few. piper alone takes about three minutes.\\n\\nBANNER\\n# No --only: this is the interactive picker, and a terminal block is a real PTY, so\\n# the arrow-key menu works exactly as it does on a laptop.\\npython3 tools/benchmark.py\\nrc=$?\\n\\n# benchmark.py sweeps and builds the page, but never synthesizes the one listening\\n# clip per engine that the page embeds - so the table it just built has nothing to\\n# play. Which engines to sample is only known now, after the reader has chosen, so\\n# they are read back out of the state the sweep wrote and the page is rebuilt.\\nMEASURED=$(python3 -c \\\"import json;print(' '.join(sorted({k.split('|')[0] for k in json.load(open('tools/sweep_rerun.json'))['units']})))\\\" 2>/dev/null)\\nif [ -n \\\"$MEASURED\\\" ]; then\\n echo \\\"== adding listening clips for: $MEASURED ==\\\"\\n python3 tools/make_samples.py $MEASURED\\n # build_benchmark_page.py substitutes into an existing page rather than writing\\n # one, and benchmark.py only seeds that page on a run it actually completes. Quit\\n # the picker with prior measurements on disk and there is nothing to substitute\\n # into, so it dies on FileNotFoundError with every clip already synthesized.\\n [ -f tts-benchmark.html ] || cp tools/page_template.html tts-benchmark.html\\n python3 tools/build_benchmark_page.py tts-benchmark.html\\n python3 tools/build_export.py tts-benchmark-export.zip tts-benchmark.html\\nfi\\n\\nif [ -f tts-benchmark.html ]; then\\n cp -f tts-benchmark.html \\\"$SERVE/index.html\\\"\\n echo \\\"report published -> http://localhost:$PORT/\\\"\\nelse\\n echo \\\"no tts-benchmark.html was produced; measurements (if any) are under $BENCH/tools/\\\"\\nfi\\necho \\\"[tts benchmark done rc=$rc]\\\"\\nexec bash\", \"subagentIds\": [], \"skillIds\": [], \"mcpToolIds\": []}", "position": { "x": 0, "y": 0 @@ -37,5 +37,5 @@ "y": 0, "zoom": 1 }, - "setupGuide": "# talkito TTS benchmark\n\nThis template runs the open-weight TTS benchmark from\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) end to end inside\none terminal, and shows the report it produces in the browser block beside it.\n\nUnlike the other benchmark templates there is **no orchestrator**. Nothing here needs\nOrcabot chat, an agent, or an API key: the terminal's boot command is the whole run,\nand the browser block is pointed at a local web server the same command starts. Start\nthe dashboard and it goes.\n\n## What the terminal does\n\n1. Serves a placeholder page on `localhost:8099`, so the browser block has a stable\n URL immediately instead of an error until the run finishes.\n2. Clones `talkito` and `talkito-benchmark` over **HTTPS**. talkito's `.gitmodules`\n uses `git@github.com:` URLs, which cannot work in the sandbox (no SSH key), so\n `--recurse-submodules` is deliberately not used. The harness ships inside\n talkito-benchmark now, so `tests-repo` is not needed -- but `tools/paths.py` still\n wants a talkito *checkout* rather than the installed package (it looks for\n `talkito/tts.py`), and `benchmark/` inside `talkito/` is the layout it resolves.\n3. Installs `piper-tts` and `faster-whisper`. Piper uses no torch at all and\n faster-whisper scores on CPU via CTranslate2, so this arm is small and quick.\n4. Synthesizes the one listening clip per model (`make_samples.py`) **before** the\n sweep, because `benchmark.py` builds the page at the end and a sample produced\n afterwards would not be in it.\n5. Runs `python3 tools/benchmark.py --only piper`, which sweeps the 84-phrase corpus,\n transcribes each clip back with Whisper, scores word error rate, then rebuilds the\n page and the export zip.\n6. Copies `tts-benchmark.html` over the placeholder, so the browser block shows the\n report on its next refresh.\n\n## Running more models\n\n`MODELS` on the first line of the boot command is the only thing to change. It is\npassed to `--only`, so it is comma-separated:\n\n MODELS=piper # the default: minutes\n MODELS=piper,kokoro # kokoro pulls torch, so expect a long first run\n MODELS=piper,kittentts\n\nThere are 73 providers. A **full sweep is around 11 hours** and much longer on a\nfirst run, because every engine downloads its own weights -- which is why this\ntemplate ships pinned to piper rather than to everything.\n\n## Reading the result\n\nThe report is the same page the published benchmark is built from: one row per\nconfiguration, with word error rate, real-time factor, compute per phrase, PESQ and a\nplay button per row. `tts-benchmark-export.zip` beside it is the shareable archive\n(page + measurements + audio).\n\nMeasurements are saved per unit as they complete, so an interrupted run keeps its\nfinished work and re-running resumes rather than restarting. `--keep` adds to existing\nmeasurements instead of replacing them.\n\n## If the run stops early\n\nThe terminal drops to a shell either way and prints `[tts benchmark done rc=N]`.\n`rc` is `benchmark.py`'s exit code -- non-zero means at least one provider failed, and\nthe per-provider output above it says which. The page and the zip can always be\nrebuilt from stored measurements alone:\n\n cd /workspace/talkito/benchmark\n python3 tools/build_benchmark_page.py tts-benchmark.html\n" + "setupGuide": "# TTS benchmark (you are Orcabot chat)\n\n**This dashboard is already complete. Do not set anything up.**\n\nThe template provisions both blocks and the terminal's boot command does the entire\njob by itself: it clones the benchmark, installs what it needs, starts a local web\nserver, opens the engine picker, and publishes the report to that server when the run\nfinishes.\n\n## Do not create anything\n\n**Do NOT call create_terminal. Do NOT call create_browser.** There is already one of\neach on the canvas. Creating more gives the reader two terminals and two browsers,\nwith a second benchmark running against the same workspace as the first -- both\nwriting the same measurement files, which corrupts the run rather than doubling it.\n\nThere are no keys to set, no secrets to add, no provider to choose, and no config file\nto write. This benchmark makes no LLM calls and needs no API key at all.\n\nIf the reader says \"walk me through setting up this dashboard\", the correct answer is\nthat there is nothing to set up: the terminal is already doing it, and the picker is\nwaiting for them.\n\n## What the reader is looking at\n\nThe terminal opens a checklist of every engine the benchmark knows. Arrows move, space\ntoggles, `n` clears the selection, `a` selects all, Enter starts, `q` quits. Everything\nis ticked by default and the picker shows what each engine cost on previous runs.\n\nTell them to press `n` first unless they mean it: a full sweep is around eleven hours,\nand longer on a first run because each engine downloads its own weights. piper alone is\nabout three minutes and is the cheapest complete arm in the table.\n\nThe browser block shows a placeholder that reloads every ten seconds, and the report\nreplaces it when the run finishes.\n\n## Answering \"how is it going?\"\n\nRead the terminal. Each engine prints a line per category as it completes, then a\nsummary line of the form `piper 82/84 phrases WER 5.0% RTF 0.07x`. The run ends with\n`[tts benchmark done rc=N]`; a non-zero `rc` means at least one engine failed and the\nper-engine output above says which.\n\nMeasurements are saved per unit as they complete, so an interrupted run keeps its\nfinished work and starting again resumes rather than restarting.\n\n## If they want to re-run or change the selection\n\nThey type `python3 tools/benchmark.py` in the same terminal -- it drops to a shell in\nthe right directory when the run ends. `--keep` adds to the existing measurements\ninstead of replacing the ones for the engines they pick. The page and the export can\nalways be rebuilt from stored measurements alone:\n\n python3 tools/build_benchmark_page.py tts-benchmark.html\n" } From 3fd774facbf70471dbe4eab928cbc0fea6a6c116 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Fri, 21 Aug 2026 23:13:51 +0100 Subject: [PATCH 10/34] Drop the listening clips, and declare the report's encoding Nothing plays. make_samples.py exists to embed one clip per engine, audio does not reach the reader through the browser block, and that clip was the only reason this template needed ffmpeg. ffmpeg was the only reason it ran apt-get. And apt-get was the only reason it reached deb.nodesource.com, which is not in the default egress allowlist and so raised an approval prompt on every fresh workspace -- the sandbox image installs Node from NodeSource at build time and leaves /etc/apt/sources.list.d/nodesource.sources behind, so any apt-get update inside the sandbox reaches for it whatever it was asked to install. One thing nobody wanted was pulling three the reader had to care about. All of it goes: no clips, no ffmpeg, no apt-get, no egress prompt. The table is numbers, which is what it was being read for. That also retires two gotchas this template had accumulated. Samples no longer have to be deferred until after the picker, and the page no longer has to be seeded from page_template.html before a rebuild, because there is no rebuild. Separately, the report renders as mojibake -- "TTS Engine Benchmark - 24 configurations" arriving as garbage where the em dash should be. The file is valid UTF-8; it just says so nowhere. The page carries no and python3 -m http.server sends "text/html" with no charset either, so the browser falls back to its locale default and reads UTF-8 as cp1252. The tag is prepended to the published copy rather than patched into the source, so the export zip stays byte-identical to what the tools wrote. The real fix for that belongs upstream in tools/page_template.html, which would also cover the zip and any copy opened from disk rather than served. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- benchmarks/talkito-tts/README.md | 44 ++++++++++++---------------- benchmarks/talkito-tts/template.json | 2 +- 2 files changed, 20 insertions(+), 26 deletions(-) diff --git a/benchmarks/talkito-tts/README.md b/benchmarks/talkito-tts/README.md index 18186c1b..95ceeb8b 100644 --- a/benchmarks/talkito-tts/README.md +++ b/benchmarks/talkito-tts/README.md @@ -68,27 +68,22 @@ advance. Without it every phrase fails with `Piper model unavailable` and the sw still reports `rc=0`, having published a table reading `0/84 phrases, WER 0.0%, RTF 0.00x`. A green run measuring nothing is the worst failure available here. -**ffmpeg.** `make_samples.py` shells out to it for the wav-to-mp3 step and the sandbox -image ships without it. Missing, every number is still correct and only the play -button is lost — which is most of the point of the page. - -**Samples come after the run, not before.** `benchmark.py` sweeps and builds the page -but never synthesizes the listening clips the page embeds, and in interactive mode -*which* engines to sample is only known once the reader has chosen. So the boot command -reads the selection back out of `tools/sweep_rerun.json` after the run, synthesizes -clips for exactly those, and rebuilds the page and the export. - -**The page has to exist before it can be rebuilt.** `build_benchmark_page.py` -substitutes into an existing page rather than writing one, and `benchmark.py` only -seeds that page on a run it actually completes. Quit the picker with prior -measurements already on disk and the rebuild dies on `FileNotFoundError` with every -clip already synthesized, so the boot command seeds from `tools/page_template.html` -first. - -**`TTS_SAMPLE_DIR` is deliberately not set.** The three scripts want different -directories: `make_samples.py` writes mp3s into an `mp3` subdirectory that the page -builder reads and the export builder does not. Each defaults correctly on its own, so -setting one value for all of them is a way to break the export, not fix it. +**No listening clips, and therefore no ffmpeg.** `make_samples.py` exists to embed one +playable clip per engine, and it shells out to ffmpeg for the wav-to-mp3 step. Audio +does not reach the reader through the browser block, so the clips are not synthesized +at all — which removes ffmpeg, which removes the only `apt-get` this template ran, +which removes an egress approval prompt for `deb.nodesource.com`. The sandbox image +installs Node from NodeSource at build time and leaves the apt source configured, so +*any* `apt-get update` inside the sandbox reaches for it. The table is numbers. + +**The published page needs its encoding declared.** The report carries no +`` and `python3 -m http.server` sends `text/html` with no charset either, +so a browser falls back to its locale default and renders every em dash as mojibake — +`TTS Engine Benchmark — 24 configurations` arriving as `ââ¬â`. The bytes are valid +UTF-8; only the declaration is missing. The boot command prepends the meta tag to the +*published copy* rather than patching the source, so the export zip stays byte-identical +to what the tools wrote. The real fix belongs upstream in `tools/page_template.html`, +which would also fix the zip and any copy opened from disk. ## Choosing engines @@ -115,10 +110,9 @@ against 4% and 0.04x for the published Apple M2 figures — close, and real-time is wall-clock so a throttled container is expected to read higher. The picker was driven under a real PTY to confirm it renders and takes input the way a -terminal block presents it, and the quit-without-running path was tested separately -because that is the one that exposed the missing-page bug above. The report reaches the -browser block at about 50 KB with the listening clip embedded, and the export zip -carries `samples/piper.mp3` alongside it. +terminal block presents it, and the quit-without-running path was tested separately. +The report reaches the browser block as a table of numbers with its encoding declared; +nothing is playable, by design. ## Seeding diff --git a/benchmarks/talkito-tts/template.json b/benchmarks/talkito-tts/template.json index fe298246..c1333024 100644 --- a/benchmarks/talkito-tts/template.json +++ b/benchmarks/talkito-tts/template.json @@ -7,7 +7,7 @@ { "placeholderId": "runner", "type": "terminal", - "content": "{\"name\": \"TTS benchmark\", \"agentic\": false, \"bootCommand\": \"PORT=8099\\n# talkito refuses to fetch model weights when it cannot ask a human. The picker\\n# below is interactive, but the download prompt would land in the middle of a\\n# sweep the reader has walked away from, so it is answered in advance.\\nexport TALKITO_AUTO_APPROVE_DOWNLOADS=1\\nROOT=/workspace/talkito\\nBENCH=$ROOT/benchmark\\nSERVE=/workspace/tts-report\\necho \\\"== Orcabot: talkito TTS benchmark ==\\\"\\n\\n# Serve a placeholder first, so the browser block has something at a stable URL from\\n# the first second rather than an error until the run finishes.\\nmkdir -p \\\"$SERVE\\\"\\n[ -f \\\"$SERVE/index.html\\\" ] || printf '%s' 'TTS benchmark

Waiting for a run...

Pick engines in the terminal. This page reloads every 10 seconds and the report replaces it when the run finishes.

' > \\\"$SERVE/index.html\\\"\\ncurl -sf -o /dev/null --max-time 2 \\\"http://127.0.0.1:$PORT/\\\" || ( cd \\\"$SERVE\\\" && nohup python3 -m http.server \\\"$PORT\\\" >/workspace/.tts-serve.log 2>&1 & )\\n\\n# talkito's .gitmodules uses git@github.com: URLs and the sandbox has no SSH key,\\n# so --recurse-submodules fails; both repos are cloned over HTTPS instead.\\n# The harness ships inside talkito-benchmark, so tests-repo is not needed -- but\\n# tools/paths.py still wants a talkito CHECKOUT (it looks for talkito/tts.py, not\\n# the installed package), and benchmark/ inside talkito/ is the layout it resolves.\\n[ -d \\\"$ROOT/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito.git \\\"$ROOT\\\"\\n[ -d \\\"$BENCH/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito-benchmark.git \\\"$BENCH\\\"\\n\\n# make_samples.py shells out to ffmpeg for the listening clip, and the sandbox image\\n# ships without it. Missing, the sweep still scores normally and only the play button\\n# is lost -- but that is most of the point of the page.\\ncommand -v ffmpeg >/dev/null 2>&1 || { apt-get update -qq && apt-get install -y -qq --no-install-recommends ffmpeg; }\\n\\n# piper needs no torch at all, and faster-whisper scores on CPU through CTranslate2,\\n# so the cheapest arm installs in seconds. Heavier engines pull their own weights on\\n# first use and the picker warns what each has cost before.\\npython3 -c 'import piper' 2>/dev/null || pip3 install --break-system-packages -q piper-tts\\npython3 -c 'import faster_whisper' 2>/dev/null || pip3 install --break-system-packages -q faster-whisper\\n\\ncd \\\"$BENCH\\\" || exec bash\\ncat <<'BANNER'\\n\\n Everything is selected by default and a full sweep is around eleven hours.\\n Press n to clear, then space to pick a few. piper alone takes about three minutes.\\n\\nBANNER\\n# No --only: this is the interactive picker, and a terminal block is a real PTY, so\\n# the arrow-key menu works exactly as it does on a laptop.\\npython3 tools/benchmark.py\\nrc=$?\\n\\n# benchmark.py sweeps and builds the page, but never synthesizes the one listening\\n# clip per engine that the page embeds - so the table it just built has nothing to\\n# play. Which engines to sample is only known now, after the reader has chosen, so\\n# they are read back out of the state the sweep wrote and the page is rebuilt.\\nMEASURED=$(python3 -c \\\"import json;print(' '.join(sorted({k.split('|')[0] for k in json.load(open('tools/sweep_rerun.json'))['units']})))\\\" 2>/dev/null)\\nif [ -n \\\"$MEASURED\\\" ]; then\\n echo \\\"== adding listening clips for: $MEASURED ==\\\"\\n python3 tools/make_samples.py $MEASURED\\n # build_benchmark_page.py substitutes into an existing page rather than writing\\n # one, and benchmark.py only seeds that page on a run it actually completes. Quit\\n # the picker with prior measurements on disk and there is nothing to substitute\\n # into, so it dies on FileNotFoundError with every clip already synthesized.\\n [ -f tts-benchmark.html ] || cp tools/page_template.html tts-benchmark.html\\n python3 tools/build_benchmark_page.py tts-benchmark.html\\n python3 tools/build_export.py tts-benchmark-export.zip tts-benchmark.html\\nfi\\n\\nif [ -f tts-benchmark.html ]; then\\n cp -f tts-benchmark.html \\\"$SERVE/index.html\\\"\\n echo \\\"report published -> http://localhost:$PORT/\\\"\\nelse\\n echo \\\"no tts-benchmark.html was produced; measurements (if any) are under $BENCH/tools/\\\"\\nfi\\necho \\\"[tts benchmark done rc=$rc]\\\"\\nexec bash\", \"subagentIds\": [], \"skillIds\": [], \"mcpToolIds\": []}", + "content": "{\"name\": \"TTS benchmark\", \"agentic\": false, \"bootCommand\": \"PORT=8099\\n# talkito refuses to fetch model weights when it cannot ask a human. The picker\\n# below is interactive, but the download prompt would land in the middle of a\\n# sweep the reader has walked away from, so it is answered in advance.\\nexport TALKITO_AUTO_APPROVE_DOWNLOADS=1\\nROOT=/workspace/talkito\\nBENCH=$ROOT/benchmark\\nSERVE=/workspace/tts-report\\necho \\\"== Orcabot: talkito TTS benchmark ==\\\"\\n\\n# Serve a placeholder first, so the browser block has something at a stable URL from\\n# the first second rather than an error until the run finishes.\\nmkdir -p \\\"$SERVE\\\"\\n[ -f \\\"$SERVE/index.html\\\" ] || printf '%s' 'TTS benchmark

Waiting for a run...

Pick engines in the terminal. This page reloads every 10 seconds and the report replaces it when the run finishes.

' > \\\"$SERVE/index.html\\\"\\ncurl -sf -o /dev/null --max-time 2 \\\"http://127.0.0.1:$PORT/\\\" || ( cd \\\"$SERVE\\\" && nohup python3 -m http.server \\\"$PORT\\\" >/workspace/.tts-serve.log 2>&1 & )\\n\\n# talkito's .gitmodules uses git@github.com: URLs and the sandbox has no SSH key,\\n# so --recurse-submodules fails; both repos are cloned over HTTPS instead.\\n# The harness ships inside talkito-benchmark, so tests-repo is not needed -- but\\n# tools/paths.py still wants a talkito CHECKOUT (it looks for talkito/tts.py, not\\n# the installed package), and benchmark/ inside talkito/ is the layout it resolves.\\n[ -d \\\"$ROOT/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito.git \\\"$ROOT\\\"\\n[ -d \\\"$BENCH/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito-benchmark.git \\\"$BENCH\\\"\\n\\n# piper needs no torch at all, and faster-whisper scores on CPU through CTranslate2,\\n# so the cheapest arm installs in seconds. Heavier engines pull their own weights on\\n# first use and the picker warns what each has cost before.\\npython3 -c 'import piper' 2>/dev/null || pip3 install --break-system-packages -q piper-tts\\npython3 -c 'import faster_whisper' 2>/dev/null || pip3 install --break-system-packages -q faster-whisper\\n\\ncd \\\"$BENCH\\\" || exec bash\\ncat <<'BANNER'\\n\\n Everything is selected by default and a full sweep is around eleven hours.\\n Press n to clear, then space to pick a few. piper alone takes about three minutes.\\n\\nBANNER\\n# No --only: this is the interactive picker, and a terminal block is a real PTY, so\\n# the arrow-key menu works exactly as it does on a laptop.\\npython3 tools/benchmark.py\\nrc=$?\\n\\n# No listening clips: make_samples.py exists to embed one per engine, but audio does\\n# not reach the reader through the browser block, and it is the only reason this\\n# needed ffmpeg -- which in turn was the only reason the sandbox ran apt-get, which\\n# reaches deb.nodesource.com through a leftover apt source and trips egress approval.\\n# Dropping playback drops all three.\\n\\nif [ -f tts-benchmark.html ]; then\\n # The page declares no encoding and python3 -m http.server sends \\\"text/html\\\" with\\n # no charset either, so a browser falls back to its locale default and renders\\n # every em dash as mojibake. Declared on the published copy rather than patched\\n # into the source, so the export zip stays byte-identical to what the tools wrote.\\n if grep -qi ']*charset' tts-benchmark.html; then\\n cp -f tts-benchmark.html \\\"$SERVE/index.html\\\"\\n else\\n { printf '%s\\\\n' ''; cat tts-benchmark.html; } > \\\"$SERVE/index.html\\\"\\n fi\\n echo \\\"report published -> http://localhost:$PORT/\\\"\\nelse\\n echo \\\"no tts-benchmark.html was produced; measurements (if any) are under $BENCH/tools/\\\"\\nfi\\necho \\\"[tts benchmark done rc=$rc]\\\"\\nexec bash\", \"subagentIds\": [], \"skillIds\": [], \"mcpToolIds\": []}", "position": { "x": 0, "y": 0 From 429f1e0be6a73e90579743dcf7f79cbb266af6c3 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Fri, 21 Aug 2026 23:20:22 +0100 Subject: [PATCH 11/34] Take the upstream charset fix; keep the guard dormant page_template.html declares utf-8 now, so the boot command copies the report straight through and the injection never fires. Verified against a fresh clone: the plain-copy branch is taken and the published page carries exactly one meta tag rather than two. The fallback stays. The failure it covers is silent and total -- a page that declares no encoding, served by a server that sends none either, is read as the browser's locale default and every em dash becomes mojibake -- and this repo already keeps CELL_OVERRIDES for the same reason: an upstream fix that cannot quietly revert is worth a few lines. Its comment now says it is a fallback rather than describing a bug that is fixed. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- benchmarks/talkito-tts/README.md | 16 ++++++++-------- benchmarks/talkito-tts/template.json | 2 +- 2 files changed, 9 insertions(+), 9 deletions(-) diff --git a/benchmarks/talkito-tts/README.md b/benchmarks/talkito-tts/README.md index 95ceeb8b..8c32d52e 100644 --- a/benchmarks/talkito-tts/README.md +++ b/benchmarks/talkito-tts/README.md @@ -76,14 +76,14 @@ which removes an egress approval prompt for `deb.nodesource.com`. The sandbox im installs Node from NodeSource at build time and leaves the apt source configured, so *any* `apt-get update` inside the sandbox reaches for it. The table is numbers. -**The published page needs its encoding declared.** The report carries no -`` and `python3 -m http.server` sends `text/html` with no charset either, -so a browser falls back to its locale default and renders every em dash as mojibake — -`TTS Engine Benchmark — 24 configurations` arriving as `ââ¬â`. The bytes are valid -UTF-8; only the declaration is missing. The boot command prepends the meta tag to the -*published copy* rather than patching the source, so the export zip stays byte-identical -to what the tools wrote. The real fix belongs upstream in `tools/page_template.html`, -which would also fix the zip and any copy opened from disk. +**The published page's encoding is declared upstream now.** It was not: the report +carried no `` and `python3 -m http.server` sends `text/html` with no +charset either, so a browser fell back to its locale default and read valid UTF-8 as +cp1252 — `TTS Engine Benchmark — 24 configurations` arriving with garbage where the em +dash should be. `tools/page_template.html` declares it now, so the boot command simply +copies. It keeps a fallback that prepends the tag when a page lacks one, because the +failure is silent and total, and it writes to the published copy rather than the source +so the export zip stays byte-identical to what the tools wrote. ## Choosing engines diff --git a/benchmarks/talkito-tts/template.json b/benchmarks/talkito-tts/template.json index c1333024..5ab6edd7 100644 --- a/benchmarks/talkito-tts/template.json +++ b/benchmarks/talkito-tts/template.json @@ -7,7 +7,7 @@ { "placeholderId": "runner", "type": "terminal", - "content": "{\"name\": \"TTS benchmark\", \"agentic\": false, \"bootCommand\": \"PORT=8099\\n# talkito refuses to fetch model weights when it cannot ask a human. The picker\\n# below is interactive, but the download prompt would land in the middle of a\\n# sweep the reader has walked away from, so it is answered in advance.\\nexport TALKITO_AUTO_APPROVE_DOWNLOADS=1\\nROOT=/workspace/talkito\\nBENCH=$ROOT/benchmark\\nSERVE=/workspace/tts-report\\necho \\\"== Orcabot: talkito TTS benchmark ==\\\"\\n\\n# Serve a placeholder first, so the browser block has something at a stable URL from\\n# the first second rather than an error until the run finishes.\\nmkdir -p \\\"$SERVE\\\"\\n[ -f \\\"$SERVE/index.html\\\" ] || printf '%s' 'TTS benchmark

Waiting for a run...

Pick engines in the terminal. This page reloads every 10 seconds and the report replaces it when the run finishes.

' > \\\"$SERVE/index.html\\\"\\ncurl -sf -o /dev/null --max-time 2 \\\"http://127.0.0.1:$PORT/\\\" || ( cd \\\"$SERVE\\\" && nohup python3 -m http.server \\\"$PORT\\\" >/workspace/.tts-serve.log 2>&1 & )\\n\\n# talkito's .gitmodules uses git@github.com: URLs and the sandbox has no SSH key,\\n# so --recurse-submodules fails; both repos are cloned over HTTPS instead.\\n# The harness ships inside talkito-benchmark, so tests-repo is not needed -- but\\n# tools/paths.py still wants a talkito CHECKOUT (it looks for talkito/tts.py, not\\n# the installed package), and benchmark/ inside talkito/ is the layout it resolves.\\n[ -d \\\"$ROOT/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito.git \\\"$ROOT\\\"\\n[ -d \\\"$BENCH/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito-benchmark.git \\\"$BENCH\\\"\\n\\n# piper needs no torch at all, and faster-whisper scores on CPU through CTranslate2,\\n# so the cheapest arm installs in seconds. Heavier engines pull their own weights on\\n# first use and the picker warns what each has cost before.\\npython3 -c 'import piper' 2>/dev/null || pip3 install --break-system-packages -q piper-tts\\npython3 -c 'import faster_whisper' 2>/dev/null || pip3 install --break-system-packages -q faster-whisper\\n\\ncd \\\"$BENCH\\\" || exec bash\\ncat <<'BANNER'\\n\\n Everything is selected by default and a full sweep is around eleven hours.\\n Press n to clear, then space to pick a few. piper alone takes about three minutes.\\n\\nBANNER\\n# No --only: this is the interactive picker, and a terminal block is a real PTY, so\\n# the arrow-key menu works exactly as it does on a laptop.\\npython3 tools/benchmark.py\\nrc=$?\\n\\n# No listening clips: make_samples.py exists to embed one per engine, but audio does\\n# not reach the reader through the browser block, and it is the only reason this\\n# needed ffmpeg -- which in turn was the only reason the sandbox ran apt-get, which\\n# reaches deb.nodesource.com through a leftover apt source and trips egress approval.\\n# Dropping playback drops all three.\\n\\nif [ -f tts-benchmark.html ]; then\\n # The page declares no encoding and python3 -m http.server sends \\\"text/html\\\" with\\n # no charset either, so a browser falls back to its locale default and renders\\n # every em dash as mojibake. Declared on the published copy rather than patched\\n # into the source, so the export zip stays byte-identical to what the tools wrote.\\n if grep -qi ']*charset' tts-benchmark.html; then\\n cp -f tts-benchmark.html \\\"$SERVE/index.html\\\"\\n else\\n { printf '%s\\\\n' ''; cat tts-benchmark.html; } > \\\"$SERVE/index.html\\\"\\n fi\\n echo \\\"report published -> http://localhost:$PORT/\\\"\\nelse\\n echo \\\"no tts-benchmark.html was produced; measurements (if any) are under $BENCH/tools/\\\"\\nfi\\necho \\\"[tts benchmark done rc=$rc]\\\"\\nexec bash\", \"subagentIds\": [], \"skillIds\": [], \"mcpToolIds\": []}", + "content": "{\"name\": \"TTS benchmark\", \"agentic\": false, \"bootCommand\": \"PORT=8099\\n# talkito refuses to fetch model weights when it cannot ask a human. The picker\\n# below is interactive, but the download prompt would land in the middle of a\\n# sweep the reader has walked away from, so it is answered in advance.\\nexport TALKITO_AUTO_APPROVE_DOWNLOADS=1\\nROOT=/workspace/talkito\\nBENCH=$ROOT/benchmark\\nSERVE=/workspace/tts-report\\necho \\\"== Orcabot: talkito TTS benchmark ==\\\"\\n\\n# Serve a placeholder first, so the browser block has something at a stable URL from\\n# the first second rather than an error until the run finishes.\\nmkdir -p \\\"$SERVE\\\"\\n[ -f \\\"$SERVE/index.html\\\" ] || printf '%s' 'TTS benchmark

Waiting for a run...

Pick engines in the terminal. This page reloads every 10 seconds and the report replaces it when the run finishes.

' > \\\"$SERVE/index.html\\\"\\ncurl -sf -o /dev/null --max-time 2 \\\"http://127.0.0.1:$PORT/\\\" || ( cd \\\"$SERVE\\\" && nohup python3 -m http.server \\\"$PORT\\\" >/workspace/.tts-serve.log 2>&1 & )\\n\\n# talkito's .gitmodules uses git@github.com: URLs and the sandbox has no SSH key,\\n# so --recurse-submodules fails; both repos are cloned over HTTPS instead.\\n# The harness ships inside talkito-benchmark, so tests-repo is not needed -- but\\n# tools/paths.py still wants a talkito CHECKOUT (it looks for talkito/tts.py, not\\n# the installed package), and benchmark/ inside talkito/ is the layout it resolves.\\n[ -d \\\"$ROOT/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito.git \\\"$ROOT\\\"\\n[ -d \\\"$BENCH/.git\\\" ] || git clone --depth 1 https://github.com/robdmac/talkito-benchmark.git \\\"$BENCH\\\"\\n\\n# piper needs no torch at all, and faster-whisper scores on CPU through CTranslate2,\\n# so the cheapest arm installs in seconds. Heavier engines pull their own weights on\\n# first use and the picker warns what each has cost before.\\npython3 -c 'import piper' 2>/dev/null || pip3 install --break-system-packages -q piper-tts\\npython3 -c 'import faster_whisper' 2>/dev/null || pip3 install --break-system-packages -q faster-whisper\\n\\ncd \\\"$BENCH\\\" || exec bash\\ncat <<'BANNER'\\n\\n Everything is selected by default and a full sweep is around eleven hours.\\n Press n to clear, then space to pick a few. piper alone takes about three minutes.\\n\\nBANNER\\n# No --only: this is the interactive picker, and a terminal block is a real PTY, so\\n# the arrow-key menu works exactly as it does on a laptop.\\npython3 tools/benchmark.py\\nrc=$?\\n\\n# No listening clips: make_samples.py exists to embed one per engine, but audio does\\n# not reach the reader through the browser block, and it is the only reason this\\n# needed ffmpeg -- which in turn was the only reason the sandbox ran apt-get, which\\n# reaches deb.nodesource.com through a leftover apt source and trips egress approval.\\n# Dropping playback drops all three.\\n\\nif [ -f tts-benchmark.html ]; then\\n # page_template.html declares utf-8 upstream, so this normally just copies. The\\n # fallback stays because the failure it guards is silent and total: python3 -m\\n # http.server sends \\\"text/html\\\" with no charset, and a page that also declares\\n # none is read as the browser's locale default, turning every em dash into\\n # mojibake. It writes to the published copy rather than the source, so the export\\n # zip stays byte-identical to what the tools wrote, and never doubles the tag.\\n if grep -qi ']*charset' tts-benchmark.html; then\\n cp -f tts-benchmark.html \\\"$SERVE/index.html\\\"\\n else\\n { printf '%s\\\\n' ''; cat tts-benchmark.html; } > \\\"$SERVE/index.html\\\"\\n fi\\n echo \\\"report published -> http://localhost:$PORT/\\\"\\nelse\\n echo \\\"no tts-benchmark.html was produced; measurements (if any) are under $BENCH/tools/\\\"\\nfi\\necho \\\"[tts benchmark done rc=$rc]\\\"\\nexec bash\", \"subagentIds\": [], \"skillIds\": [], \"mcpToolIds\": []}", "position": { "x": 0, "y": 0 From 5850500694eb4156d62971223d83ad64d71eb15c Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Sat, 22 Aug 2026 11:02:39 +0100 Subject: [PATCH 12/34] Frame every block once when a dashboard's items first render Nothing ever framed them. ReactFlow's fitView is off and cannot simply be turned on: it fits at mount, when the items are still in flight with the query, so it would frame an empty canvas and never run again. Blocks the reader creates are placed relative to the current viewport by findAvailableSpace and are visible by construction, which is why the gap went unnoticed. It shows on a dashboard whose coordinates came from somewhere else. Every hand-written template ships viewport {x:0,y:0,zoom:1} -- a typed placeholder rather than a computed frame -- so a block at the origin renders under the header and behind the workspace sidebar. All five benchmark templates carry it. The three templates exported from a real dashboard carry genuine viewports (zoom 0.69, 0.79, 0.69) and always looked fine, which is why this survived. The maths is extracted to placement.ts rather than written inline, following findAvailableSpace: framing is pure geometry, and geometry is worth testing without React in the way. Which paid immediately -- the first implementation had the sidebar correction backwards. setCenter puts the given point at the container midpoint, so screen x is (p - centerX) * zoom + width / 2 and pushing content clear of the sidebar needs a SMALLER centre, not a larger one. Adding the inset buries the content further underneath it. Two tests caught it, one of which I had written to assert the wrong direction, so the projection assertions matter more than the direction ones: they check where a block actually lands rather than which way a number moved. Zoom is capped at 1. Two blocks magnified to fill a wide screen reads as broken, and zooming in is the one direction a reader can trivially do themselves. Note for follow-up: ensureVisible carries the same "+ offsetX" correction and looks wrong the same way, centring roughly half a sidebar too far left when it rescues an off-screen block. Left alone here rather than changing two behaviours in one go. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- .../src/app/(app)/dashboards/[id]/page.tsx | 59 +++++++++++- frontend/src/lib/canvas/placement.test.ts | 95 ++++++++++++++++++- frontend/src/lib/canvas/placement.ts | 58 +++++++++++ 3 files changed, 208 insertions(+), 4 deletions(-) diff --git a/frontend/src/app/(app)/dashboards/[id]/page.tsx b/frontend/src/app/(app)/dashboards/[id]/page.tsx index 709a1cce..d22599ce 100644 --- a/frontend/src/app/(app)/dashboards/[id]/page.tsx +++ b/frontend/src/app/(app)/dashboards/[id]/page.tsx @@ -3,8 +3,8 @@ "use client"; -// REVISION: dashboard-v58-cancel-pending-edge -console.log(`[dashboard] REVISION: dashboard-v58-cancel-pending-edge loaded at ${new Date().toISOString()}`); +// REVISION: dashboard-v59-frame-on-load +console.log(`[dashboard] REVISION: dashboard-v59-frame-on-load loaded at ${new Date().toISOString()}`); import * as React from "react"; import { useParams, useRouter } from "next/navigation"; @@ -81,7 +81,7 @@ import { Canvas } from "@/components/canvas"; import { MobileFullscreenCanvas } from "@/components/mobile/MobileFullscreenCanvas"; import { FullscreenControlBar } from "@/components/mobile/FullscreenControlBar"; import { isCoarsePointerDevice } from "@/lib/openAuthPopup"; -import { findAvailableSpace } from "@/lib/canvas/placement"; +import { findAvailableSpace, frameToFit } from "@/lib/canvas/placement"; import { DesktopVersionBadge } from "@/components/DesktopVersionBadge"; import { CursorOverlay, PresenceList } from "@/components/multiplayer"; import { useAuthStore } from "@/stores/auth-store"; @@ -822,6 +822,59 @@ export default function DashboardPage() { tryApply(); }, [data, dashboardId]); + // Frame every block once, the first time a dashboard's items render. + // + // ReactFlow's own fitView is off (see fitViewEnabled below) and cannot simply be + // turned on: it fits at mount, when the items are still in flight with the query, + // so it would frame an empty canvas and never run again. Blocks the reader creates + // are placed relative to the current viewport by findAvailableSpace and are visible + // by construction, which is why nothing framed them and nobody noticed. + // + // It shows up on a dashboard whose coordinates came from somewhere else. Every + // hand-written template ships viewport {x:0,y:0,zoom:1} - a typed placeholder + // rather than a computed frame - so a block at the origin renders underneath the + // header and behind the workspace sidebar. The templates exported from a real + // dashboard carry a genuine viewport and looked fine, which is why this survived. + // + // The sidebar overlays the canvas rather than shrinking it, so ReactFlow's padding + // cannot know about it; the centre is shifted right by half its width, the same + // correction ensureVisible makes. + const didInitialFrame = React.useRef(false); + React.useEffect(() => { + if (didInitialFrame.current || !data || items.length === 0) return; + let cancelled = false; + const tryFrame = () => { + if (cancelled || didInitialFrame.current) return; + const instance = reactFlowInstanceRef.current; + const container = canvasContainerRef.current; + // Same rAF retry as the template viewport above: the instance is not ready on + // the first render after the data lands. + if (!instance || !container) { + requestAnimationFrame(tryFrame); + return; + } + const frame = frameToFit( + items, + container.clientWidth, + container.clientHeight, + getSidebarWidth(sidebarCollapsed), + ); + // null means the container has not been measured yet; try again next frame + // rather than locking in a zoom computed from nothing. + if (!frame) { + requestAnimationFrame(tryFrame); + return; + } + didInitialFrame.current = true; + instance.setCenter(frame.centerX, frame.centerY, { zoom: frame.zoom, duration: 0 }); + viewportRef.current = instance.getViewport(); + }; + tryFrame(); + return () => { + cancelled = true; + }; + }, [data, items, sidebarCollapsed]); + // Workspace sidebar session: pick the first active session from any terminal const workspaceSessionId = React.useMemo(() => { // Only browse the workspace through a LIVE session — a stopped session's sandbox diff --git a/frontend/src/lib/canvas/placement.test.ts b/frontend/src/lib/canvas/placement.test.ts index 369cb0e7..e5fe9269 100644 --- a/frontend/src/lib/canvas/placement.test.ts +++ b/frontend/src/lib/canvas/placement.test.ts @@ -1,5 +1,5 @@ import { describe, it, expect } from "vitest"; -import { findAvailableSpace, PLACEMENT_GAP, type PlacedRect } from "./placement"; +import { findAvailableSpace, frameToFit, PLACEMENT_GAP, type PlacedRect } from "./placement"; // A 1200x800 container at zoom 1 with no sidebar => viewport is flow (0,0)-(1200,800). const VIEW = { x: 0, y: 0, zoom: 1 }; @@ -117,3 +117,96 @@ describe("SlopCodeBench dashboard sequence", () => { } }); }); + +describe("frameToFit", () => { + const rect = (x: number, y: number, width = 100, height = 100): PlacedRect => ({ + position: { x, y }, + size: { width, height }, + }); + /** Where a block lands on screen, given the framing. Mirrors ReactFlow's + * setCenter: the centre is placed at the middle of the container. */ + const project = ( + r: PlacedRect, + f: { centerX: number; centerY: number; zoom: number }, + cw: number, + ch: number, + ) => ({ + left: (r.position.x - f.centerX) * f.zoom + cw / 2, + top: (r.position.y - f.centerY) * f.zoom + ch / 2, + right: (r.position.x + r.size.width - f.centerX) * f.zoom + cw / 2, + bottom: (r.position.y + r.size.height - f.centerY) * f.zoom + ch / 2, + }); + + it("returns null when there is nothing to frame", () => { + expect(frameToFit([], 1440, 900, 200)).toBeNull(); + }); + + it("returns null before the container has been measured", () => { + // A zero-height container means the layout has not settled; framing against it + // would compute a zoom from nothing and lock it in. + expect(frameToFit([rect(0, 0)], 1440, 0, 200)).toBeNull(); + expect(frameToFit([rect(0, 0)], 150, 900, 200)).toBeNull(); + }); + + it("brings a block at the origin fully into view", () => { + // The case every hand-written template hits: viewport {0,0,zoom:1} leaves a + // block at the origin under the header and behind the sidebar. + const items = [rect(0, 0, 560, 560), rect(600, 0, 900, 620)]; + const cw = 1440, ch = 900, inset = 200; + const f = frameToFit(items, cw, ch, inset)!; + for (const r of items) { + const p = project(r, f, cw, ch); + expect(p.left).toBeGreaterThanOrEqual(inset); + expect(p.top).toBeGreaterThanOrEqual(0); + expect(p.right).toBeLessThanOrEqual(cw); + expect(p.bottom).toBeLessThanOrEqual(ch); + } + }); + + it("clears the sidebar rather than centring under it", () => { + const items = [rect(0, 0, 560, 560), rect(600, 0, 900, 620)]; + const wide = frameToFit(items, 1440, 900, 0)!; + const inset = frameToFit(items, 1440, 900, 200)!; + // A wider occlusion moves the centre LEFT in flow coordinates, which is what + // pushes the content right on screen, clear of the sidebar. + expect(inset.centerX).toBeLessThan(wide.centerX); + }); + + it("never magnifies a small dashboard", () => { + // One little block on a big screen should sit at 1:1, not be blown up. + const f = frameToFit([rect(0, 0, 200, 150)], 1920, 1080, 200)!; + expect(f.zoom).toBe(1); + }); + + it("zooms out far enough for a sprawling canvas", () => { + const items = [rect(0, 0, 400, 300), rect(4000, 2500, 400, 300)]; + const cw = 1440, ch = 900, inset = 200; + const f = frameToFit(items, cw, ch, inset)!; + expect(f.zoom).toBeLessThan(1); + for (const r of items) { + const p = project(r, f, cw, ch); + expect(p.left).toBeGreaterThanOrEqual(inset - 1); + expect(p.right).toBeLessThanOrEqual(cw + 1); + expect(p.top).toBeGreaterThanOrEqual(-1); + expect(p.bottom).toBeLessThanOrEqual(ch + 1); + } + }); + + it("handles negative coordinates", () => { + // Exported templates carry blocks at negative x and y; nothing may assume the + // canvas starts at the origin. + const items = [rect(-64, -208, 400, 300), rect(1200, 240, 400, 300)]; + const cw = 1440, ch = 900, inset = 200; + const f = frameToFit(items, cw, ch, inset)!; + for (const r of items) { + const p = project(r, f, cw, ch); + expect(p.left).toBeGreaterThanOrEqual(inset - 1); + expect(p.right).toBeLessThanOrEqual(cw + 1); + } + }); + + it("never returns a degenerate zoom", () => { + const f = frameToFit([rect(0, 0, 500000, 500000)], 1440, 900, 200)!; + expect(f.zoom).toBeGreaterThanOrEqual(0.1); + }); +}); diff --git a/frontend/src/lib/canvas/placement.ts b/frontend/src/lib/canvas/placement.ts index 1460e310..96cf2428 100644 --- a/frontend/src/lib/canvas/placement.ts +++ b/frontend/src/lib/canvas/placement.ts @@ -134,3 +134,61 @@ export function findAvailableSpace( } return { x: snap(bestPos.x), y: snap(bestPos.y) }; } + +/** + * Frame a set of blocks so every one of them is visible. + * + * Returns the centre and zoom to hand to ReactFlow's `setCenter`, or null when + * there is nothing to frame or the container has no usable area yet. + * + * Two things this has to get right that a plain `fitView` does not: + * + * - The workspace sidebar **overlays** the canvas rather than shrinking it, so + * ReactFlow's own padding cannot account for it. The usable width is reduced by + * the inset, and the centre moves LEFT by half of it in flow coordinates, which + * is what pushes the content right on screen: `setCenter` puts the given point at + * the container midpoint, so screen x is `(p - centerX) * zoom + width / 2` and a + * smaller centre moves everything right. Adding the inset instead buries the + * content further under the sidebar - which is what the tests below caught. + * - It never magnifies. Zoom is capped at 1, because two blocks blown up to fill a + * wide screen reads as broken, and zooming in is the one direction a reader can + * trivially do for themselves. + * + * @param sidebarInset Pixels of the left edge occluded by the workspace sidebar. + */ +export function frameToFit( + items: PlacedRect[], + containerWidth: number, + containerHeight: number, + sidebarInset: number, + margin = 48, +): { centerX: number; centerY: number; zoom: number } | null { + if (items.length === 0) return null; + const usableWidth = containerWidth - sidebarInset; + if (usableWidth <= 0 || containerHeight <= 0) return null; + + let minX = Infinity; + let minY = Infinity; + let maxX = -Infinity; + let maxY = -Infinity; + for (const item of items) { + minX = Math.min(minX, item.position.x); + minY = Math.min(minY, item.position.y); + maxX = Math.max(maxX, item.position.x + item.size.width); + maxY = Math.max(maxY, item.position.y + item.size.height); + } + if (!Number.isFinite(minX)) return null; + + const boundsWidth = maxX - minX + margin * 2; + const boundsHeight = maxY - minY + margin * 2; + const zoom = Math.max( + 0.1, + Math.min(1, usableWidth / boundsWidth, containerHeight / boundsHeight), + ); + + return { + centerX: (minX + maxX) / 2 - sidebarInset / 2 / zoom, + centerY: (minY + maxY) / 2, + zoom, + }; +} From 2ba073c01e4dfc3daa30cf53e3c2789e115c225f Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Sat, 22 Aug 2026 11:16:56 +0100 Subject: [PATCH 13/34] Ship no setupGuide, so chat stays minimised and silent setupGuide is not documentation. It is a walkthrough brief for Orcabot chat, and carrying one has two effects that together explain the duplicate blocks. The dashboards page sets orcabot_initial_prompt to "Walk me through setting up this dashboard." whenever a template has a guide, and ChatPanel auto-sends it on mount with the panel expanded -- which is why that sentence appeared as a user message nobody typed. And the control plane injects the guide into every chat message on that dashboard, wrapped in a system prompt reading "Use your tools to do the work (... create_terminal/create_browser as needed)". So the platform instructs chat to create blocks whenever a guide exists. The previous fix put "Do NOT call create_terminal" inside the guide, which argued with the instruction wrapped around it and treated the symptom: it stopped chat building a second pair, but left it opening, speaking, and being told to build. There is nothing here to walk through. No keys, no secrets, no provider, no config file, and a terminal that starts itself. Dropping the field removes the auto-prompt, the walkthrough system prompt, and the reason chat was invoked at all. Verified end to end: setup_guide inserts NULL and dashboard creation returns hasSetupGuide false, so no initial prompt is stashed. What the guide explained now lives only in this directory's README, which is read by people rather than injected into a model. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- benchmarks/talkito-tts/README.md | 21 +++++++++++++++------ benchmarks/talkito-tts/template.json | 3 +-- 2 files changed, 16 insertions(+), 8 deletions(-) diff --git a/benchmarks/talkito-tts/README.md b/benchmarks/talkito-tts/README.md index 8c32d52e..b7f85aa4 100644 --- a/benchmarks/talkito-tts/README.md +++ b/benchmarks/talkito-tts/README.md @@ -22,12 +22,21 @@ parts are a terminal and a browser block. That also makes it the cheapest template here to run: no LLM calls, no API key, no tokens. It costs CPU time and nothing else. -**The setup guide's job is to stop Orcabot chat from helping.** A `setupGuide` is a -prompt addressed to chat, not documentation for a human. An earlier version of this -one described the dashboard instead of instructing chat, so chat obligingly called -`create_terminal` and `create_browser` — and the reader got two of each, with a second -benchmark writing the same measurement files as the first. The guide now opens by -saying the dashboard is already complete and forbids both calls by name. +**It ships no `setupGuide`, deliberately.** That field is not documentation — it is a +walkthrough brief for Orcabot chat, and its presence has two effects. The dashboards +page sets `orcabot_initial_prompt` to "Walk me through setting up this dashboard.", +which `ChatPanel` auto-sends on mount with the panel expanded. And the control plane +injects the guide into *every* chat message on that dashboard, under a system prompt +reading "Use your tools to do the work (… `create_terminal`/`create_browser` as +needed)". + +An earlier version carried a guide that described the dashboard, and chat did as the +wrapper told it: two terminals and two browsers, with a second benchmark writing the +same measurement files as the first. Replacing the text with "do not create anything" +only argued with the instruction wrapped around it. Removing the field removes the +auto-prompt, the walkthrough system prompt, and the reason chat was invoked at all — +so it stays minimised and silent, which is right for a dashboard with no keys, no +provider choice, no config, and a terminal that starts itself. ## Architecture diff --git a/benchmarks/talkito-tts/template.json b/benchmarks/talkito-tts/template.json index 5ab6edd7..04ea97ea 100644 --- a/benchmarks/talkito-tts/template.json +++ b/benchmarks/talkito-tts/template.json @@ -36,6 +36,5 @@ "x": 0, "y": 0, "zoom": 1 - }, - "setupGuide": "# TTS benchmark (you are Orcabot chat)\n\n**This dashboard is already complete. Do not set anything up.**\n\nThe template provisions both blocks and the terminal's boot command does the entire\njob by itself: it clones the benchmark, installs what it needs, starts a local web\nserver, opens the engine picker, and publishes the report to that server when the run\nfinishes.\n\n## Do not create anything\n\n**Do NOT call create_terminal. Do NOT call create_browser.** There is already one of\neach on the canvas. Creating more gives the reader two terminals and two browsers,\nwith a second benchmark running against the same workspace as the first -- both\nwriting the same measurement files, which corrupts the run rather than doubling it.\n\nThere are no keys to set, no secrets to add, no provider to choose, and no config file\nto write. This benchmark makes no LLM calls and needs no API key at all.\n\nIf the reader says \"walk me through setting up this dashboard\", the correct answer is\nthat there is nothing to set up: the terminal is already doing it, and the picker is\nwaiting for them.\n\n## What the reader is looking at\n\nThe terminal opens a checklist of every engine the benchmark knows. Arrows move, space\ntoggles, `n` clears the selection, `a` selects all, Enter starts, `q` quits. Everything\nis ticked by default and the picker shows what each engine cost on previous runs.\n\nTell them to press `n` first unless they mean it: a full sweep is around eleven hours,\nand longer on a first run because each engine downloads its own weights. piper alone is\nabout three minutes and is the cheapest complete arm in the table.\n\nThe browser block shows a placeholder that reloads every ten seconds, and the report\nreplaces it when the run finishes.\n\n## Answering \"how is it going?\"\n\nRead the terminal. Each engine prints a line per category as it completes, then a\nsummary line of the form `piper 82/84 phrases WER 5.0% RTF 0.07x`. The run ends with\n`[tts benchmark done rc=N]`; a non-zero `rc` means at least one engine failed and the\nper-engine output above says which.\n\nMeasurements are saved per unit as they complete, so an interrupted run keeps its\nfinished work and starting again resumes rather than restarting.\n\n## If they want to re-run or change the selection\n\nThey type `python3 tools/benchmark.py` in the same terminal -- it drops to a shell in\nthe right directory when the run ends. `--keep` adds to the existing measurements\ninstead of replacing the ones for the engines they pick. The page and the export can\nalways be rebuilt from stored measurements alone:\n\n python3 tools/build_benchmark_page.py tts-benchmark.html\n" + } } From a2cfcfaea1f224a7d0d8969eb25e297133d3a50c Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Sat, 22 Aug 2026 17:46:44 +0100 Subject: [PATCH 14/34] Tighten the run-it-yourself note, and redate the article Drops the "no stored results in it" aside, which restated the sentence before it. Says the report appears in the built-in browser rather than "next to it", which is what the template actually does and is the more useful thing to know. Removes both em dashes from the prose. The one left in the methodology section became a comma, which the clause took without complaint. The forty em dashes still on the rendered page are not prose: they are the placeholder a Human cell shows when a configuration has fewer than three ballots, from TtsResultsTable, and a dash for "no value" in a table is doing a different job from a dash in a sentence. Left alone. Dated 22 August, since the table has moved twice since the 17th -- Orpheus Q8 landed and four notes were corrected. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- frontend/content/benchmarks/open-weight-tts.md | 8 ++++---- frontend/src/data/benchmarks-posts.json | 4 ++-- 2 files changed, 6 insertions(+), 6 deletions(-) diff --git a/frontend/content/benchmarks/open-weight-tts.md b/frontend/content/benchmarks/open-weight-tts.md index 7e077941..5b91dc66 100644 --- a/frontend/content/benchmarks/open-weight-tts.md +++ b/frontend/content/benchmarks/open-weight-tts.md @@ -1,6 +1,6 @@ --- title: Testing open-weight Text To Speech models -date: 2026-08-17 +date: 2026-08-22 description: Every open-weight TTS engine that can run on a consumer laptop, tested for WER using Whisper. author: Rob Macrae ogImage: /benchmarks/og-open-weight-tts.png @@ -13,10 +13,10 @@ Every row can be played, so you can hear what a word error rate actually sounds **You can run this yourself.** The harness, the corpus and the scoring are open, and [talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this -page from nothing — there are no stored results in it. A single engine takes minutes +page from nothing. A single engine takes minutes (`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it spent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template -does the whole thing in one terminal and shows the report next to it. +does the whole thing in one terminal and shows the report in the built-in browser. ```chart tts-vendors @@ -51,7 +51,7 @@ face value. `medium.en` was run too, but it did not finish for twenty-two of the thirty-seven configurations, so `base.en` is the one recogniser that covers the whole table. Being the weaker model it hallucinates words onto trailing silence, which understates good engines - more than bad ones — so the true spread between engines is **wider** than this column + more than bad ones, so the true spread between engines is **wider** than this column shows, not narrower. - **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on the sample in the first column. It is a second axis word error cannot see, because word diff --git a/frontend/src/data/benchmarks-posts.json b/frontend/src/data/benchmarks-posts.json index 813e7458..9da7ee34 100644 --- a/frontend/src/data/benchmarks-posts.json +++ b/frontend/src/data/benchmarks-posts.json @@ -2,7 +2,7 @@ { "slug": "open-weight-tts", "title": "Testing open-weight Text To Speech models", - "date": "2026-08-17", + "date": "2026-08-22", "description": "Every open-weight TTS engine that can run on a consumer laptop, tested for WER using Whisper.", "author": "Rob Macrae", "coverImage": null, @@ -31,7 +31,7 @@ "slug": "limitations" } ], - "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nEvery row can be played, so you can hear what a word error rate actually sounds like.\n\n**You can run this yourself.** The harness, the corpus and the scoring are open, and\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this\npage from nothing — there are no stored results in it. A single engine takes minutes\n(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it\nspent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template\ndoes the whole thing in one terminal and shows the report next to it.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for twenty-two of the thirty-seven\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones — so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the thirteen\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2x for readers who only care about what can be spoken\nlive, and sorting by RTF or compute per phrase draws a line at the cutoff so the two groups\nare visible at once.\n\nThe line falls at 2x exactly. Nothing sits awkwardly against it: the slowest engine that\nkeeps up is OmniVoice at 1.51x, and the fastest that does not is Chatterbox Q4 at 2.03x.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes, feed-forward over one.\nDifferences under roughly two points are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" + "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nEvery row can be played, so you can hear what a word error rate actually sounds like.\n\n**You can run this yourself.** The harness, the corpus and the scoring are open, and\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this\npage from nothing. A single engine takes minutes\n(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it\nspent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template\ndoes the whole thing in one terminal and shows the report in the built-in browser.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for twenty-two of the thirty-seven\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones, so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the thirteen\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2x for readers who only care about what can be spoken\nlive, and sorting by RTF or compute per phrase draws a line at the cutoff so the two groups\nare visible at once.\n\nThe line falls at 2x exactly. Nothing sits awkwardly against it: the slowest engine that\nkeeps up is OmniVoice at 1.51x, and the fastest that does not is Chatterbox Q4 at 2.03x.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes, feed-forward over one.\nDifferences under roughly two points are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" }, { "slug": "agent-skills", From 982f5795c9ebff79f5861b38c0e4058abbcc9299 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Sat, 22 Aug 2026 18:19:50 +0100 Subject: [PATCH 15/34] Add a Cloudflare Web Analytics beacon for readers who never sign in POST /analytics/events calls requireAuth and stamps every row with a user_id, so an anonymous reader on a blog, docs or benchmark page produces no record at all. /admin/metrics is built from users, sessions and those rows -- DAU/WAU/MAU, signups, retention -- so it counts accounts, not visits, and a link shared publicly is invisible to it. The only signal a shared page currently leaves is second-order: whoever signed up afterwards. Cloudflare Web Analytics fills exactly that gap and nothing else. Cookieless, no cross-site identifier, no fingerprinting, so it needs no consent banner, which is what makes it the right choice for pages read by people who have not agreed to anything. Two gates. Without NEXT_PUBLIC_CF_BEACON_TOKEN it renders nothing, so it stays inert until a token is set and cannot leak from a local build. And it never renders under NEXT_PUBLIC_DESKTOP_MODE: that build is served from localhost inside Tauri, where beaconing a desktop user's navigation to a third party is not something a local app should do, and would fail offline anyway. Both paths verified against the served HTML rather than reasoned about: with a token the script tag is present with the JSON attribute intact, and with none the string cloudflareinsights does not appear on the page at all. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- frontend/src/app/layout.tsx | 4 +++ frontend/src/components/WebAnalytics.tsx | 38 ++++++++++++++++++++++++ 2 files changed, 42 insertions(+) create mode 100644 frontend/src/components/WebAnalytics.tsx diff --git a/frontend/src/app/layout.tsx b/frontend/src/app/layout.tsx index c55937b7..b5f08fc7 100644 --- a/frontend/src/app/layout.tsx +++ b/frontend/src/app/layout.tsx @@ -6,6 +6,7 @@ import { Inter, JetBrains_Mono } from "next/font/google"; import "./globals.css"; import { Providers } from "./providers"; import { SplashTransitionOverlay } from "@/components/SplashTransitionOverlay"; +import { WebAnalytics } from "@/components/WebAnalytics"; const inter = Inter({ variable: "--font-inter", @@ -102,6 +103,9 @@ export default function RootLayout({ `, }} /> + {/* Page views for readers who never sign in; the app's own analytics + only see authenticated users. No-ops without a beacon token. */} + + ); +} From 291ad05528f6ce05430733fa56ae52f793a0c20f Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Sun, 23 Aug 2026 12:09:31 +0100 Subject: [PATCH 16/34] Put the preference chart on a log x-axis, like the chart above it Compute per phrase spans two and a half decades, 0.12s to 51s, and the axis was linear and pinned at 6.8s -- roughly where the engines that keep up with real time end. Everything slower was drawn outside the plot: ten of the thirty rated engines rendered off-canvas at up to cx=5278 in a 780-wide viewBox, while the accessible label went on claiming thirty. The word-error chart already had both halves of this right, including the reason its bound derives from the data: "they were pinned to the range of one export, and the next one moved a configuration past the right-hand end -- Dots-TTS stopped being a partial run, finished the whole corpus, and went from 41.7s to 51.0s, which drew its marker outside the plot with nothing to say so." The same lesson, unlearned one chart lower. This inherits both, so a future export cannot push a marker off the end. The scale moves into the component because the plotted set depends on which engines have ratings, which is fetched. Label side and tooltip side now test screen position rather than a fraction of the maximum: on a log axis 0.75 of the range sits far right of three quarters across, so the value-based test they used flipped almost nothing. Verified against the rendered SVG with the live scores mocked in: 30 markers, 0 beyond the plot edge, max cx 736 of 780, ticks 0.1s/0.3s/1s/3s/10s/30s matching the sibling chart's decade and half-decade convention. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- .../components/charts/TtsPreferenceChart.tsx | 45 +++++++++++++++---- 1 file changed, 37 insertions(+), 8 deletions(-) diff --git a/frontend/src/components/charts/TtsPreferenceChart.tsx b/frontend/src/components/charts/TtsPreferenceChart.tsx index 0e0b68d6..6359e830 100644 --- a/frontend/src/components/charts/TtsPreferenceChart.tsx +++ b/frontend/src/components/charts/TtsPreferenceChart.tsx @@ -3,7 +3,7 @@ "use client"; -// REVISION: tts-preference-chart-v2-human-axis +// REVISION: tts-preference-chart-v3-log-x // What compute actually buys, plotted against what it costs: reader preference // on the y-axis, compute per phrase on the x. // @@ -40,7 +40,7 @@ import run from "@/data/benchmarks/open-weight-tts/2026-08.json"; import { useClassFilter } from "./useClassFilter"; import { useTtsScores } from "./useTtsScores"; -const MODULE_REVISION = "tts-preference-chart-v2-human-axis"; +const MODULE_REVISION = "tts-preference-chart-v3-log-x"; if (typeof window !== "undefined") { console.log(`[tts-preference-chart] REVISION: ${MODULE_REVISION} loaded at ${new Date().toISOString()}`); } @@ -96,16 +96,32 @@ const M = { top: 18, right: 26, bottom: 56, left: 58 }; const PW = W - M.left - M.right; const PH = H - M.top - M.bottom; -const X_MAX = 6.8; // Chatterbox Q4 at 6.39 is the slowest that survives the filter const Y_MAX = 100; // ratings are already scaled 0-100 server-side -const sx = (v: number) => M.left + (v / X_MAX) * PW; const sy = (v: number) => M.top + PH - (v / Y_MAX) * PH; +/** Compute per phrase spans two and a half decades - 0.12s to 51s - so the x-axis + * is logarithmic, the same as the word-error chart above it. A linear axis here + * was not merely cramped: it was pinned at 6.8s, which is roughly where the + * engines that keep up with real time end, and every slower engine was drawn + * outside the plot. Ten of thirty rated engines rendered off-canvas, silently, + * while the accessible label still counted them. + * + * The bound derives from the data for the same reason the sibling chart's does: + * a fixed one goes stale the moment an export moves a configuration past it, and + * the failure is a marker that simply is not there. */ +const X_MIN = 0.1; +const xScale = (xMax: number) => { + const lo = Math.log10(X_MIN); + const span = Math.log10(xMax) - lo; + return (v: number) => M.left + ((Math.log10(v) - lo) / span) * PW; +}; + const DISK = { min: Math.min(...ENGINES.map((e) => e.disk)), max: Math.max(...ENGINES.map((e) => e.disk)) }; /** Area-proportional, floored at 5px radius so the smallest is still a target. */ const radius = (disk: number) => 5 + 7 * Math.sqrt((disk - DISK.min) / (DISK.max - DISK.min)); -const X_TICKS = [0, 1, 2, 3, 4, 5, 6]; +// Decade and half-decade ticks, keeping only those the axis actually spans. +const TICK_CANDIDATES = [0.1, 0.3, 1, 3, 10, 30, 100, 300]; const Y_TICKS = [0, 25, 50, 75, 100]; interface Point extends Engine { rating: number } @@ -127,6 +143,16 @@ export function TtsPreferenceChart() { [ratings] ); + // Scaled to the engines actually plotted, so nothing can fall off the right-hand + // end. Computed before the early return below because hooks cannot be conditional. + const { sx, X_TICKS } = React.useMemo(() => { + const xMax = points.length ? Math.max(...points.map((p) => p.synth)) * 1.18 : 1; + return { + sx: xScale(xMax), + X_TICKS: TICK_CANDIDATES.filter((t) => t >= X_MIN && t <= xMax), + }; + }, [points]); + // Nothing to say yet, so say nothing. An empty axis box would read as a bug, // and a placeholder would be a promise the page cannot keep on its own. if (points.length < MIN_ENGINES_TO_PLOT) return null; @@ -205,7 +231,7 @@ export function TtsPreferenceChart() { ))} - Compute per phrase (seconds, Apple M2) + Compute per phrase, log scale (seconds, Apple M2) X_MAX * 0.75; + // Position, not value: on a log axis a fraction of the maximum sits far + // to the right of the same fraction across, so a value-based test flips + // almost nothing. + const right = sx(p.synth) > M.left + PW * 0.72; // Labels sit above the marker, except near the ceiling where that // puts them through the top border. The best-rated engine is by // definition up there, so this is the common case, not an edge one. @@ -256,7 +285,7 @@ export function TtsPreferenceChart() { position: "absolute", pointerEvents: "none", left: `${(sx(hover.synth) / W) * 100}%`, top: `${(sy(hover.rating) / H) * 100}%`, - transform: `translate(${hover.synth > X_MAX * 0.6 ? "-105%" : "12px"}, -110%)`, + transform: `translate(${sx(hover.synth) > M.left + PW * 0.62 ? "-105%" : "12px"}, -110%)`, background: "#0b1a2e", border: `1px solid ${hover.color}`, borderRadius: 6, padding: "0.45rem 0.6rem", fontSize: "0.76rem", color: INK.secondary, whiteSpace: "nowrap", boxShadow: "0 6px 18px rgba(0,0,0,0.45)", From 6eca9b03b8a01ab22a14d5332f89ebf1bff1ccfb Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Sun, 23 Aug 2026 12:16:51 +0100 Subject: [PATCH 17/34] Import the 43-row export: VoxCPM, IndexTTS and Pocket TTS Forty configurations now, up from thirty-seven. All three new rows clone from the piper sample, which is worth knowing when reading their PESQ: it is a no-reference estimate of signal quality, so an engine reproducing the highest-PESQ engine in the table is not being scored on the same thing as one synthesizing its own voice. IndexTTS lands second on PESQ behind piper. VoxCPM answers a question asked publicly. A commenter listed it among "most recent advances" the table misses; it was in fact already integrated and had never been measured, because without a reference voice it ignores the input text. Given one, it runs: 78/84 at 8% WER, Apache-2.0, 0.5B. Its PESQ is provisional over 42 of 84 clips and the note says so. Pocket TTS is Kyutai's, and its licence is the split the proof map exists for: MIT as code on GitHub, CC-BY-4.0 as weights on the card. The table states the weights licence, so the card is what the link evidences. Same for VoxCPM, where the bare openbmb/VoxCPM path 401s and openbmb/VoxCPM-0.5B is the card that matches the configuration measured. Every licence here was read off the card rather than taken from a search result. VoxCPM has no clip. The importer says so ("! no audio for voxcpm2-tts"), the table disables its button and labels it, and the ballot pool syncs to thirty-nine rather than forty, because a configuration nobody can play cannot be ranked. The second line of the article promised every row could be played, which was true until this import, so it now says almost every row and names the exception. Synthesizing that clip would let the stronger sentence come back. Counts follow from the new rows: twenty-five of forty lack a medium.en score, and fifteen rather than thirteen sit above two times real time. The cutoff sentence still holds unchanged. Nothing drifted. The sixteen cells that differ from the last import are all the normalisation layer working: REWRITE folding stoch-ff to st-ff and "under $5M" back to "<$5M", and CELL_OVERRIDES holding supertonic's release date and dots-tts's licence against an export that reverted both again. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- controlplane/src/tts/handler.ts | 14 +- .../content/benchmarks/open-weight-tts.md | 7 +- .../public/benchmarks/og-open-weight-tts.png | Bin 78514 -> 82561 bytes frontend/public/benchmarks/tts/indextts.mp3 | Bin 0 -> 19555 bytes frontend/public/benchmarks/tts/pocket-tts.mp3 | Bin 0 -> 16867 bytes frontend/scripts/import-tts-export.mjs | 13 + frontend/src/data/benchmarks-posts.json | 2 +- .../benchmarks/open-weight-tts/2026-08.json | 315 +++++++++++++++++- 8 files changed, 334 insertions(+), 17 deletions(-) create mode 100644 frontend/public/benchmarks/tts/indextts.mp3 create mode 100644 frontend/public/benchmarks/tts/pocket-tts.mp3 diff --git a/controlplane/src/tts/handler.ts b/controlplane/src/tts/handler.ts index abf483e3..4f8ac160 100644 --- a/controlplane/src/tts/handler.ts +++ b/controlplane/src/tts/handler.ts @@ -21,14 +21,14 @@ const RUN = '2026-08'; * reader can complete and the server is then obliged to reject. */ const CONFIGS = [ - 'supertonic', 'piper', 'chatterbox-turbo', 'kittentts-micro', + 'supertonic', 'piper', 'indextts', 'chatterbox-turbo', 'kittentts-micro', 'kittentts-nano-int8', 'qwen3-tts-vd', 'kittentts-mini', 'dots-tts', - 'qwen3-tts', 'kittentts-nano', 'chatterbox', 'f5-tts', 'kokoro', - 'orpheus-q8', 'xtts', 'orpheus-q4', 'nt-2e-q8-metal', 'nt-2e-fp32-mps', - 'chatterbox-q8', 'vibevoice', 'nt-2e-q4-metal', 'cosyvoice3-rl', - 'chatterbox-q4', 'styletts2', 'melotts', 'omnivoice', 'fastpitch', - 'cosyvoice3', 'bananamind-tts', 'tada-3b', 'zonos', 'csm', 'tada-1b', - 'speecht5', 'mms-tts', 'vibevoice-1.5b', 'bark', + 'qwen3-tts', 'kittentts-nano', 'pocket-tts', 'chatterbox', 'f5-tts', + 'kokoro', 'orpheus-q8', 'xtts', 'orpheus-q4', 'nt-2e-q8-metal', + 'nt-2e-fp32-mps', 'chatterbox-q8', 'vibevoice', 'nt-2e-q4-metal', + 'cosyvoice3-rl', 'chatterbox-q4', 'styletts2', 'melotts', 'omnivoice', + 'fastpitch', 'cosyvoice3', 'bananamind-tts', 'tada-3b', 'zonos', 'csm', + 'tada-1b', 'speecht5', 'mms-tts', 'vibevoice-1.5b', 'bark', ] as const; /** diff --git a/frontend/content/benchmarks/open-weight-tts.md b/frontend/content/benchmarks/open-weight-tts.md index 5b91dc66..f0f1250b 100644 --- a/frontend/content/benchmarks/open-weight-tts.md +++ b/frontend/content/benchmarks/open-weight-tts.md @@ -9,7 +9,8 @@ toc: false We tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text. -Every row can be played, so you can hear what a word error rate actually sounds like. +Almost every row can be played, so you can hear what a word error rate actually sounds +like. VoxCPM is the exception: it was measured but has no published clip yet. **You can run this yourself.** The harness, the corpus and the scoring are open, and [talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this @@ -48,7 +49,7 @@ face value. **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it compares directly and cannot be gamed by padding. - **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger - `medium.en` was run too, but it did not finish for twenty-two of the thirty-seven + `medium.en` was run too, but it did not finish for twenty-five of the forty configurations, so `base.en` is the one recogniser that covers the whole table. Being the weaker model it hallucinates words onto trailing silence, which understates good engines more than bad ones, so the true spread between engines is **wider** than this column @@ -72,7 +73,7 @@ face value. **Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family. -**Nothing is filtered out.** Every configuration compared is listed, including the thirteen +**Nothing is filtered out.** Every configuration compared is listed, including the fifteen that cannot keep up with their own speech, because "how far off is it" is a real question and a table that quietly omits the answer cannot be checked. The **Real time only** control above the table hides anything above 2x for readers who only care about what can be spoken diff --git a/frontend/public/benchmarks/og-open-weight-tts.png b/frontend/public/benchmarks/og-open-weight-tts.png index 4496b60d647dc81af10a2b106eaa3edd5caef8f1..cf3b38b352be92cba2288281296e69089388ffce 100644 GIT binary patch literal 82561 zcmeGDRa9J2vjqy{ZUKU82ol@^!7W&V1qkl$jk{}b2n5&Q?oQ+G?(Wh!G~V3Kcg}al z_}~7=`*6qDPpqbQ@4c#O&6=}jxRQc2CK?GE3=9mWtjt$s7#M_V7#KKL6gcQ#x;}u= zVPJ4zWWS23x@Ukg5e@JbNxL@t(|BA6zkSzcC}?hLYa8{44ML5Iwyowc96Rfb{&dc{ zw!WTg5#5;B+9XTz<~RIjm&96G%W4ylJF2U&0ra{a_S)PH_a?p5K3}$#DD$ox1lIY zh=t(%-;F{8(Ww9LMwp2I-;GcL|Nnk9XHu~tOaAC~ffEv3Thr0a z51APo1NE895~nUBh%pC^?k`}L>>liVsBeguLCfyzH}3z>w)*2f^3cir3GCf=KAy8% z`<1IJ&;@`6uuf_fG>ENs_Hk-dpi2+qvEJN^8U-V4t_WsW)kKb!AV6R}g@jSbF zs&&k)jP(^A9bMML@eU-WO0~0x&ENjg!!Hgqxpel!yH7D# zSYs4JWlaZc|3b4WSbZ~`=GT5sSarmYPQA<8?2bx2jn$*7Ey?4l%J<$~*FtXHJEu+O zID3CoRySj(kpi%6wc~3=8R~25>Szo4Cl|v;Sx7!$mCnuf<>M;W)HO6T_|BE(E1~(> zA{5B6^m}l`WDf57+-_(m3`Wm!dE=AI{)p~5?KR(7dPm6uIqBKA&JXS0y;)8j<4F3B za=HBX5&A*sZ`1Gx%Iuw-jM*F<9PA(LqVx`RAL#)`_ZUC990 zVlkPLSzaN2q>Uh|p)%U-^_p$WPWWvG0b>`;p62|>%5WNc93c|k^2v>M z*Bdavud?0uG{#Un;$JL(#)CdIl0ppDCW~%4eRswe7vIOUXt(qAHKqDs$r2@0+aM3dEe_NoO^F_VN6s-PslHlPU&|EqdENbh?}y8}I(+JS;g zULCK;_xsj79O>T*ICmg<9+cj$;GRE#PK`cybg_h&+q}oRLm!i2-C*(DvO1M-TKw+h z;imrA$H&{1(-t#aoZDuh3M&lOM7`e=kMX?h1AW31oUpZ zd$bl6nri$!0166y#I^JlgZ=cLbwa__)z$L&xB!J05lf@^!19>50Nn_~Hd({QWhD3A z%@ufLa+3G6-*69op!(1TG#4bsu7BH-w9|b-zaBZNnp$gVX*Dr1`(B`IB;>r(g=sgR z!OE+y?jCeZObjAX{lJzwmpr(zu&}YgDkPN9&W!~IPfWyUYVsL{_g&ZJWO+8X0p%`` z?i(83r!iuDH6Iw6mi6Q_>N82l5);u8{>iO1EhH!0J`Hg0@E2-*7aaUxSF zb9ol{9a)_)x!lHf^2h4tMH6Kbr-*K=<7=Bf#D7G@>7;Kv=ouMlE|3)`zYIB+qDrIf zacVlDFby6jQ5uWl9!17`Qo>H-`+!U@Y_r3?6Y|dU%}p}$$eri?{e8u( zN31pHmBbo6Y3#7M9)(mnisz5MH{f;Brt^;lJ57|y3zhrQe(GgEjSJhm3GgHWPhKAG zipCCxxeooFt!bZc7mH#jmx>7M9?zw-IVKpW*t`7JigV$gZV{%FT)V#I&j`3(0M=_R zF0T?iuZ#17dRzC|dLo3q$yzS!087ZC`oflLL({#;&+cBh6=|vuM-I1dNv~dz>(4_; zyD#-ZEw2309oZ&JMzt}0v3$Rk(9bE*0!ed z!cMlQopfYu8d=&r1%DB*o9hNrDbr;i`e9ap5)` z`jI@J+{xc6F`0rlqZgwr@uWF+;NxXX<<(|GBB6UCD{kkTXKc#VR%5od;92j7^YpWo zdJ~>a&C75dR{hq8-pb6?-h?bZKaUlVTlB`zWnXa7B-OpJ=S4+fbno#}mL?zHWrtej zGDd8daFbs)#T0}+UCDDBR>b{&95pYjmnMh|Z==-uQx-~a9g8(2kgegxd#OqLX7MeN zu>$XTZyCMw_3>||=6oTstxT>|C$C#lTbXO&`Q=sBW_a@?*;K3K?3Hn-i`8aQNhlRp zCeWVP>h#ZPlZ*of*WOk<9%j!8@Uykm{RH`OeL-R2W<>M--F@e=DINUY;}$G%JNzUB zfkV&Pf->jlX04AKC;`C;uIzUyGEo#yxdt5$N{SGmqiu>k#54MAC0g6Lnbv+K zp97OrG=Tu5KT7zNHwLmPropC^BLCt@7M0a!l=J6wxnVXXUP6o8)&#)z*X>~cFq_}& zs#EuwEftLW?CBLv#$#{)wsDzcIm^gkbETMlCS-3im<1o&Ya2xXxylP0cejPpfdzNb z=(KvjMgn|uy$t`J#tNssnI+I|S58b!d9JnkV5c>|x8|+U=F97ciFq@CWK5i`dLCo+ z*`Gf~s=sMAK)~zIppT!?s^0c@uv5pOhd-pw5vW(YnIhZ*sInV7p#Rq2vUOnBNOEGDQlN!=2 zz?Amon^;UOvb~Y#W17)(pURVjyg_znppOK6-YyEgYmr!0>x5j=T6_j&+7oaKMG~}g zo!-gb5O|mJGB7chv)-EAy2YVyvpulT$B<8u!F>3x@!I-bSaYgQoBq-~spnmMdir9q zBDPtbFA%5kcB;|E2^rqBX@N~4Mj#DU31#-?xQ zxS(m2h#K6ly`W7X^di7PTzI(=(R$DPElO#rT(-Ai)DQk*p-F_rIJ@jHe%81rWgArT z_LY8Q%+E_rDbxR+z*p=73J~m6y@$v!>r_mvr5PzxHT#)2si;VVFwo4ax{aeC2yj;^ zKZ3;0+ua{mp2k;mlW7a5nX*3!w2AmZNDyc|vAr-F8$tq*7)N9E{T3Qj1i%XJ7xE;Z z{U?{=C0d;=+T4%t4G5+Mu6o@z)A(%RykfIs@2&9$ZjYyJS3r+_%)TAMZ!b^es!Y&- zBw+BODpg4520!aV5Qr0B%Q`zL+v-J~VDIX+55bd@pFP-m@x;D9*lc@&JhI|?RR&@y zINZC&+=QDQ&*dGe!4c^;l_ zVpHZgOOSxKtF-9Icc&G}7#8kg`~H(z@X`c65EifVGQ&`W`4Y^Cq=enb&s;ek7pL=> zzkvqR$%6`+^KI_wYU6&-zNjkx#5q1N(@tZT-8n^>yuU`YBl1sR1L=-nrb4;A_}y&U z9y`}P!!Svguvg5@Ze07|pV0SsmhQTAdW4t8`wNU$7cNsd4R6i$Uz=?Y^^9OI^GAIl zB#g-$z7_0_dc zB;F_zc8vN42fTVV*iGB?oB{$Fh0Xd5ib=*;a_(}BgLrM{>?dAAqMD|vi9|bJE^03< z5&ij^++tZTr%5?J;)X9B27WMq$nOx=0~v!{IFX8R;#@qX#SQcN!(U#E6d=knaG{Ce zaz?gwHXg4mv2(RG`FKr?-EpWl&~3GT`sg}Zq8s?R7|T0nPvVES6GU974N0*7UY`}| zhQo{Agi40)GT^J-N~85~lSrrc1kE2}EHOqQA&pli9EXee$1Cw1d0n99SmJ7#m$?eE z-__MKs4+^uQ*XWF`5Lb`v+NBbS%qx?pGnCe2y{V`RFTX45?X;-d%Tu=cB(RfciN(% z(_x8EF|dI!iQ$E3f_0h3nLg#rgMn_!!=?`$#(30tnQo8W@JhGl&k@<%IIk$P5HSz; zKX$oU5m{cLA{|N+d;Nx0DhnU_(oad^wunK#L>RS@&*-%4im)1isx>tt;^*q-*6@OM zv*G!g=eB9JMCE)?A1>na%vTf2tCwxVAeS&hfE0Id|JtOWNIGZSHwCM`2+32g*vNSN zDbBP0UoIdmi)RQ%^5_sCT(_NGPv}2Sp!~N;peCS5;{mdINnLn6*YPMZ9eLm3H3rxbZA zb&wdtO5k;LGSr!$!z(a3{c|r*Y0>X|PiFH@pY}V6S-Q9{(bQfZx!hew1WUR)BnFq# z$yb{e1YoQ&1ZhzBm0niozMv-{96hwKY}n%pnnP$+M*J zk|H=@6{VoucDM8C`+(!2&xJ(xyveGGkVegcxwdFQeS7PiE%1EDAJ8D8f7>yw?f(VR zs~#z@yD%{UP6mjWTL;B19$M}CrgiGYI&YN%s_;Xdb4 z=a?XR;^exvdXVri=<51*smUiZi>Daby?uM6v0q{>xxzVE^$WBEg1rqD6pD&syh@ag zjT}$7y1C-;)M#5^>C*Y?$IA~Q{pioblFpGnWzYNyqH%pi9z;*ES+s?)@$=$r`M@EI!Pak>=%XwNY_L?Gps}FF?AQ zo;l#EK;@aI4*@WlCH1Yp2>oz&!nQeCqg=!qJ^!S?u>YwfVW2uWw&+toUdd;Y)?h+> zd{S+6c>YvG*a(=0H=6Ie@mQ@bG zE1Il}1uV7}==Cm@=yJSw*xderh|7-%E-fmW(0MSpQnvpM5^_&`Me4RfIwv$DS$F)L z-Bttg`KoSBg?u|O5VUD5{JjBQ`y+Qc2tFrfGu++&qT3j#?O!l3c2b+oGLr@^L_B2d z%bkj`n|>PK)G-;lLTT=i?oLfL9xTliGKZJ}*7bEr(wJU)qR0M!|5lLnrx%uIkhPBF z7rt-#Gm^hR)UTDAFZX1j@(;I^g@t4`@1r8I@jJVY(PfgZx}7bOe!wydnIz~SGT`)j zGHKNQ!bwN4wgp9dC>wOE@hEN?U(Q3wJI4xW@_V|e^M1mx#Tsp! zlDnr=3Q(`wgj=T=%6Wahd$8X~x4D@avrI@Sp_%fjZ6m|6+SIQPUiQAoO_}%$*zIc;d71+4Y=n@%w(5cDxF+~Gu2y<%ct-(=O=S3y1I+x!4;!G1n z(dyt1e}sfZE`Uh6)b98M<3-++6*LexGJfCr(sDhSy|nXAko?xwIShGSQB{sv16Cm$ z=zPBO9#}$>L;X=u9`2qBh-$>0$USOT_UF&G6KHRlMF6}8eD>e5fAF@gOsZrTl{}iM zVR0h3ZJ-(GRQNuo^e$(q4u~0qbPIv_`ORz=$&q1J{qY%WNN=>=XhhB;$q)V^fgE(B zr08l=$|6HOp2lsm>3rL#8N~{Ftl0sQE-d!v-Cet6>Lbc=k(L^KY49fhVsmqSy}qXN z%T$yJR(x*;@P>R6T(7iJC|L(3@{C5WpdAbXUc+>}jlGJ@No*&?=k?$$xbLGr;9^E7 zUX`%`{V`Ubg@Wfn5(qCA=CjFZIcss#*ICAsA?UQdy~{vPFFnLs9oY$4`Eo#{6gAim z9_W>pWZ_yuakbUz|Ed7D4HzG&kTMPywUl#z0`>NP2W+$;Bewl;Ct?t^k$snjwXi#8 zB%)yE3cg*wIqtqu5p*e|v#^aZO;ClYeeFwiiYj{7o}6R8(drb- zC#v{|-~Hj^bibORKyEmQ6rnA_S_p}!%=HjVBz=@E#a20bXK%^dCDH4sH42dpQSYm)!9;UdV z^R>b8Y`}JkHe?R<uB|+jh2#Yhf}9Tb2ij` zLYK3hxn0dYDckoMnS0M%c6iGxmERC_NzFEF6RV)l4XSvyh-s=fX#td3n;%sHf1*g_5nEo~tH<;h!ai)@0 z+TZRGD`9U5pupB0veXOW()?PX^Z|c%{`i$whDe7t{cW{ZILFZ4NBIwIfaT6@(=l;T zsHj5VwV>(C(Y)WAqtRPqF8uQe`Q>EkUH#^Jgzi7qyxQ+ST|JaI0VedFpFon$US?_p z_~FWgNQ#QQQ!J^I&*;hQ5IlC&s?~bkk+k^z-TmP)lCy;pcmdg|bk@fM!d5rrciV0H zNoPww{wW4SUBY8-&5%>44fEtOu0ysu`l&!Fz0NELE(FyiS+0#-M5Sgp-6dvNiZftr z$CA?)O0yU~+WzC{Vvfye3%MMoR7j8D?l>a!CA#Ai!+X3Zf6m9qmWk!GneBLdyua{$ z++7)8MKt2b)P)k1Cae8~z7WhjBUl4ZzvoxN-r=oUcnt@&rLKa)3131ItCe>o4sYAf zlx&8SI7)5BCgPEcMds>YMRbwpwA62y6_k6p0_O^42l7Y0x#hpV}wUIh9N(MA|n9RkUV=Xv3gKKj}3+ z3H%|Lqux}W)VfWsTB2_7I8)ed+P-ge*(o{o;=Ho5&?vxWy+s-@8NXY!obKzpF?KIs z_9oZ?C@Yy9_%oV9GpX-6je1OA#++u?cjTKxPbsWsQW&UzCMlVFxOCo!KG`s`vEEjj z04>zS%4A2^7i%8dR!2HI=N$PiSB7PzO=okYQ3&MlM0>wXZB}d4FIZ0Zt@0Qvad6jP z{!S;D*+KwKj}|Ue!>_mF=9`4s6%Lm};aEbU?VU%F6F#A@kb6cW%u#i~Tp4}=D@M+I zJIB4m0lZ?LF#2u5U_6_YTTs&XRikdCK3CZ8NXTB~#qIgSvAT#%lw}LFV+NO}gi%!z zPaGqQ1p}8`@~$4E1;Au5kNz6i#R%qXI%XHJ-QsLfD(Qz{V5XLt?l|w<@&VzuIF=pF3GwRWs&v{4iXwviqRy_6MBs&@aVWx2i)5Wo z7Q$5ym+1rlLVDjIfv>lRP*z5-wX5Cqk7`;SG@Hy@>IQ!W{+(z9oQtU9>+nKBuu#4G z!xp4Uk$7?exAWC|ARSMcBC|xN6T+iuptym{_>_7$P8q7pr>}2(T`q=IaM@2 zc1=_=WRc|^z`DR$``z@5gE7#jZc|F{m(?XX%tm$i#TnthnAD@iWyK%spZ1rpReQSJ zk@<&Ko1fvW<4}Ryv5<}$F)JWbSAwK*e>0PV$NXe z7_?CeWmz|#ubbKans(KZ4hR_=AXhd!JhO5qMt#Pb6Z9;`}5Ine5NHZ99A# zecIgC2uQ}&Wg#_c-_6&eYzpun%!dEU1^rO&|B#Fp0_EV%R0)CX*w_@pJc0)s@ZWb! zH-S|mYpm4U#9p(<1&0Z$NL{&C!@ny-z*}v z08Z}@LL;T}Uiicn^!UGbOp#_fO{3$drfj-bRt+=+qNf8IrVMIh53lND^+%hQc8Mx! z(8%~4R@ei*HEI7KdG|;wW>9>#!2vGk|A1H0%;o&Ob1qhFfgD5dDLBq+tNUo9fuS7n zVVJPv4UBv-5fBB$yV{xnA|JBqz^21}LSx#4a>{DnO-r2uQi~Sil-+h{>4>a2Py(h}5Re|M1xS#+q>;VAOO8E` zr2b>QS&QtR4+al?^dH56ZKPK)`E(Kib^p*gr$3tEd#cYunH4w-*qGGM-@edt*DwA0 zq$1QugO{gBDcxVq!tX4q04#~>gQH9gt7(ZeD(S1XH4U2oFBia~l2ppK0IxcJ+H~P; zv-;$5UGFR6#!EP`=JL1_4b?US+r`V`p=3E?NVVQN75G;Sx_Sh@c-2niN&pT(iml+v z-i>#5?>N<$l2|)|oFCVSn6Kq2fvjB1sl_T+l~FOXMO2X~WM-*e);+9|Mfmf@IRd=bhEFL*Nc`UVb_+_}W4;$>3+gBO zURiWLkSDAme5H(?*Pj9$gWo!_@|H4kEQR=YK?CKLDfvgbS&A9X_}xxg;`zu|>qT}> zPkBR|Pfl!;MzMJ0c+$){7zrcQZ(AHv)WMval`iO7=%&czK+jhKfrmcnY@a&AeE>*y z&fdsfgTWih>>>4>&XYD_lL=!%(6}64fD!a#J!Di!renq8M7?Oq;uz&T7zc|V+3xez#$pC^r|yI=+W=wYfh-Xw$g>{Z)nIC`?49F5JK%<6ZJE6Hv9_fWehMVNqDYq4m=E+`{@OvpH}*?K3=Y?K#!kct`*Fl3wf5Y;z3paPhymfp5u4lelh>m;?{wuVJ9G5dP5S9-H{`L!o$H=w0f&FH-MN z9xAl((<@zH_W1Io7ph2&oc-3f_(AfXe=cvMj9dr()NP@i5iZAf@4$i|3}csf*qt?t z!BY~-g2h4tm2&n5Fd0j=lJO{xZ-yT;Et4zPavJ&uwlPJe47GWei;B$gt;x*Wfn7({ z+B{P5fgDFHjaaFGFSxxu21F>+(Q?m^_kOL;HiKXB5KC%2>IN1DiW-8I2}Aj~3fuW^ zKAX{Nz}l_1xPg22=jx-{+h5jRUgrH?jhglJ)zoT~`0vC6qc%RwX~<~-^EblXyS}o1 zmW&A>5-KS41;(wq-AVsKoJTb%>(g9L+Nl4Z4O0rgK}TEaWu}+&8-4k7l1xDxpQZ+K z7%M*>EEN2W_TeCxhMqo}QWoGk!-u5JW8E`UvUMZk2&&)EG&uF}oBpks44GQ6rVTi(jk($U`ndD`S8lZd?R;v)tY2D9rotw;Wy+o^P<9wpvp zjoq9@=P4r0CxD6;mr~0&IJJ?FrfbsLkv~gV0HURjyUA6{URiy#8DE@)g{e7s*(h7n zHD~D>jpYZ-;NOFjP2V6-hY>MnTKC7x4VHOGp}U$^kB@dbchW_A60)>I9{GZzA=PN2 zRiHiw$sEiHA18iy898zh+nFzo&lKg?VKG_kp7}*ZZRus!W=XzlC_vM^;{)b`2i35b zz!xa&in~69O0-Fo7+U(S3amga*4seLUU6b<>Vy?2snG%D+@~!kY-exEYaU-s%Tg?)nviE(`X||{MkE%nVT!s7e6;=w4>G0 zyfr_k!ef{#rDnB&m&Hu@d*mW~q(MF?l1qyqal@L|u584}pq)6ybjRWrpNOQ|uZFNw zv&QVEE`RS@=A@%{kSGM_>u?v09~eR0!cra)Cv?K@Fe~Viy7ee;z0DQREiRQ{`ey_OSEDs3IAUeloMuTfQm~85xmCPX}QBP3se~YI02VlT<4b zfDIeS@y@x@Dng{TsOH@qq?{pM?MG$90Do$J`Ll&Jv^wsO2iNo`6})azj^6z(aXDLy zslXwL`4N}~H}H+0g6Ls;wcd)ywJ$Dfou2~<5r?@O=G}Wd?QOo;MC5;*YgZupx9MNG z@F_x2J;oRJ-nlo4JQVrzQhg1gTrMKi`06c`Y)g z=4`r#A3y%pKDAujkcidfLVfja-%R02O%$2x2(q3#sxJNp=gbbe^QMobyt^pH3=Iol zpX+EIzB#TlO1RE#X8c2EPNu1Fj1~lx>B|zl*)q(};z8;83f%b%FXW1C;(RH>Ziu?| z>0mrH_mCi+!!}T_+Lt6w0V+-!^$ms^-buQ;da9^&-3mC2p_oz4Mz6YIl6$iCUUt7O z05>HA^ci9UOTY%M0k07NOT(e;sXSC-mF3c#DUAP~op|Td!S0sT0pZQKz%rpZIS` zx3v<3dR8T@7EdXtGoq1Z(WfB@P(+ko)^5?Q1C9{rd~K=C_>J~{4@x%R!J!Faldl3b z0_@+mjlQcmTdXoRD^S(7KuKy(ZscNKC?~mD_jbA5+m>>&b$)`D91=;ofgcmw{Pv6Z z7Pew-C>MSi_A205Wk8;SH)6Tn@uxQJq{ObE9YUScQI~Jo0>E^VHuS?FMUjMas+1n} zWnhlhaY&5!#o01Q-ontLS92`k3m8Hd-LutDYirf;`kI!C;$!)1~;Z;xd~ zOe$(@M@92CtcID#(lP+T-#&fw%yEa-a4&Lk3yC8GRVvpkJ~vchs6^JP57m%iqkAN_ z{YT<%WKx6B4Cl92v)_(pO33S~zs&7Hr=2=YZwB5fH&A?FfOw4H+r@!c5}r6AVT5+3 zY(HZ=c@2J^-OzD?oF@NK@r0|#zecCg&>AhuR)XNelH!tTp$ZKzI1y5TL_uf1P2Wcj zyjjGFm1-AckA0nX@8};W^ikvrhe*x_ivaxaFI$!^9!!6O6!?h+tWan~x)TU~^^dts z=rkjUqt=1%R{?U!?{46%3wq5hDlhDKD(PMi+(nA@GFcpg8QEseK~Jr zwk=G>@rm!mg9!ELcQ@xun^&?mFCZxQmi8>UfC&S)gqxQwn`3(s6V(X4R_T(!k9+aR zg7qmzP6O+&HYV9^nNL_uKp*x9w#h`Gn_2p|vkeJp;=AGH;j76vA0eNKXPbWa^Nhu1 zc+IfrbcGRo0fAK!-$zeyJNB`mcqJ=qHfzRUF&Vd%`X-MgpU{t%M$2a`igbPljJn6p zi{z-3OMY$U^SO@ks!cXSw63O3W*Ro+kJw{=Lai<)xNoP}`WTcVK|gFfT~5aqkZ6pJ zj@hyI6X)Q=6vi@m0wBWP_I?MLfuBy6!s|dZx;hym+T7X87^qk;&d1nebcr7BDsua9 zW%`B3R|C-dnoh#~llsz|Xoyx37#KnHAs@by(oJM^_F=G4uWXMu}a?NaeFzq_k{?&wq^sPlylEb4&lvsF9qwDKuv#|$$kh|Dh z26&5d$o)DlRPjV^OQl_r)LDT%tVVX#?X#e2uqGsWQ0k&FE82P_q@__%sBwG2m0A^n zZSTqXM_$nMf;|TciST7UW_y20CMJJk;Pf0yDFGFJW&x(qDS8PoNleg~VnGL+g33W4 zNqz5|3xG(q8`jApsqhp=fw@U*OUn|mwkKrt;w|}sdHAKDaKg{)9JH9LU#`y^JWWi* zW9L(2bAI7hg}}r%>*|k#B>q^yw783#1r@Ymb-CHM`U9`u&Q{c^g&DwB!E(3lUf2|M zx@ZPSdGEzt-GPznm339{K8KW}i=qcp=~NwWucvN<3IW*{NIIp48b=Qty-Rq2Z=jnk z`PZkx2&gDH2r9wT@LIp#6tzrddSBlZwmy`^a&;9BBK~}-V|aHW92M2-@+_Rlsyw_@ zYkYsIGRQ@fo$Cc3cGCPqBAd2AFkBG_ErhBjAW!t7{5^CCPhwq^G@m9vfcbQPe;ROv zaE+T-$?l&Do|!*|LQ3Nat-E`+u-MF2x3?d_RxzNLv#FT~3EUl42B};<_QQ3rdbV$? zJtXtjGbU*;U%IY}$thT9>XmE9V9?bTI&2B6%Slw~0|Y|HP1ZWj&{eLW)YWLOKW50; z(niH}jl&iQQKcqp1O1szlWVN}e_)xrO3{QKXrV<2h*~-n6U`KpE$FpBz!E=H>2g!p zi*#>Xi)e!^;70j@ZT#Y`9N{XH!!ZEG{c*2@i;z`Xs(%i!0*8y~_Y#n}kl|)R|8O1e$YOy&GR3 z_o7!M>M!qtUEc{HKdoGE#+>kAQ;ZdR^{h#@T65ZBAWy(^o)CG{g{R{?D0(OHdaphuNhv8AbRk(a7{+6@T1j{{cp>E8ht(r! z*0*(@DIn-hwr{{@rPP^KQZQqAo1_f$d3Uf%1TmI@yYrSL_g|SF=iwlQG1j zXE7!J%8;7FniiMCjq}&A<#az*{bw-gY>JSsK*On(O}x=jou4L{tpp{Dj;>ppP4SkF zPp8A|D-qDb;7Yf}%&95nEOp+i>w?2XloA*qAt_zCY9q^>;F8|t28BLC$w&qW`0l0` zzPFDfBEBE?yUu=jMVBz`%zz$K|Hv4e7&;i8HM^t z9ZF-QA2;sp!a}W_BJ=22XceR{OH2HeNh+;9Fv_3vaBwrzZWqc+#^{!uO&vXKg|{Vx z{>udzCwP1~7g?!!?iw*@T zEti1TIdY_|CVWL-2C>So<6;W^x687h*UpdsDZ-RKt+!nhm{I$1B&)K!(zS!pvle+T zJ`XM1a{0g^);m}c1AY|w#2|JaRqYhK&ceiAgzJiTRv0_>p59@4xIJ5Y_`q2qkma83!^D0HHrz+*Y_u7;E(>hL` z301DtD>i2eRJ2`pF}_hPTF^*rFQlBRo+BQ~gL(S5oR zUSTYkw{3yb6DXFB@1oWX>-s{Rya#}f^&q^RyBb4H5gdc ztoessn?YZH8jXU5c8kVqOdQy82#PwcDyJ>gZP50wJt%##)F|^h17ThV+Oh9kAs4ax zQcpM)*C? zD*{nqhPk$V%DC?iwQcUR_w+n?iS1zM^!P+|PQ;q4o!a7fc3=${gJ19HqoG!Z&LUg@ zCBG9zqwsfH^XiW+zyq0DA4rS}a-2g97dv_j2NnY?1gy`)Jg<!uuS9g=<6Z&*&%D#%K zzP4IYX5iv*uI_Qu10suG!kT|SS0-@oKvjZd-+K`*74z-)qVIIC$Js!|xX=UHr~eQQ zH#6v}&ypr+UMv2OuMangnpk=ybMDydbfC=D z>%kzd6k@W<|84R&YOz0*&{4SA!!h1JDQun3lu$iFNbXL=?{THM#Ob%P^0*G!ZP&7s z2;6*(k0;`RpfW{{xJExrkV+GT=D_DNwWL|E>0;yUa<+>9j%0v>&oZ<)WobiVfLvIQ zSscdsWpvSoOvDAN4(k)?6qmF@MvL_(HE@EIQDg{j%1X6)R#uT(l_8PWjxf>ycAW?X z6C_k^=?pYh!r#=aeyF$??yx^j(^F+VIuxxW7d&X_*Y^cHI`&c(vpl_|pTkRug`s(% z;EJT2h~b_H+v7W*c%JM3sU4G|eNQkVDznmZI8{b@tZF5Cefrsb3_9F}(7$0(D;9}F z`H=HNMRaf~!5=y@{4?LqQP3i26VhHTm(>alipSDuCBs~QjyptLr^}<^4hj0NYkn37 zu9qWbz#G`31kX{4^>BA~=QVE60svP=o4e^$@$g&aN=kfgXDdF@pMSIOdO-t;w#>vl zls->)&ygKdk+5F$JHo%1bASoTKmHRaBo;-`9{J_uS^qAu4ae4f04^@66`Kf^;Dd5u z`9`(N(!=qV;ez%o_P0VcsqE+Oh=%kbCfSpKpeMBL;n~9i<=uBD-8+2VYo~*XncuyW z`YOL}-Mc?9pta^3?iS9zWBZ2iG#Efb|JLALvAu({RsV8uxVh=^hSysS3<(Lj@Oz28Z!VpouYc0hZ3dkhJ~wY9Y> zvs~X3v9|FqF_XfubRv4X_8m_b(SMXADs`5Ze)1M(V_?rwW1%d`zjRP7Z!kvR?tpe_ z-tB_h^Fi2WvTG`?m-(DxY`D6ehIqsUaCW-9`3kwJNU5qj-e%~ZI_Vh~c`z9>t74nN z+L)Smv-$?pQ6V~NC?P#74KG_+u8obB#Hpy^lX4Y8@{If?o=eNiP&j%8zV#o04%h3S zFk)0xHWs^kiOs@h8asX~4S}c4$RumvD?%@a+M$q26+EcmhJ_-n;F7>B4l%Ahg z-bn-^il-MTCVU$svk>YLKNT~Dt)S@hDO1HC5)#n|i0D(sahQbEHV1#rlRqDwORN3! z&syNyU&5je2mg{uMsA$lcRIev@{S@_oj%7k6Xq?k{Qt|!K~Zku$b$ub=8 z{qx>c#UN7pwOEc*$~n7G0wy_Vka`C%WkSUNc?4$5xc=AdC&L#YW{R^n52a#O&8M1i z;g^Nv<?JX+-{ zRuHvU@=uRMe@*zx2z0zVxSBnL%FwC-k*-v~8brVM*lCztN1*opqM8pa2a&mEa#FHZ zo+vk%;3nq8Sz#dk5sO0`=wBY5o~CAE8eFk3_~)*HUe~|>e_VvnIi~+zfEtGWU*|)# zN5s-YC@Hu15L8#^)y$R4nGTh>vzX;^5?Tu~4CIr4=DZNAol8TrMv#d~g)u&k@q3xt83J5-Muv^D-O_S+MyBpZT!(OJ zw2(&O4qwOAHP!X|+gpO=@VDsoFDJFh zgNOI$LZZ2<2TA7Ljd3Thjx62R1Y7&Cs8h@RKNW@ZL8d-Y1mw>clj$H)8L zok%Jp9?yDp4JSw&8~q&`75hQfJQt29ZDgEx=j-U=5|*7cFc`pWeXA4qjt}2u_vw8^ z@9`p_T22jRGyJZ(rKLL>cPHD#++531aIMMXMymjbvN~$Cm7`QH0RKuCCP|KO`R*`dCTW=og$TYzSiPpVG)$hTV-mJ zpDKc0k610AUsTXR(%pjr24ev_9xE#Yy6BFXP4@=$F^y0z{#$&ga`(M<|=!qz<=B0d?jI9IgYQnCS{>-0~NR!NJWUkcU*N zKSZi1Of*~Xy-zD1`IeW{d8$HO$5%(&{NAVmop+F?O`qI34>+-3lp=;{{*xlUASIJM z32enl-`CKN#{;iTzs^2JY_78@Nj-qI?LmA8!TR&euM+@^nhy3)`-`Haj0@(4v+JBBFkx&!X$A(tt=S zJBGA)zcZ3Y7u9Q36HlLNy69SM0_e1Oi&e()+O9Z?zU6)pn9h}WK7t34gACs~fC9_r zz_Dt$y?LlAY?^O6+NeWuapC8i*U*-qU&PtseZ6-~O?^ipH1}lR^=EJY_-+-jS(N{= zmrf&SW@e^P-Yd_tjv$#WlfDWqL>%&P1s|CBZdf)v{{Lk!JFi`P0+C{2oF_CrtY@H7Caa0N6ZRZHl`#?DD71 z;i=Ml^|2~#l$QAwdvgs&Qr}P%@vC4KPR|hwk}_D-2Mq z$~LzPmp8~dH1}t`aPjcCNn8eF)+=>d5YA9R4mf;!`v+>%0mh$+|HUOX#fI0z5yk4F zx>8O96Z!+3XPs;U!D5wuZi5(!1tK&8xl{=28|qG%a;;yVFv&pVGQx;Oh|Bej6PwPLUJ6SbGM%QMr{J8(Zw0?^4i)HWg-AKP$6_+IvxTG4 z(a?wmEf6&Ff2;gHnMXg}1Y8ctmx-m@;%p2}DBx2PLVvx27AVJZvjguJjc`Z9R#BUq z-Fk;p(BmiWda)vf^=|v0R1>i=5rWPZu{x}aUTExM<4}+8@$BFkIhj_2U6ERe@9Uk{ z5_CqaNWIE;W5f4b0o)4RrRQ`7u9INcpzUH|URbiamKCE;YL=quFFO=FpUc1bM^SF` zKe7JH1!S5jpYH6e)TlI7KVjxNw39lZztrb{)M~QZtT9%P2l}Ya?Up685jCtvQ@*8r z{jd!5N~gUjP{`a`u;mvG>^Hg!xhdk}nhiJQ226je_yJgJ4i!45>9%H7LdKLVOgj;b@>>`yWxHa3Hp^lAHyzix`0VK<#*|~$}8lhCRcO30pwk? z7*F$iz9Yro$Z;?0p zA9f3xK3uwzn0iG=2!xi&;JFHOHb)s*S$>ae4b?e`-g4G;JP)tOddt;!S>JQxB&CG; z-5~fwcuFlBzEBR+3;Bxy7}P@2U^ZL$+48uAw$aR>;AFZkHaq*RY15mPfUsTev`Ahf z9n0l(K5P+vua5c8&Q*^x&+J=a0T}DR6*Lo)$K~ZXaG}~?-XSSV?O&G1v`g&D2 zsP#24vJll&Np(1Kvpr*D^p}h#rukj4idUhOR z(Uy>JCl{2W+`;`RQ+xE#NlNW6}#T z8`lm&SPUT$7KRMxN_eGCdnyTU*HU-i>8!2g*tQcogng2s1(ym^RDDm*!X)8#c5sM{ zj^?*Fh<&_GXRL7!(CM`6-=d}o3OZr@IpTGHLBZFaJCVG|sJ0z{A<+Tm+DLt+ux}3_+M5ADxVjiQ)Hy zIR8J?on>1c>$0wKx8M#5?gV!jK!D)EgS)$Xa0wpV-GaNj1|7HscXzkbbFDr1bbI+&x}PfDEZ3uP8ERE@Whj)@UhNh~-wi*52IznsIXJHo;O2l60Ir8rJTLTmf z6w$IkKg3$^L}nYe#q`T;X}|6|~mH?0mJ%xuyi>D6x0lmDuJov8|G^!6* z*eV`NX7FI1T)Ha|tu9mJ$H6gEW#5P=tdvbdLb-p2hs%cjz*MU8ZF40d`x}!(0uuDP zb#isKoh>-5>+kP3Lj*FTfQ!teq+e-iqV;0PFM8RRMZ*(k6RTR_g@8-^GuV;MOn&Zq zqwhZ6{4PhXuX{LBOeLtND{iiC_kLN2J%L`{Lc$yzjFbg$i41s$al0EnM=NfMj?!#@ zmuF^>jb|dAo1+?E0~)PI5(r8y{vQ*>`v6g6Tw)?7F`_+GfSs+0iwG<8Ql1Lw^g_YTGScf^@Fk%nJHpi}$YR$TcZ!JGkQ+Wq4RB zCajqXGkm%65TY{;9{tk>&67?b3>`MY`Ff1KqqFU5kzo6!ZpV|b=M97|lCe}ol8s24 zAKTY1ShUcYTqE`2PqCU%P{$&O-cKCm3VEIS3X5+BLoP)P+o@$z0yWNO)04MwH;M+Ij2iV0Kfup_j^kO@Xc!~|z> zE7NbpZE7n9s?wVTR)^$3`0ex*6sS*jk_V9p)Q4inpur+S^On1wLg1NkyC>$9h&`f| zW5ImEj^@zZ03B!XxfF?3`fYn?Gt-Vm%gbLl-@vAG+6kJDiGR;M*{iS$3UFIZH>aWW z@)mzsj zUl0n92o{8W^l;j1C&I_~ew_9=#dkrkA{Bh<_U-#RTIl`>BiwW{3;uMl!^7~ubGcrD+9v)gueT&J@r}zk2X#1Lxp>5jr zd0}Q0*;#Y(dy}nsn`a z6h-*L+;fnP$X{;VH!w$Tp4|mxncXiTR%3P!6ZheW+)E*UQ0{oH;>XP}(8(h z^cgO_0e?xePy&X7gpUW% zJ>8Q*g}_NIR-`X@XzMQ63WCV;CE3yyTsy$krB&lWl@AlAL`a2f?FL584R>v#kaPA& z^-G(ZQQKHqA`{O~vqy`iR<5qpBsfL=x-JhFHJN-a!9(zElqy&yz1pPc0-_61a#I}2 zhKEpyg=LCr2!HI|?qR5~w1;A{Ch5ueU6ocENqPt|9nKrexct83#dSa**x6H%#-LZw zwo5Cr$(%TibN_R|n)3NP?e@C0lEK*6kL)b;lt4nEydcFakuwr|YS6;5q^fY4!r+@5 zoSdGlG#*1BO}~1%b%(zR^bs|Ai%x&6>yUWMxNj(ep z>C$*fJeD^kZ|V(9|MAa8)L#dFnT+}tRdj)KJFoAc*e%rPRT$LYP<`gYvB;rO&u42G z!a~5^{`|;PpJ zXeK;&9DFMd%D26!W3Rh6N#w6N*|O<(a2hQs<@RoF@25><7EadU&^CRj;{-NCh7j9o z_mv$z7%~h3d&l-s~i12PZA@PV0}c8ZfZQQO#}(FQA-jx6@@T((F_mh_v@^|2}f~ zl*n#u>Z*}a!PUT>|mGaAL9i1tr-uLA;(*nD_trQTnzi4tHQ(2)LU*^L8% zbv0cj^jX*p(-ZsQ2B_6640|uPzt}Efj@FTyvXS-kRL8KPt-Za5I3F(e^-P5ms3WG$ zUE%xySE^hW2KWr6H&rQ^gE4}5S#|oTL;EVo<}o-p@^cpf6+(pXaeFdv=`Lp~iiG_` zP~d60O-Hb&7lSc2XU`1Vqw7BK1e7n}K@wuk5J^#d>e^89%5 zC~$`tnWEHurH*{^qtOtWe6~=VS4^iqcC*X1rLJwGT49%0y6eu9`*ZKM^=jj0k$ho` z=j|3aH(&6t0L7oaWbaRT3}f^SN0Y)4gaomNW|OJN4JR8ts~6L4&Zpn2Y-#bY4nJV6 zdfm|Fa$es(0`HvqfJ;SY*P0YX!cEYByzcv) z|N0eqs+O$gAY}6Th+0`pgPQauLc0?*40A<k~)p_4ZH)ZZ^h{Y;?2ZZSTqVYZ3InEl$Uy zl&UO(d$1|XkpA#!XGL1V?t!uU8?(X8JeL*%_g7Y0`%l$<-Qe`j^*y&?Od`Ph_X!@MJ2Sf62W8Xj{Zr{zhav>q|z5G$cu zPg+;awp`(+NJQSv?&OQN6KE!PjK+ysZUH7#9%2O{jKyYSe zW^yrlP1b);jUV@CHYb~%F%%7K1S(B<`?|c7<_>!@ute^QKxGN2WZ0Sb*ih(R&q;qS zgF?~IjXK5bPy* zxZ&WU=2N^E4@r2q%NCR|mgHx2IfnnO@Arcoo77aF(p;$P{cXlzl~!Bt*Ld8&TEHkW z^pvVbY~#SR;m=P9s`d_sQ08-Z%o%1k@zyz>Q#Eh>TZWo>vdqx%34wvopHV4K8tB+_ z-p*FKW)2{85dCuZXG@JvR6pgMfYidoKY1$W#=bsmPoom4eTV>J$Ba5#?EBB~=&PxQ zMKv6_M2RwOf8~fZm|`Ss{ZR_^R14ePe_~YJ15SN4!&=Y*9#Zgz zUtt3XVq~AC%NzOaQ{K!>4q*P0wnlyYv+%V>3ka7+ zUoqk(I;jNAJfg>+f+GW6KROIN)<)JmDa>=S@2?w1W-c^(K%vCO=o zbwL(^rfPSs=w?V}VHKq=*g08JCe6ya3G?X~UF&ozb2+}#w+hq)9fzC?w%!L18Sil{ z@$)<^T)o>p7g5RJGQD0IRl}|z*qFAzQER9oD20R24R&{z>WXs?{gC`HG|RQZBPRFR)tC=Lfnxdca!4z^@u{lT zPv0d>J!A9Pz5UTJ-;Pp!$E(N^d(!?SrRKg)>|So#+59-Yf>WmMns}iHS0q6;K_FAz zl;0i&Po&>}6ZO`1mfZoWH)XlVcBsX^RC8}{tGNmyq~uy?95r*vLw&040Y9wN@@jJ7 zcMtfgLZjkJ@J}i#nwtHK6a^+{fhJoWBN}q5r7)NFMM${(#_nC@*U+^MUqRUpJUbRL z9vzklB_$atXjE(*i`?0{@zoat@_xW1aL{tz%?`1;6CtzPg+GTv3a#i5j37{RVK7rF z-@%cw`EC1D7E{GSGSo{i%{b|qGWp_}YmoCAFbc)TMf*jZzyE2V$D_R0v{R+pE?0%N zyBTAx`f~+du$AOcVY=r6d3RwU4)QZ%Ds+;gX}r$HtMjy0b|=lspm=(&uKT-F>a&y>Y_<`~vR|x@Z&&3p#FFUFeywffs`p)p0ST5`Vjl=+*pH z6?S%akz1DA+}@WNjc`-455);9lQp(h7OcpmSD*a5pmrPuPmxs+G$X3pSt07rC=;VKZ#3$L!(2Gm zbInIcZd9e}XnQ+3F11Hj0#iO{jB=2+reB&;>FyH@K`YUvvkw&|HIrcBgD3?ddq-BO z(1Q21i7JIrCy-WO&#lQ<3ChaJA)?zcVj$>JmwKk)=_^IhKmbJ%t-ISzQGUF##%svi zTSLQ0Vg-}4#MepPF(s}!zQb?JxH$;1v^f6;y4aYNHBImlV zx%uk!IQ(wbr?V|uPuBsSTRU+#^|L`iJf2tu^U%+3F~q!Ro+U`ye%N#$02P>@@RaXc zoPOckM&N)B-@V`OFDx9%dfu}bDf!dv(@-d5>h(2I=KQO8Ofo$U3`(ikV>*-P2HuI5 z;TkJ~i-O!?(_ts#{TvD5GE=84{HP+XYSi_F;>AxuV}XF-?wm&0z(RT-qW! zx&$imOC{{{xdS<&u7YK=PrLQiP}D117S+yvzK)K+lHCw1c}ewh5{`uvls-1$@Wm?` z#A@=vKlDp8(ocFeab-S|V*p(PsGoA8-qvcS)FlnMxdsV{3`LJ@46j!+rNhjn=OJfUpAGRMb)FTz2cZ?Dabg z>Ms@svBvD@O-n}dAGDk5koS-GMSPteNL-%roZ2r!itnB%_;Nd#CqEM9!RKg0PxhpJ9c zxX%PXTM6LiCcym25vhIo)od!N!Rq<@(UB2A0Ipf^cn5t9LLx9@#gsKRdp}8iYlQ_j ziC87Zr`N6b=a91(6cmaOeIQmz?ijI=TJ`Mr1Q12P%Q;!XoY5BhO%8AC(-SyU{P|i{ zqU!V{{IRq=o(?bcph0Z!;u}&Te^mTRj{ZDT>}HSSr+KF1;w~Yx;f=Uhull|bZ%a`B z{oS4E`Q_CW=MOv<7P32IX9bqi!#{Mmk2ff}HY+)^g6ofm74xVS{u>4Cvq8mS1IM3Dp$pI8CpVX+%%)#{Mes!Jr*ebYAM3W-ufw z#NJ&}*?WF^3gmLXIq1QnL7NfwKS#h8RSquhXB&gdz zZLJZ5^ziUZukECvvSN5hEW?YcR1}5JySTbCmDtD3WEi`UVr8c)7GQshCpe(X6hC$? zfE0YQhp-Y&ckj|zTBp%F=FalOUlhl`;r_jzWSc><{g`a{%)395$l05SNBL3?CE zO`NgFtjhNj)=)N|PX=22FobRb&>wF^$1T|{UX@kgW=3;8T`tpUi`Htg2^z6|C<_eS zzB#OOa;%*%e;XNVZIif#{3zQ15)FhGEs zaHiOvSk%31MGJ?)AXENEK`CUjSREb_L1Yt2-RdrLMoLPJBZ`$`*J&(_3;gFn4oQ2Uvp+a2fpIY?BC#%w=n?Aqs9Svf?VnQK`7mG31S3}IqOx?83COuj)(&7?Xc2?2ZN0*jHMQ{}?2gfu0=P=l8dilzl z{@Nm)dqsZF{ciTxc{&8=>IZa+BWCyaID_ zaZynKKLQMoC%QR_-9)Oe3?66%xLF1u)9{M7pj*eVG`UgI#Pi8=70umWfP^wPgw1K# z#VYs~=tLAnNh+RzhK$hQ)@9Z^q3=&uJ&yNO`(*sx1#ON1t##ph9uM|oggy_$Gc4OMBqYl62FYKQrN9cZGJqK3Z%py+_wkYejBu8=NfP)_qW$m)PuT;<1pf~tA(EMEjIy5>E%^Ss=xESdW55ns!oJSkSNFU_e z8Y79Vb;6kj0ol-)ZK*$885na_7IV-(RK;!!lJ!(QR5bkYQ*gQZ4^K}7(LN0@*Gzt-o()RW$%Tc* zoMvCK!SfBatMZJ@tSFjjPitEWNzr7&!5vN;-MKk&5)zg}q+DUR- z;ef&4#F*Mm>e+APj;kGBGZ&T{(4~Odb93m#tjzJKELZ$DtqUpL(K$pC0NseCnKJw{@bXx zkJatj>N(h;>7Yc){P_8X0AV#|bJ#^J=Q6k$u7iqShNzfGTtP9OgHUEQ5Cwc;^SAjZ z>buiWWNN+Xc-Nk?7l4{P-S*Kaq&jTA5V-TB_N70#CxVOVcYl`7nCf*ng7W92Jmf@RlHt ziOI=8hCmYM_cxvTW&$jtIk6c$$G6^rx<^vLI_4`=V^cqj^%j}?lmlA(2Mw(( zITN9WMbO;Ij2pyouwi(C!!|VtLk`!}UVtF8BH zWh(b)>kdv1=inL{2~R>n0*#pMA++FmoN*u2Go!oldZ9G6jqfc zn*|UxOOPgFlmY;4$g1JX{;LIaE4u8Bg`b>Q^L(b{Ng(m4omGH2S!{W{U0Ba}S|$RB zp(H8FoG*clwRY!d`S4oLcX~y)D>X*fW8dxkU7Zg{U=yX6%D;{Rx)%|5#m9L(IM$@+ zhbOEE1H*dF;hODG^7X-}Oz*4NsSIGWYz2A9+vPsX&EhIR@%WEip5A8zkZAyw!_M%f z;uXkvmQH0-Ipr5I&TZ_LJ+-{=>OvLn4QCETjhzF4LZ#YfFb$XQv^)nNgv4~-?s>&f zkgDm1~>DQ0W5* zw4Ams`%OmL)ZAMlXY=*Pu|q70Q0gN{p_2KEWi$dlfRo`B`p8V4BVU$eLqL;7I-jJ$ z1@z*znr!`;w*z4#s4(enruHW?;S3<;Y3B7iA8V`QYT9Q)JNGtv1fk(jR7>YJ_TgXQ z+qV=6Wwa}5`u`(IKw527Z{~NfJo@Ex^6(TW7oROF7PhjAPJR9cmb^I^zM0DR0vZ=O ztv9p{=;PkVIiUs>e_MR6kJ;B?!nd9pMx_%mk2<@4oVpyV;b$Rt+yT-Xe)#OvRHrWYxS1cyA2CwpP!iHQF|HF0t1?>e2|XI~y4het#ht%LDbeoW%61+mcR5L+TbAM;K! zVc%xAb63q?KS||bNkGJR6M6p#_NEk))3|$%k=-`TayeF)O<}bGZ*1zTL)#d|q>J5qJt4>zzz}+CR%~Nqn$x59j!_D5waqSmy zoR2;Qv7qKNm$I_5>-k#@iCQJL2;x$M6;ar;(U4pS?)%Z4%>B8fY_?GP8xFA_W;hSa zq_GvUAyx)j3|usnr-H5g#GahhcM2lTbnXfwK4Z!-PC+kA=&&ze1OtrA64bPUEh=@o zpgi#wKY+|fCnf-_gSM{7$=aHOhoSZU1TZzY;~rRJW(|HuguXj@tFl@k?eIbiyJ&52 zS%I?}tt%aaFCGV!?MOmi-;9Yx4?u~0K)eF00J%X3s{eFKT-@2&S#DEE6{)_tUfbFF zR6b4PIrzJJoDmxFUz0&)hrJsI(qwozb$3p2{t;#mRI9Y{O$zu2C{X<1cy4?Wt~5H? zGCWD+w_ptE;ooyV&lZhCggKo-d*(^^U0i zKqU~k0+hjUTc~RQD3QEtj)do7ob?oI*dTdbjflWa^O9@BcYDyq-90 zw+H6{zJi$3kT`iL*s;yznvZWbv>|wNv+p?&4mFTIrzuGPJD}t=TI+E8{ei+&~t+y>s{lpB?{}#BGxskY5UWyT?ScX&TnD=n5d~#_f%3aoNpdPRD0t-0bUhO{Wo+ zL;xio?A`oY$Fmq`P%)3s6OS_LIP}+mD_-x16#2}y`&rb*m39**bBaOq#@hZPhxfNk zmadm(-PNa@uK{#S%tEeeUn7_Z24t@jf2CO)P#Ad>|F}E(%A5>8VQh|`tK{c!-F2ES zcQ}?rql$hFC2MJUb$hH%zW;pVkR;t8imhyoMz*xSu>q0P%&gP8zrTC>x23@+l=&t> z#p?4q)PaEp!)CDUa&wW)0x5-L)WWXiR^vMtfsS+BLDJ7}^f&;!TG)}Ph$rG|80y?= zzgaI`ak65jubgRwkwuu`tvGS0w>17isWeBtGYV1Z`{v~%^8|D1vOV6Q(N}){Ha=ed zz@Eq~NNdvSw*U{!o1m@D-my(Revv|zx}i99Vh-ZvszMpz8vEhckFoj=Ape?DI_$-Z zoIu9ldB}M~R+VJFucoY-nx}GfKk-$2qoC|&)Xc3cd84Kwp|o?$!K~Os?N)MMl|P8s zLmEuTQ$oSVrIs62jjFd|GwZ(0tqJloYYRY^2E0us=qRSo8P=3mJH~2OL)9?^sO~Pw z?)5>Bn9K)zb12hmC|<8?CB$d8SkA2%MMJFfBOK^?Uw)VhwhBPjq|JuB`vlLrn^v1H zf^_Y8&G?7lRHnvlHa%$&B7m{Yrl#gYfSOSeO4k_H62gWs5i&zqL&bMGO<_29W6~Iw z&!4R(e)Vo_Z2|3-Y=0pOQ?A2NXqWz((V-9)DLI*Wt7JwC3`XOLj7Lm|p0tK9Z}mW# z0^BsGE5O{^`F+-(x*+-wfHYsv8v7{05b#JHobX#NM7FWXZ5Fm}P9xnqZkUu2O5z$) z7h^Vg+QlH2l0U9vx1!!;d;ys}R8X*^n`8)^-V+1qfVa)fj-s$v_FC1DLd;0D7@ZgW z-caq0wyXemC=B9y?U}5&u&7+G>(_w@rU{OzCPY5|y!5aKYn+B* zbf}ky*K===Md_a+LJtK6JDYM-n`%G97_|@}7(#$8DoW6TsgFyN8^=PS1tQ-Y^Y%ee*>nvJkV)1xK25@H!Mh>W+bg}AL)GX96Nnm)?z>5nWQ z(k5w(bc3iWm~Nu!N~zFgJqc_S(WMeozkM>TO*>Gsa_1aHj!Ye68Guexpr5+EB;j|x zy}DU`^|ndDtl^RleGEgRDvAX?D_zNwmeVl%EnBzjS)XG0s~EK0ncG;Y{hW^hRrbwR zB^h(l%8(0k+fB7<Jp+OE7tj zB2k*lT0_SW-yT@;g27{qO-B$nyf!+MuDHw^vb!H4_L~ZOz`@N8e!FSc$f}Hwi>ueq zkB_hED`LEAeZ&W3o79cfIlrTOag(t8ctBNSj)^<1Lb-qklGs|YE1(J=MJn8_L4s8( zg%HkdYYPQ@3~Fj2wx0u&N!B&6>|I=x=~Ycb^vR`mA^)FkPiO>RF#czi)c;reqAI(4 zBFGVxIyROUz3TB#=Ui*;_P?X~S3E{%1*g^!7G|h?u!i9R2sohlRV!01iUu;m&ayJ# zqEOdBxVcTjp%T(iQDKf^grB5y*qrSiX8!smt_vkeQydbNMolBga8A@g+3*70-tJvj z%O*w8(}WiieS|I4A{n3h+M%^jckMh8C8Ii;NR>rfi`Bx@CS9N;%2B{*{}zv*G9*ybXVI--;q0cqy3;~5fq3E|YZchSf#%ndb*y60Cstg^;N8jq& z{PdBb5b=m`S-B5Eu>ewwhbvvEga~+mJP^TRfN=+PDCe%hvwak6c_#g?bMwN%PJrjK zzsjQqPd4-R>DX?rLQ5fks$|YmQ650~bvqnCb8;@dbQ}eT{4uV5Xx30F0s_V5WbyVg zA0Eh&i=r*&>w}dFBcr1^Ttol|N?tx0z^L@x^SK1@J0EzyK2aGGu$y)0#Ag3^zC*oo zboX>eJrLd*yk*$+$xki-X^1H(RF#y37BF7e;LRJZvHEw!t~uKQTMafqHg<1~G>2Qz z%4Uzx@X^Ga6FqC+s2N}cAGZeTBx6Vm5}EJH>XU3&J+t|pK2cCawZoh^+J4yHn9lSm9t}{rR;- zuI5T0o^LIV;#na}0Ld4QMWG64LQiC|+g%qm#E=QIIsq{BvVV&1o|1~A^fdRnx9MAo z&ASc950Nh4!1rf6ii$7VZLWwy{cPIW+8rIau3RuOC-d7Z6AeX^G`n9CPkE&3X1^7QGYWZ~QPRE9Gr+j=HUZ&Q zswaHX?Z5irgPRoZ^@#1w$OYE#E)?ZOf&(9U?^;UP4<4D)j5{g39TxF>fP*QDkh4Mj zBV%zZGvGml61Im96b#c(i?h9bo4_{eqT*iV2)pB*2LU`i;}6vlk|CO!yk! zx9bpe2gF^}+j6_d@yW>;V0RK3p{1n-NYTCP2c6bC7)ewgk;$yr5Hyliz_DazmVA&R z#bf<+v((db>+SQ}?GNpluR>qbPzPl0JJkk8fPw-!A>kQj3V?Q=_XOrFm~v?kd;^7! zPEG(h^Ex}5@F%|b{*WDF?C%n9YCN?Z8Jn>RU?s@QBO66RCLt!?pLpHa8@G}!bOdB@ zU*ET!9DsYtwOv)L(U48=d%h#V#=yAQ`O4uPsuD#i_z!>s>F;8g-G9;C?r|;T=GF#9cTI_@B>2U$18)5fhgdgvI<>h05zA~A?Yml-T!X1E&M#ScM5c`cutI5UD zSxA#f;$JNwIy^it#@pylJ6QBH(5L|ghDiQxWB0miQ?Jwx;t#;QJ!?Y-RBemFz^ z&uDTi1|B}XfWeCj55A+WCT-zrdu&vL<(%zO?d`$m_&lS$t1HKq##0`{od=$%F!!k3*cOnZIdilZ*9@?E9Hq%rTzEbvWE~m>*K>J4NRR0@PXBxkoZBYUB zwYztRdTGa(_&u@EC{pkbbA1teQ#CjNWS_Xf>K_Ls06rfY5<*2ov%kOJBkw-8vty>y zc1I}WE!!8eY|wY(OPY=D!DiUsT)FBgIwAf=Fr*jm`8Xs9FaWu1bQ>s|PERWVKkdy3 zZ_JPp63tX%R8OE}j~;EIq7`=eWd=ZKpSRa_4-5cSh}a~i741&%tb^as38|g0aL-SV zG?e{;{0Tyf<^)cTj{ewaWI{>u9)QZgqQ|F4&I}$E%;!ETl+n09>+tb@+1fJ1K{^d{ z=rCmUxH^vg=HbD6F$1tauijQug_RcZxPZLV>2@AHRa>B*zrCHUp|Mx7+lTQ58h*P| zUMQ63inEw)j2AIC_GG0COV@n5u0KP&f}nS3#ci+Ljai3Jk_@6k8juYEMa!U`j=el` zVxki43b@b|(m0?A8O#-AIic*8-5s1mhHhMN>_OApTa8s*mXSB3aLeHpk{vi3i8dWs(J?m!p&&Fc9c`$7o*EOhlm6@ik)EA z@d~s>Lqb0I8pi0EFXYyhm9<56@AdzofC0r?!h`}3$f#=eT-52sR#?o4miLXU^z^o< zsQD!XLz6mFrk?j|r~RbZ)>{y6Xy)_X8BF+HPv0{USLOUyn%M^jWx%9kaQ)Jv3=M|{ z+XIUh$l%wGPmI~>GqlIwWarRA?B*7OM$&e&+;F;)+X2`}Ql6H}trnPB^e*9(`$%~7 zd8UpkmyKwnUDIMa0)ag{I><;@3*|alsi}Yo$@u3qm-gM=o$IF$-iTIU4D5x|74G-P z{YlG2%h_QE@)l7-W5*@HjN^K;h%$s=6BIqcEY3Y_*&AfR%xtUGSkCY5P8Qb4S9k+k zqmXxZwhF1Y;TKO6lj`ZJjE?+iyFYve6#u>EC5EPEDknH?M7M*;-^g_<$Rc`TBE6|x z(9v9&63E5?F0Sse1)*%>;rHy%e5cPButk9KS01d%rj_> zM9@)GyuTu`>2(#_l4f;dYz)iR@{#8w%OmE%vAy66YvB9ka*Pg7p7<9-4;((-mRT*ljpp(Ihb(kORtl{VF`C;}P-%{1xmMH5$r4r$ z7!wPBt;Fpda$Kuk$kX{9(4J0DNxC;mjVWt1v zTTCq1^O$@3wmVij{7zk-o}Oa8dHJ6}a+Yg(I2~A|GdlMO99&#bKn`LEPDuS7>AW(R zFPROrB}s(pCsimm&dzuprfQGSpMFEYL_iE6bhVd{ zXusu-(UPk&Xb=Ka`0a;j15_Ab(6UT@clk9xe`X?I4`7P|@mJoU}r--Udp+X4>oYL%h$_Y@=fDL@FI2*1^Hu&5t%1j?GP z{?-u@Atzzl1AM}Pp?6?Bz;ZYGcwKC+uCz51)N;-U+y}!mOJh~8y3{{(TCS=-Pz+9_ z-J;R@qm!+5IX*(QEBC1{)fg=x@5|570RjvifHI700Z@BApu3VY&rrx?We>{q8)6tA zHhJ$)RA_Nm^(Oa8BCTiM=f>BuBY^i=G9YgeRpb`Yr@gDzUc==h6s>0O&sJfFR{z9H z0=Ohn(>%jIW64BjE8w3?BUQ`%l`-4kWtR(XtX7nMFIOsc4p%oZE)tsx7R=;#gI^#J zOm;bLyd`7N+|py{R(ZXuBH1|IzP{zoo{ERdCHGTaHJ>N=8y29g+2w1ZGE4E`0*I(& zOfSTh5;391c+mD=gXeARLt?xA0M`ARuCCCAHB9Ko*|^oVMt_!3mG7U*30Et$?2t#> z8y=RrZ{^@(EIrPKY@pvdW4b^(TWS+4$->9eLC9e0LrIB_Z47 z4n`0W0(`pZ2R~_zEsp0;Q@m=9Yp3pQ>oGi^2$uf%1_1_Fz*m%~bC9#L-h#~Z8;Hv2O_Lom&!Uro}WG%xlL3=NktX>mZw5L!>*+B_b-CX z$MzrR%)?8-10TW|4u47Ow$I)jhkyGRP$fJ-0cEJjN)CMc}MPB zE8o07CDl0=jC!48b91QPet^+9h4c6n&|3+k4R3F6+ngVX>rye^`zICBuP;oOo)t29 z#=I5en0{^`jbJYFh`XZYU<5pWE|Hfy1^CkqxwJ<6>IEU8Fkm*;At0W&dfo5kOQMsP zgE zySspW5B7C_=?}qA7<0v<5zID;bh>MnKkNj>(H7kL8U24TXlO;B(NEw^WEtc9h`7pq z`@DdPY?M`%07u&9RoW}dzlac$Z1`wD72U6MNDU#i)%v&1XH5hd##DXftrDY?Vg*_dH{f~z?nOM+qXfQAeVdEq3izwb&;^SBp(??qy zHMKo}J>FmkQqR+6H!c+f4*}oHgfjP5?XB)~Rl*;2gtrkVGXj$%K;F-7@%LrzwJj%X za8Mku7oxZZu)Yq}k`aY)m#*h*S^>@1bLZU=;Y4Zy8O5X=ng={H?&?bl9 zr`eTFA}6 zppY*iHyGu0I$Q&NbHO?yvH_i1s*vq};atx?KJ}+G$?g=`i7KnJG+USMslgMOwISK} z=6gM;YM55A(+|uTIKEM>utwFzQ~I0Q2Z0o+RA@exCwGEnlM_)gEO{||Mw;byZ!vVj zY55n$nVve#d!e!;at&?za9f>${>kARpO01IoU~6&W5%Tg)(CqT8xk$@?vz*LkS8p< zYT5j0qqXkn$ycc`bYx^Ru2l8ZgoK0b{h%oeY^pECLkan0<%;F;GLm`6uE>nE-du*F)A$BcQZb6GU4drwi_;A!y$?F zT?_(G)_iC$d@5>cditjiSS8^S&(l$fpY74^FTB4|noH6~J;I8T8{}YkdOcH9Q-jfu zcTb(B#j#^k4u*1Xdxd9JV%b%W5E3VD|uri z@ftXEVSd~OK7ZIym;5JGsPI+bfPkqje=sw9V|~IpNj4GYbgIl8UK?M9=6eH5M!9S* zBEGlxw-s?`xT_U6J2VomW#IfSSewvkWqG@?;wu1YYFYx!6TD<>SxK=1782F;$3RKH z4cP3Shn2%kNMr)1l}@`8p7* z12$U^76z!j)* zy}>+ZOGP2G#zq1iowZIMG3ZKzRe0jVLjnuHv@SC#o6N97vg`(|7w^@wo%;k1X>0i&dt{fES<+xq@Zy1gPO386Zh`&czII<_pmTQNnfudY@ z1lrr0+dd0m6>l?;hVh0>Fxcxuu_$WsxK8W(Arcy`(`}~!=!Qi(p+ImF@uJ7bD3d-r zr_niAjMuq*ldeOq!~8Y*YNbmA!{>5>L>5ckvwKC!j9cFWhf-V=;>X>G5G?s=EJ~0o(!euqlWiL0?y0rm@XJzetJH~<0O}Y7ha#E#e*p< z+#fHA+H{kn0N($1)OoY_P{;?aqYKzmrRO_FjH0xf#yJX*Kg|c*EOerRzAqQu{+R7W zMOmu{ebh8mXUh!@Zs-1eAyFYHyq-5A>z%l;`iitWR%-9;u49zVSQL6MSC(Qhi6#&B6<5R#4E0g~aXHr4?eU}XGWk^Lr+4OepOZzp9QSVi zX>$(XXA*EZ;(29wR$XiM1%^|Z^!AqmvwR3r54D=CrRF^l>{eUPPEPm$Ibji#kG9CZ z8HVkh39f8@I&bd~A>Sx2D6-Y)`jxW#bSpE#JP}VVkGHfESPOi6`Q>o4-;n_4X?as} zna%vB`6=<_!R|U?!4zZE4;c^}9oE{>bM1w^lA2A&g^M`mN>YYDN6R}u5abcXqMVWQ z1Z!v~N$cz}8R*fU>yn!JRbZ#ru z0n%EBFxS-3N=0|TVnI3|1}uA`t1Dm=&E#B05P{TfxrzaL8IKS1d_H^b06i9jSVbWB z`#;sd-rg=gQ>t~hPjW`a4tcFv5naH!!}0vB&l4IUuOTr{RF!Uv``KTMK4Tb|`A7X8 zL(}VnslXGuW^k{c&qom=SK2h-$t;)qPxm>1JPg)PON|M0V^V-=9K*x#hj8D5Kfr+K zaY}&EasM5g7BH*$&(OI4j8F?8mqIOMqYqjxS;Gz$`*t&;jBPRiHUIBlk|)BT<~(N; zzE!=x{Z6)Gj*~9H5Ph@fA!Z)g*p8i zh2#gM2GIZX|3H>#IdAqY?ISrbs_mk~yLnt;|9<>&(K0N@=7PLZsbCE6f)dn!zoioU z)L~HjPZB=OIHZs8Ijleb_ZxF!S4gPj#&I&7tT`Kh`}aB^Coma`_JtOox3 zp~#7Uf65HPaW|TU11{r#`;qkwEU`#=TlH)xR6u_C-xtV>U>2hfL()niHH!H6BLtY2 z0@Z3O!vFrGRPb*JoP&?zK$7r3oy`J?|EtRhqQ^zK4gcpARpEf?g6isc9hpWFD0i1v zlvTv&=uI^gp(?$q!3;LNz<<+!Oc-wZT#E1@AoQOX>u1~R5-j_1ZVOBW`#(&*bySpX z*FH?AbSd2oEeJ?TcMk(dD%hHzYFj8`PS$1FPC$jx%au_ z*!$?dF&!)c7~{70>(VZjv24;audV^YBmKHlFwpqK;KT;#M>#$+Qh4KZyuicFePj<| zsR~XUCAR%o_g!{5lQShf0SEK?dVY+I>t5}E4gUQ6JPa{G`eoQ5k|3}s6C5dL_{%(@ z8J`30$CCMd0Fud;8w#A$yLW|3ADY$2A5p|0JGlheaPCL zkZs4!;L=uyQkr1B6Hp!@RRSUB^GSVPZIL#`4C%3Y+4e11%t~osnlW;U{g~FF!H!N|QOF7c9d@SH=NAt{{D5gIS9A)_eu!7I!t9{IPfEf^a0Q&9}cfKIHjf zzXmHKBYhYB`RbtIj}+F^@o;oyCf`JjC4&b;ddqdzUp--0lxG*9`?G9*CD6auegE{D z#*@eUZ14ft>mQB-ovWVpz0Th5EnNlNWd| z`X3uC|F#G8uggOe4zr)z(zr3BBYc}p|HkThVr_2`ncYbDN4@6H)|!?A+PQ~+n;Ro| zVJ%fPe&ywU%d1TvZ)*Idi)i7y`mpOlcDr|o(pTor?d22pb|(viX(^0~wPy4*pqnPp zk)J@)72`XG@3UTAZ2MIka@Dqw zu35zvE=eUDn?J!A+B~+KM@L(qY_TW=1a!c-7(KO_wX#OiZ#_Hx2+0ZMm;+TglrN(? z9Ioguo|dk3o;+OV_47q{f%%yg$UmT@EZU=@Vvx+hw;e3jY*zX{KFw1k>b5!fG8?&n-siKJ+WvuPI8PG=51G%eCKRdTy9py)m+V|c)#7P7Jm*1iLNV> z7IecW*NE04CKrqs6@^_-4LDJXiUK%_>?U{Ccr38_y<~h1ISi|Mdq$yA)bO|WZYjd6 zOMz4<{xb@tHDbH#8YCYyGN>KS@*2JTXY4>x0Xi_E*atXEEf>}j2qkBsx)!SGoM{R6 zN81?yn*cx&*!>y)p_x~p$8=>k=VT}Sz~m=jUDSonP_asA%g#@ z_0iu{meJeO`Qp?r??&+3H=DJ>0$)~P74$r#tWGlD=lkiHZ2oL^43d?T`=*|&eWaA; z;TGFbt0mr*$;r>%$s27guB&x7Zx03C_gg2VZU;jy*5?2*@%eZ^INfXd-o~=0&FlH> zQ2^)0^_cH=RN0rs>o03Z+$DeTLo~H0Y$l&IwIEKu^?ica2Tnusa zL{|Vq=PH7`GyYcbadGTU@Hfd9FNUwB!g8iS5+TBw}a*m8-3pM z>qB$&4_f!9stF}6UJuh0!rm8)v+s}#72n)!HT-R{yCXhSP#jObyY~Mv5%UMR{^-dz zvOL4?XF4L-eB*1okU+g{+vW1#E)HZ&iVp9aF3l`&2^Ja-j=cj4ABy)LD4U(a9lj0T zH$E5rI?%VkBMnx2u@+?|J9nThG*K>C+otp%ebC%c=-kbTXXQsNL`ntMshqCH$Gvvln2Sz6itks_T*!v;?7 zULTZm-0WXB{^@K>%*YUDfjN$YLaiIu<}^@j6t4C3al>#^pm0)S3>)uF2QC0aly-D&F2392u)MhPdt2Z+neQHnXaTOw{{jSb^HC}uq| zqBpd1hlHsxTqk;th8GO1Oj7j!XHf+E0}=B7GfQMJ5dVWzxl6va+)@pk!j4>(^8iE; z<$VADelwIr+z>_mc^{3~5 zv49sX%$GYUD?>?A=Njkgq87wg+?1f<<^kz!O70VirJFF5WV$2-x)nn|EjEQJA<|}g zIXCB;i3*HftW|w3!-cQ~#q|G)kjP!p{lhT|^sTJ{vL!FmhTxc`pd_@N9LW1y%?e`1 zmV&~<7aU|A8PxHlXI~3%k>J!bWn2dG18am*T*rh2Q=1qln=}S6b!udkQGO7MgUCpR z@y$ClE~{Cilwr`y&=nfui`i0i0PTws22DEt-uGO1&xFPrk#LzA8yUr4vH;jyJycZt z5uhB}X>XyVEHA)FN$DC$?^YVLC;(D`o$ru08JUU;-q&MA?`2}zX*wcp7Mwt(V=52a z?43@xG?=&v2Pw;%bD1ehW9M7I@UMD0qX{&Fc<~WkDi9c@OL8~~h^ayXLuZ^NO9!9B z8%~yg$~O^nxv6{bZbI?qvNzHfFF;CFI`3`ga(zDDX8!v7bdq#}){6*$pBcs(AdSsU zH+w?s_jU5ZUVpdRLIs{4uLXTKEe9se&#gV9n&9XQK4hJUx+&a( z&k85Nrb2NtEOw|S4lT7N=Cq(C>xIL7h)ig;qQAx=IjKV$cGdVRx`DcvSL!tl9Bw)^{qnqO@wod9#2|pcx68@O_Qu|9zU;gEc_2q4 ztcr)?Ei~Mo8L$^=zOVQ+_)-RArY^0y`f! zZxX}vbiGVudKgI{X-YSOPE9Qgd`!uHCt*;o6dI{xQV1s)Sf=sp43v6Yle@Yqn=IVD3qLQT@j(i1c)o)Df%7dspxmX1QaF z3n-~@PIFEhuJtZ($CNV1k!b(&B+5Y&bd6%elc;v0E>&yoHT zXtiyNfRvIh)%fV>m+O4f{FbJc9wHr`*o+Ji4b4-Z^4^}(*q9*sPiX+DSR(gUj#3?A z!#?!*@63fkS<3fpRNN9tUHRzxEenQKY=2HX;3)9TD%W0t5=v`Vg@?z764ZnPWu147 z9yDZ2@GYQeX`cE@RRg(Ae0(+)$Ct`Wiis7;X=yzhb^so0dU{%}m2+)QbyGXy6f9#9V(PPX(w_~Zdz2m|2-J9^j8U-4udTwQDi1JVR>wzsgZO{ zDu$$elUEK#i9}l%Eki-O^=AR10tu50VOjBF2E+Y4%DslFdLB$0K|j`#*w7kWS(5^= z6CZBaf)vR3#7{8{VksXZ(|7sB$A|nP-lk(Fj+r?hG49aVki%tcHR$dQF*J^n`%QSY z>sd|dCM8N_5 zNYEuQZ9xm(Pr6^vkm;#M=_=+pC$h0xV9d^|-C46Rj*KImpo=oSfr?R$jL5j!*=?S= z$(h8`Zyz4UrNp8c0D&8b9JD50{->iaOSEQl;LW0}amejUc^G+JX{ zcebmleY@bmjfLLWe3&aN6DhI$)$9j`sFSrR-M-@P4q)!^gC?`rx|9&-YHnU$_oA3h(BPF1xo0z_c_Qe;?pAsJ@ zgyk?4$E*4}wV8}rNv)|C7PNBXZ~P>*Y3Fs7Y>Lac{T0({qYFQWx=+`m7f8G!lqjf- zn|fr8ZIUa_`&4Y%;n5~6{jS%}%@sRRH#;azha^(vnJQ)N{wnj8lJZb;(YPURV`SR_ zAeEPu$XG|-)E zxbVlbYpCA^eWNmV>M6=pXa%kjVqkM4@A=Jr!r&(mZ4nmPMcsOzsyDXP5^OLsORe|HDpaV~e&$oJU))N$oFiUeM=dVC z2^Lnx@Bh_9Bg6{*_$@%k`Zg`gJm?zaa2CH=U=Bg^pD_lE(=uP5FM^`j%z%Abh7S%> zP{7Whg?)PN{$y@xs$RGf{szs@rIgWCBKwVoUX|!3SUxT-SA$UmfOxU_7Dcfv)++5X zK&;*97m|$wf*k|do!0H{IPD2VEqKI`u?iG&6Eux&yLpj*-R>I@%b+{^RpO5`_$=FT z!O6v!U~?kxN!%d(ZA%%sfYAXBe4IbbyX4gOI0q$ran>AN5?8wd$) z^G9|K^OxX?OrO}^;n3&neKDx>uMP#t{V>SoHv%v;he4ct5c@IhlbU9JbK&lKrgy4MD^asGbgI0z-~Z{_oeLMTlll&nx{)#b}yF&53JU>vE>(l&-` zu+AMSZD6(7Bo+M*6ezY>E?v1}^3_p?M}{hhkG5LgnEI<|QlduXmr~)|ejSKlH^UKO z!nkBP#vBVD7eh0m(|y+^xlzP`d;EL7dyFl-X7O=4MXL70Dc$PGY>*v;j0r+gyCil# zQkR)G9FcF1<=03{)<`+6?HEV0trO456o zWM~?R{ukTc0zazf@k-#2Ab`9Mz(whb_p9~Ea&sdd9TaB1=ft<)E@`uxiSAq3>A;18 zr9hXtlmxGg(r!ycH`=@+TBT(!P(#0pPob5F z#M0}_=?!i_$5XPX49&{X)r-j><{hvm1rRe&?2pQf5o&`OhB#R9lestkXCWfOx>T5V_b>`De=RBPiHhCu8Jc3m zN#v-{YpWh~_)3r(m~QXFMl&I>!#+Wj1mf8$TkMd6;qo7&Tg;3UB+hyQvs>vzXgTo8 z7|%bmGEpX%v{SnH0Od4;pUnVg9rIOKMdQSO+prbXnbd z8ZwfwKU2S)@}SRocQ#g|6b;9A0fsV26@=A~GxI7PT9RdJPnEnPotS5uI2E>w45Ng97H?Ik}NBk=ZG73CCl@mb_eUi)Q+mY zte$<1Q(ClEr>JSiaSa0oa`X=%ilCpVJd!$=pwgPOxvxQaQn8w^^wz1*n{*G-K9fXk zyR4!M#pRX``d81A|C2;ivrWG7+v{UGa_KWGlmZ<=B9kASVhpd$@`fK{DbwBOm3pUd zmrGl%XV|>z$fS>`R36aOE4-W~$Qy3dL-5~(To(fmH{pcnHT>ZWm|1$QgnLO=m-hs% z1UtLWN4IlJr;^SN1U&YHtYrZ~4f|h*l>iHxMn$f=PL4c8er9`<6Vm{s=q(Y$DM{3* z&TQB#$4g@yPRw5{E?N2$lL@v!Hqr5T%ki^sdaX>b$QmWMZQ>T4I~&8|JE=3Jnp!03 z?oTvJgmbwZacjV~F&;!snkr@?Z+dmzDu}qFM4mMDNnXG21zr(j5xPOiyb$lSubC7K z&nuXdf|r9l-0rIqpj28NoIAfd?SUpMk*R^Y5(~bZz5ujDEsaSu4}XG2y2*3f?cOXt zPb3d-2;c!-wV|mK!%FYyGc+c?KQEmm;N4$-y?nASuV$P`DC|ly`dcO^Ri!c677G9G zsuGOAO`ImtV59yDaC2d+{?4nVA=VW?9B1mPn-+z8$Ia=%Ha}Ogz8Be@Aia@V>3QG-aNBg@Kyh(&q}er@$62SW$VO$))g?V z)6LZwYJ?NOTvK{v`Mp?0!C6ref6L=y&Y8gA^eJoD^)!K2SVrh(+>sCWp~fMU+YqRG4O!Vnpq=DS4b)U)VN_qa&H9vmW%mvm%+4<-V}k05*!luN=Y5VU^i3IDFaNXK^#TfY8D_P3(!xlsFndSXWM!!7{d&-i^?);x=; zU;0iw2s;NLq@eK`bXK>W0GCB*B;kV}6Iw~- zlTrg&Z)i>I9PFR`m5mQ*jU^@xR4%%aaVj1Cd-Q^NZhAB&c0n;en4~0NPwe=GXY^;g zwJ?Iz27;(dT?ot`15;b>R)V4r$4oB_+TkT-`WORWif3Xh{FEqOy>5l%Y4~AHjQEB7 zJ2lXY8*8EdE_xrY^rURLZ7P}FwT^)TF36Ea`*xlk6?pp~ke;lhQB^}Nk zzD1i2s9Yw(_z0+-&l5P(evmg;viW3(XG9jJn_?wVadv+Sv`?ZcvZpGuwT`FZiU|cDSxn7k_xOR z{|OW0DVka^C;LLVa-M8)m9JO2K!vHm;kJ)^rK2l>yj+-EQOLki?Zly(PLr#n57iOB zH$}7hZcRM{1f;S#{?3)qT_e{-w9FBhm)B>|9kcXn~y{P zEV*TaQ%x69iQR3&<>Nm~nELeFyb>+NP)QMJ1O@3v(jk|jZgtNGE|jqMe4x7DFQ^EbWb3>02-FLPJ2nF| z;5*ESx|C)aSu2^GVGzh8l!$~wKFrCOU(<`0^1ft&@yo{)#F*XGhrkF#DiDT-rk zf>RGLZzKQC|7D$?c3ga1r3H*=TH~g-k`PG-`Fzc^kd;U2df^WE(+uzGFr~tK=8HFZ zNzg@OG?D%}k%XDjDxlP)>|ELV%)RW4T$z<0rKp$S~ z6?Oih{yac(y-p$+HWp|US_$NN4ZyjRXs8uO3SWOIR2MjhoGp$wV0j;MunHmIRv8?Z=VoR-m{qz zY(iS+cL}H8?)1h~a~KY`fG?J6AWmohHmOwhTcFX9uuMOdOT~fZoz;g(?ITEkS#QZ; z!iU$t&;Y);e~&xU>(PsBjL6kpBFO zMG)kXMa-y5q99=tpLUgp+Y+S&@`EP$M|T({H=+lNiKQvKDR3NmFaSSbQW*L7kc^oU}mQ)bH6`Mru3U&z(#!Mb-br@$BpA%z7Pp zCaV})4f+U0i3X@oHpTkG8>5qxXWgV3ID@Im3ZhWmFuGF&QcQ-EGg$}{Gv7GP$vb`g z(Foh(H*>E+dZ=#9*>BbKyJK;oE~+8N73b4hf`DZU-yKBCokcC&i^ga;B~TP$Bv(yA zH+3NU5iVqSG|Wq?KN?7GQL4lKyGSZdj>u&2f=`}YP-s2|r_7YYGHc2JFY4ZGjPaYQ zrdZQPmod6bHm7rUk58zg*15PBVvBh)6j*nU{8O`$eSf*}POP;~y}bvGg|)16Qlo~V zZL8?ljq)n!X3SWywGSKff5uqms(3;N0<$JkDz9rNU$sziuDzNFH?*-r3$ZL| z?ZrcL_yY#OJ2JEX%D$O!`NO>qqWfj09a)9Z9|KFpByTTQC1Y3iR4$3*d`!x@wsJZr zhiAe7k8~FYzs9@Wb z z0h|w0@6^{1qY-R50htDgUcYxPNpym2V0Ck7LTu%umvRB)CQ?ZHle|4I2gw)EMckgM z8A3_=t_HdJaq-GDc!Y)+8Vtnbdzs`EK~1Sc>JAS0>AHxf#Yu*a?>O3c=}H4L%b|Awd|63! zRXgaEPZA8U?8R*t>U|VF*+e;xHoYMNg6)9CN{$!H^z)ybSQb5>5q}DLLQ#BvN!swCB0YNJm=vOy- ztr3{ZC8;V~Tb(VJSoxXND)Rml9rEf#^c5XrP{PmZ_lQ2-u$keXbm+R?ip~;Xs9Ny$ zPq|06#)aUzp$~2wQ+eWltr#QueyEJUghJwD|)P zbTaClnPuijx#27zzOTsQ?QyC4I_Wvivg9|w<ojk^3)3B_k3y1ei+2Q?-(C$jhy_T>cOuvnFDE zF-xOY1f`6=I+qw4rn@<-h)fop+z%`TJR=e42G;$vr;JC$xKV=6jRE`1W*f&`Y3J9t zL>RmU;cx`+#8vE9Uq&E$0XNA6ls&=iIL3qhraWM86Ve!u06PXi%Z4 z;ooHmBT$2_FLguJ<||`w-WDsQay(wV0&>O9U~YRAKSDx=t(CDqHcGg~wW--sMWvtP zRq`g^p_wtz3mSw;G(`5ni(dWDDbW4(9g@E!RtW=3I=tCav|!5{ghj>PJt(HS0ZvtkZLY|W zs{kVR>90|5?qJ!eC#X@x0$!5B1Ys4Pg}70x=O;_arnY%ZwV!f0#H>gUkDnc77-J_R z;_&jdvuBaHH>;9|8_uCi4s{l^qR!d$*?60KE_r5NFjr#8lroA^YFtB|;@SGxv`4$8 zLvd=Hb-~iAaG3VnmY7A39ii1Uk{XD*WC)`AU-K-1_=8L`%qbC3<<4oNmp~e>a@CcS zhs)W#DAuvpp(;^Iyvvp8%xQ0 zN$z5P1M}*;`<+CnqNw=1ON$=7VP3lK^*ltv`ZsPDZ!n^;%i24KpXihZ{$P_tNnU%p z`t2uuB&iDT#`UA*#w+Itr4A6>($j^SSL4zy8^3XN$rgl2AXAAkQ2Cqviz!0T%rdZ> zcW!&$$FU1JpjRGgcO)9cyjQi%H*eyb9Mp=&dw24aBfv}Lx-t=;0A5V7!?GbamR3hJ z;fpk)CmMvP+Fd{mk(j@m)&m76KUmZywk1mXhdH5b_`q1>r-=@ij-Gmvaak(w@WEfvfd41V)7VN+yBJ^NDFY{ zg(bOGc~UK#T2BnEBMgV~B409a6O65RA8%)Vznw+HfU^Dhb`fhWhnr1?DvI)gKFSA+ zs(!!4=HH5W-t9)5h)^?{*L%uw?gsJ-Kp?B-PFFf}X;|urUS-*#wk*aIZRUg6(KeAP zPnPuJhiE!KL_m-x-~J$0yZ4>_QcJ6MCD6zOwO}^ZCc!k{`aZ?OjUXLKjRGthp(J5< z@CN?^x#Z2;h|kb!OSi5x;f;bS<9X{Be63 zyZP_zF>uVTKXnpi0EYDRqB`FMv+pzf@&jTkQ#qbkOAOPWM$~^%@LUF}9S|Qe+Mgjx zl(B!nVJ0C62VfvdP4=zkk~XFqS)U&)mEfWC;X>PaPzGio6>;z$GnE+md0xg&1*N(1 zXkEMsEPqZJW6a9kV075R;w?-*p%no)<{xxZMp3*c_=k5i3OFc7zB(pvb*^FSHtb~`*H7?7`pS>JUV=!CxDR>utBrnHN0) zCbeWKxr)m#w|y#oG|Lxy5%9-sTD3VjreG@W*O4}gV~I*k>gMt`IpGKkrLs_r zYOxRsB+Z@mgKJF(Hy+Z+5Rs!1O>|yjov5EdmBEUq7|%)FSvXWvO0-he*rFf*`Hk55 z_qo%^Kuk9?RG#peb8BVt!#T~~BLZVVSCgd7tY1_lWKEw{9a=q8ib|&$uPx@BTRQ}J zp4q%J?DpMA*F!%sOTQ?Uis)d3Od;X7LHk>hN9Cc8;UhvH+i?EV)s!;al0;wf0wTJj z^4L_gxTxvtVp8!`8Ucc`F*Sc`8nF5Vp(d3kjcQ{1#FwEGY=K!d#>)LfPkt4q*N(?K z7v%V-^|T$gvLsF+WAqS2oL-p!5YQ0Q?j!XmFL|s)%i=K`rla>5a*;AnKk^;M?9g?( zUViCzdhxE3{jE%Lfl~CSJZHqIzppDPnF3G@|05`m*+*ridv6aZVVOJbY>26(^{Z#n z{XtwJHX$yRCBs|4GL@twO&-Rw4oPzuHw%_TkIrQar7V208i1EIa{?}#HAE&sL#D|c zEj+5i4+!m8qjtoP6C`Tn;MsBUp#gS)T{f5hjL3fiRmxCn_G&1bbsmjs-ml1A`=j?G ztxTlIdsjB7tsp6#<#p_-V539fe|<+ zF{OwJ^ULn73W$h=7=}XRJgFVHftMb^K;nz*e?wj7Q)MjoPN^Y`Khg4rV3%l{Ic-P3 zr?#@x5nf;Hmuh2%X0iM+17VESiN_HCpXR7TG+utEERsTLY;*B)JUMo`KkJegU>rV_jMC$RYj0DYJE{hxN!*-gHvp+05WI|7bk|qdY&<~&{G=txDdf9zoSQ*+BjDiQv4doEbJA<59J)pR}Yb7C5Op`dw%3 z)nV-p+&TG7)Ef387eMAaBy7eOX^2P&H*;Wf`h&y_m;RetL&hz^ClMgf$9k@Rz{vSh??)vI~rF8|W61KK$s{$yS1@5ubZak?>cuNW4e8Ec2Lp>uHkc zZ@FLW#OE5=;IFL8j0aR`A2SD!S!=U?OSCi0U-35Y!vqnJM3|6^Vv|8RieDuSAFUMh zH#suQS{x;qdB%c-gz!cq{e5PrH+7Uk$~c1fyZfyitig$^qiO(Ep8~P_tdmt2-7+qPeFS+ZG_93$VwX zAZ-A&)8`bz9(H?EeE-%qC(i6sOF4oYqRLs1KhU`d8}!B@0#k`Of3-~Qb_^b8KY(D+ z(X#)dQZ1gXA;8XP`6;4X)kJ@q$G$&3c~Y&g2Z`H@}I&mzkz zah7}sX7xksB2ke1@6r|XNan&q@j$UO*74e`ji66VY$CtU@0oZ8Od>+UL^tjN1-Le% zs@FOU%1yk|c^bI+NO?IRPgC1};5d}S^jWzSFdLoB=+C)HdHS)r;_{Cat#aAmL2*fo zlp!=^wamD{rLvQ4B8wlYbugKcp*iEseIMxg>Em)i2EukYioqQnePdQHuaeNf0I@jZoyv)j~Nh&8Dd<9OCn`{F1$BCtLk3;2j5` zCYV^~7z5eZ#*UWjGq`-xBO)RK0@|{&C>RAmm_?RJ|!NBLUFYmhY`r>O9Anpv<3?b-fz}@MbOZzlG5P|m zsB&_+7XPZjz^W7IOV)CPUl>8sHmDqsUv5K|e68qwk3{GjFVR&fmOO!2Y#XQ}Tg0{~ zQl%KeM1ZE75Bsu83xwJ)?~f{EL;-Bipj#d}oO(jSje9~l4-aD$(F=66_hTvcYO8aZ z=*OoY_c}>nk@?v z`O5}%GMPWVrvpmjW&XPq016HGB7ewRp6fwB@sN(E{W6jUo95=`Ru0Et8RMO;t*5@Y zf`9uTLmc1&Q*5KZK4rE{s5C^SAAV1*&Z+~%lI_8mW?=_esJ{zzdHt#X)zv{*=Tlrvh%ZUw{;pYJ;dhyop-_~gU+QQj zKsH$hLt=+(T8!X{CY8u>r_}Nz!97k1QM7h#H$reMBbiqY{QsczIQ(uP6Wjt)&-m0J zF)a@MP=bizIn-Al@B2(g3@`JfBfYGrXfYc$i>F`Y1{^u*F|=Di;1sw}^3XV}!_P$K zCg}-Etp7Or)pYs;d7nt8FvrXP{iaK-s&T;G`)KLPz}H}|K~wpk{^pL~@#$)SNAtCybpCA)xJH?BO(|$LU@C^{ zuW9so_imsDE@r!(GDbbGvHg9fmnzFrP$E+x&~DG;b~jc|-DIQLLa;M{yd9+sT{k%BEj>LC zEpKuUql;I-kNY=oQWnRNV_$4SnNiqF(TS4}n2fLL)dJ>Bw+F;97^-4w912H3|BD3_ zK?6Tv{r9b{E^yGPQm-Cn(#2JhyqLJSH3H8u`LVahmB&swpXpc@(Af&u<(_P4GjI!e z#)bq7=rR}y7;=1y*}FoSUj{7bp33G#MErg8;dXa;{KL~#5yh_11kv$QjcwnM`hWXR zoN(RIW;kvl_IeW#|YH2fv*hyrH(Ax4!iJN4f+N6Mqh2T}=Y zI?lL^5Y-&&Bff-dxMIAQts#geK%>eUw_W|Nm8QXJeRR0+(DAS#{ofeKzrK#&QxkyP zG%^@eaj;12FfC>fkiWhB4hje84*=iZk@&R#`!_Dh1MNFzppPuDbVbR$3=x#23>Op8 zGNny$;P|8&QJDx95YR&b^ZkHO9nRI&1O)Wdrr55#slEtdF#LVCR9+aqBcI0oomCU9 z_V*0O(EWd_kp1Dbh#D*7bh9e-B9JA4W=`w-HUnir;rNb+O>S$Nx*+=_`$F$`6E`6k!1Su)Muj|1mCSGT|fma z@G}`39$ES3(~GO})GKK~NjGQnbbcy6AVNfoJoGY1E}e^~6B(QmOc-LQ4L&2XT>Qja zFrSgDlLG(@(G{V3<6BA>nwub&>WU@%*_|40PRrVh~nvT!s9jpb04v)?2U{`4{UfPak(I?6V|KO_Zf1D%C!|@bpsti znjw2(((OrtEs6($Kp8BmoIvDTy)ms=%GXC_193^&c>X+QJygDQC=p&uA9(?#9W&%t zfVOcLkz|-D`0ssy(atCI2)@74rW{r>_Pl}`{+_-XO{npW$XcCn$5B0$8o~h?>BL!Y zT5y&mk~)b+qH-(usxLbXhLTWQPZp?JG%Ez)SU1)4?9Dk9 z64MO#O@nxs+qOO)-LJj=jKrwy9X6C#gU^v+!lqr|ilNTc-}sz(VbehaMO zTGA-Oupkh1(t0U?gks1G1!iOTJj{ZzTpNrSl#mc-{e;}z?|{Cq=~wrcYkzNF?tUw$ z)3C?#12-3gRv7_x$4_m|Z&Crz01so1x>&~6#yCP+N2Ue3Zng@P_Rj$KIL+ErFzUZ| zCT1w4fV}$@=`&Cmx6^*}d>@%sU0rJ=?ReoJN^tnAh9N}f_)H8qrORwLu}E$6ea%R; zN3Nx7j9Jl50uiAQK?5{8QFDfA6-j1eT#O6m7wH9bwgyU14~39VdPYXiYRV}WxnKxT zSom*SVEM^W7K2ihLG2SW9Hf5b)a)sgfK<&rtY?)603-=U%;CVY0eP*0P<@MzuGDKu z)UPDeX6JhME&`P&fby^D$p@IU;%DBk9W96Sd-_gOiERz97R3`b5O4^vRUwcSc>9yK zowdjT6>QbuL#l+f9M)e2p3#S!$ItA1#1?xpeq&z3f0uk6TxHn@V_Qz1R@;y#5PDB? z!V^BO8Wvl>S+!siXwmBgUBLiC>0erPnQP6}>&nh3wyDo{7BOUVc9lsTy+WU_Tk-a-{FEaTj{}b|=A^ zgvaZ7YhG|XJ~&?1BtO>tiF)@WJonW+Mea2!cft;OI`pf6(Z_OIH%V1~zr=m7o2OTJ zZ;dN7LpIX2#+OH&n3&am;5L=e3R(Uk88eZ=16w@ul}%dA;n~?*WW5#FvLQ=qf}4>k zoT7Og!Js#A@!_Tt^%tb(F>O*YAwWo2VlAax6vuFg@>Z}({$ z3=fVL2qVB`Yw^AN>^^lIzWM7nmqYu5FVN`T7Y62QVF3ix-6ICjx?Lf``_1*tiX_Kp zz4w>_!NE)zLaxt!eNcEkcb>0V20rDZ!oZlpFG&a6>efJ7tvS<_)58V4E_=h*o<~O6 zi*KrawK$*8c>FvTme4JnHN!mCH~8~il!oEy@=98-#i`Y8E+AT{!Q{cF0DSjMG9FFL z0ilkaXglxf=_mYfXDDmV=r`9nD+2JNv`^AVDWo|4YQ`X8Kgtz_`TFx~a89|riI?dZ z7A|9YN8~sTOtzG5`;C^3e!Em&yT@*}fa!{$)oJJQ4%-jygC>c)d)axXy_px= zRcMOrEU(evV@-& z{O4DnlXcB`(vGrN1$vk9enx%LnsZNb_#$u1K33P_ow}GgQ7Ku!aWfD0!c-iz8GM$|% zdzM7#Jb0W`Fe1Nx|HcDU(?Jor_xcgm2cn3;0Ho`dgQz`kPgmijlp5!UX z|Lu%DQ?p@aWoDQlZX`y*i;vsIf`x%;MDTC3w>=#|783JtD0Eo0pf;4t&kx_-qbPWO zer-C34(9<62U9#e=6l2H(d2G_y#2Y|TcS6YfdX6tp#Vy% zTPVEmI)!TTmZSy$4lUmWNumWk?)>#uAoARAng_a`u6bSr<;H>2r3X7!<(UM+5MjO&PiCvCebUh7F z?kJpA&f>GI`nc+QAxVs>Hq9v~Cm-2u)?%ZGtYn0GX@HlkhU|WVN)4NoRALiBP7$zo{T8R0xxo*_~`_5zEOA+72NS z%vQCW`>qG*IT(w6CP5*As8(K#a5MT%I&ftL-q6^iIIsj4CMfu5LsdmZ!gd;MBV65= z&yN2IXO$_yP+34xF?-}E>yq|*@vfMSS zsMC|O{@x`Sj5j(H12>>Jeu^DTeN$`%8w@QK_dp|Rrh3^vr<18LlSRCuKfBi2Q8n7- zV_H0EPo-i_y81^{g5%@f>u1Om1`5N$H^01h#(Xs}DELE_gvpXluUzFbgt?UChQL z*}2{Y2X54{oQ14^F1Qqy7IZ7IS@G=zq|ZFt?C2wq*s0NUS&AYrH{-#*({xAP&;g;I zwAsz+VecT4)w0?V!IL_>UosqDIAb@LSw|baVNnvKEiK@7Az#UUi_L)$ zX--^pyu0hd&|iz<69T9U^t)T#@^DRMwM5y3!YmF~I31neBl$1eo;c*v&U+Uw=;E7s zo3}^OJ6U7C$4mJlwB;E%eTF-kWtXl-eI^#HSP={~fPC zMbxaL11vj{Jm1;RZp%r_sU{Iit$RdtCFvv{-At(LBw+UuD&zQ?%Cgs$f#7Wk|lY+^`${) z(VY3uH9Sk))v6f=2_yW#N}I_st!?d&KkDT(Dr^<*g)l8}h7 zus9A5FhQ*4mMtVx__Jmvj`Nk5UtsdiHRgNH!KdjuvooFTj&G2&EWkp71V5YiWJ{Kw zt}4~DOJGcN<}v)?PK+fAxrR3L-`nP)Co+E0sRV3zV)^Kv&oUnW&TX{ES9 zto5yRZlI%`pI-nq!yY|6<{4>w0Fncgp!&0!FLF!v+pYbWv^tp$!c z6Mc$Y?>JdK>jC+x3?AYds!uq8;52u;g6fEynp(7NWMga-AZBLfP_2rj41GjJP zQamFxjVxwt!eJ3}^TC~2Mo37+-rqw(Uvqm68hMHFD>Wc*fyYkB1XAJLK)q8#!0LyPA%K2Va~Zh;g4ytsC5&7M#IF!G%KIN?}_ni&XlFdHq7~K2}$7Tl|(KA)sU3B-PzoZlK`vf16qy z9PI5)`PY7`9}JE)`0}!ZSVGACPm zY|ht*u@rLWAI1ND)N*H6BFMF>nW|HQUMMLXDe3~U*_svamxK#+{lcKRkU=}HZS%zx zdgO2_mvW(^_jA8E?`S@d%H(q1?B5b+;$O=NmtJrDz|p{Wni!CfFFw*wiX%kP^f-Rq12;Y)DvTm(z;*;z{mH(>E>1 z`Q`BESX5L-!eh+;>`I`8&!D+tX>2k>ZM3xMX3{l&(?$WK=o7|oLEMku+EcUoD_z^PnNSY=K7)0mB@eZ*M+HF4e=_QnwV#ecIZi^DoX{R zV_XXcTfvnQsza`|JU#3oTd48#KLdHG|4pkFZja*hL#@Pwv{fYvyhR-Xu4A1HF&uGMk|F5!I{VP(sdkRZVRgO-PPRh9^Q^c0U) zrv&geKR^FT>&46-iJ98n*NHW_M*z zWGg$(!BS0e%I;A(?#upI`Nhxj*&@x#nAeYj1P_QlzSGjx!f0e$=okf(A6>Db+J~nRS!%$M} zI1`@PNR!ppfSz0i_u9^m{7yoAJZB;;4ULKBOmA;(CSfZ$Kc1SEdX#XTS&dlH^3iU; zeRy8LE68Fj0g}*=%t4Jhe}xxn3Ab~iIi&UWE5s^yQ2d1dXEkDyRq$>}fQbpSi8eUJ zE_M1B!xQi4dT#Y+r?7&=gv!${YDO5`=yFlAtX@OEIHH4#T!NfG!B2dOeevmU4=gWw z5_&Y!#kHsb9aT)MCvP|fNQknW+e$Az*f(a*3gF4fz-e_Oa8F6e7)7sNJ^v##v_F^I z;a0z6zHTX%){Te%xvI^PL_93lcT9h3k_sWW;urI`iN(v>c5|mK3)hr+OG`oBc!{SX zQ@k}Uh%aI3o-zILNDB@iQRZBVQBBUm5Fht!SV#6sIg~OE-=OF|l_F6~HnygRPv+Xj z)C^$DZEG73TVa+`{YsOReiAcY4-*e@qg&CrV63 zMYV06TAL_uV8S=HxY%pVy8o5lQ~f-3+SHzk`-k~ma%J+KH`Ee#=Y+ysCLOFGX?&q; z3N$Vg@HSLN92r=< zZ5jvc(3fT0l+FjBsoctwH6U%#U?(`O*;9n_Q@F44Fa{IfV`6?1 z6${5^SzFDz$Q~_SDf>2Z=x#yCn}RZ%ZFZ8*=jMIUI{KHf-_8!_Z>4_wNlV^^~xlZVDY1gO7t1yrbwoGo7e+hcDxVyR)_p;u6R~23(;f z#77FQ9Upq(INml?{N3(1q|sLXI+`VQIXR#08rMilMn(q{C3A7A7w)oW4sez>>i z;pQ^^!!MY1Zuz=}V+Q!3z&%E%I<-4nC_w62r2A2Eg2wS=HBu zrxDiX#2=^(`R)$}m zcul(f18PHxB>|NhI0ty_*-)MvIoP?_;z3o+*FP~`I~TVrQgOC5okR`f4&&YlR~|A;P=Vkt_tkD# zGGxcekT>oe*XwzlpEU@m*G7iDMMiYDUK6etmIG;#_JYtwvB$EG2VJatm;=38MeO|k zF3?g$p>I@Pd#$dw&23p$`NK`9^1vXC+w~!nH+UH;Gtgn`5i3MN+Et-TGptC(m-8=u zUjQ9t4yeHwe0mxfDNt^_R|=?vHQ6RxCK+DOTLJbP8j{&Jm#RY=Ry>Io zk1~N&uVx_dEJY3ambu^D1_Z>!O#DG-8Ofa-$V$wocw=5bt#mW}wYgBVL|TA!ZpfSo z;r0F3@BK_q$45m~E7p4+N#n-Ez~JFYd1LPDQdzQlDwV{ZnU@qcLBZ;N(YQ|+)FkoULS%ADIFZGUm#bpFZx&mXOw_ z{=$74id`IBViWP46?GFiFVicw<1IQKGv{YeM9 zzmzosgE$U?Nobe@gq)fGj;IFEaz1g@*I#}IRpF1!WL{emb)Ie`^xD|t|1}Qmv;p(T zeP`Heg}>t1_cc-0X%g zV6T)ajOfp?;$$RcsioUjZo7}wOYO5kyRV*;#X|_^ zBb*!hE2+ObrUhNd`MqopTR3z*w@#z8fu!d+#d+?vU*5cV222KE4$Zw-_uz|!7uRYS zih`Jxg%Hw=>HBjg{exD-l?`1d=pV{R{_bA-sJ?2U5?b-mmJ@JgeooHR%4n(CD85uK zGd=zD1w$+E*Za5iq-biwI7uCyq0co~xC(7?g}eZxpt`(b6z{1^VQjd@n!2tt-*!4R zUASwZy{~(zvfZJB8GWao@?)ZCiz}ROGg#%A=#)LZ63xywqB-Q9JGwb#h)R1I+K46T zDmVv$Z8=`21Kc&Q8-#KeMEHzm%VJJGS#joZama=tkaZ~+`EE}fSawJxr1YAO8o z=PH1E!lTQSDNIIF`~W8Dt9erhk$^c_T!(YX`QCG_hH3O8y?E7H?)zUk`r*>{B2?!6@ zdm^?@hhxzjoU#4D{!EU4B6cN?OV!bKvSy=Emr_YDk|9%cMNPo4&YN2K!2Las5ZFdG z-^V`LAS&6pjZEX|gai(~V?i7#F>gDnUoduh@r68thj%nxz8g5OJ)EnzI{RhjcD*}Z zE+TXI{A3!M^h&#Up`tiBZ~07=r_`iZR0mC5`z&;+;}8m$15s1OE;*W+>*sQ-=v^7$aRR6Tx?mgi@DPqHl_ zF7bYc_}wdC6CbZNSk}QDt3=A*f@*|;5_bC;pB^n_D?vJO1k`O@EcXXQ z5b4Mi1!ro8-!dNX%crnx5o`f$+il!S&KL`0Vik{=rwWlnlJ+etII1L>Z==hOt+$&S zU%g8Jrx7;(hwHhHK!?Cope3UW86a!M#NfNY{of*zRM}R8!6nUF63h}%hotT8)5G8Y z&x0EnJFo@`E(;f7avkiX3sjoEYH&xjcJM@;fUXO08xAUd%jp?K0_{_E-5me3sHW?D zeug)f!4jt$@$1Y1-`$M43@ay+727JZ`z;d&qD1ve`#+BD*VldWdU9}b?YTorho%*< z^3cH2id-dC;vp!9fOWgKw}*;qFiqdLw^ZP$sAw;_<@47;3mOYJ>_6TTjTfWt`*{zk z8|CLtqG(X9#~(~{w8jf4m*Ymg@lw<40E8705hIn06K>kKT8OV@_mbXR{#ST{l9%q7 z5}=rm%D5Fq+$Hu00t}>%BX#CipdE0CQDNFc{%;TNN?ZCDJYywn3uCQTxckOml$}@K ze|rieZQVhbk*$tksU6_zD#poya3l)Oc8bIbHRaoy3p_jCn$$zsl7Ntilr1$whKd-- zSHW=$pp<$2%DM&)dnjHp5>!C2C{*~3^k@`15e=XTwV!6;Z-%no4d`?S8YE=~{yn@| z^*?qTrlz`|cxQgLv-|#zi93y+iAplRQwaHWIt3kn?M37@A>h_O{k*?MnH*9`$&Z_z zhhspg51`bnXw*1w!!{|1x7XnaW`uiLQzFdKmQ_)4FHlMN`SWLNinMpv9_^qmi`OF{ zxc{@BGUh$#A9FkOD&9+1#^a-9RVH5~AV|Vor}py``(!f)Q^K+R7C;HAvbwp(Ero@; zom?c#%{}qH9nb0%-#r4r(cu(#^@lzQDB-{Bz1_|EOu@NpvuU0rN+@HZ{>zX>5yGzg z9&#;|fp%u^O5pL*QPF;UQ!+N)XQP_Wa!^oP{ygP zjGKg0CFz2@J3E-z_|K2h3*BOr3R=c zfJZ)JBLbg?9?X^`o4>I>I5+^lTNO#)_-90cs z+m$-gf{QOoQVZKQjr`9`;9w3Q$N4-3J}m>y%f?Cgzqd2jP~fzEkp-T}xyU{jE6S#^ zuC+9tu%-;@dDfD$(HX5H)#m4Bko+Si5z4m=W3FOj<<)<@G1|(}T(?_3?<|w;8wLIg zxWB@OzE*JBNk}O*4=(QF0pAC?N8F&Wj5-U>AN; z&c3@SA;^2LyEE+v=jW%37yfis`&3T*T~)}L=j+b0cE}fOZFfjv&wN&xkUnjynt&kKq9h0ssld7wht9G6GNIv@HH80|#?A~FTU$FxZ?~@4 z!C0`qk6u;NFpU%^@W{RW6r;PMDZblFYyb?_XsdRQp|5Y!NQ` z66~=xKQ{L5a;I25osN!X#rYI2?fHSgdxiC?0XJW_aM~OQ&v>~V7C&h{BmURa+!sSw zPfsAW-9zld+%s9^BXe%qv$kQ zY%H4h?HY{|L9rgx(`UsvAsxWn`4>4;MID;%K}pPe*z!Fip+AQ@UMkFQ+Qr=!I}9c? zH@>-QyS+?n<@CodCx|gYsc-=^HmzaK+^yp&euJP29p741pP(qJ>LO9o8B8FWM#{!# z*t$%!jKiKmP5a{G@7g$otPH)ju*KuB<%g5w<;BM*2-{Ypkt>d+3d1`z?m?04aaEPa z`fu@$2L8I@ndE8>c$zepsauEs@BeNx!6_vOgI>9`K`Ci=TP<;0Gt<12g?gKzIrw$j za4l^uH8kgKQeUl9Uc8K0ZNa}yKsdj0p*kKPo4nae9Km!{^h(sKN(Dovm6lWcKYk!U zR6kffnf?=KV}f*KD0!q5YR!w)+L4!GDk<4`#ah{HyI+lZWOiqo6Xn#SQM}37zj6Fq zHlM*sv|gga=|bC4uxINX5GYWhRlGNoJv&YUTWhU#LV(3;y2Qrc;&m`vY`;~ICx($0%UCjpuB$*f4I)JpEwJ>OXI7y3Sl6Y_d@lu+tY$GDd4&iI5ry zHrD7V6yb3!0ROZ-Sq}s!O2xC{p_xlUb@aji#a(~FI$mmj@9ddW8&h_@Rt_k7LTMI4 zj_c#&O!QXYGl^n{OEOrBcAU82sS*V1#CM#&uz+3+FRcYbK#;-Z zJS^Z$PeR8=#>pxA7(KYvHU4I=?rc`SdBa>d2RJz)T}c-#B{pr;*Y;_$nWu&Y zXc! zCdRO`{<+0i^goKAf1S(`x+UdkEAuSMgdk<&TW@!I+rm=iH$m?|O3P|Abb56YK^RE4 z{cmApk7PhYn%=F=p`dO|{vm5-n?(rWctd4tTi9Yb5&h3!-4Rd_aHRfMXuz9Ulm^_w zqaRmB6NGfR{pQV_5rf0S-}sPfpmd{!5EvreFk>l=+x=hNcINaEthdH-{aw@XFa^Ao zo6{rY@FMbCSyF(?VC zW{S6-?LY(puCL&nMwgr+6edghhTD4~L1G|e;DZb-vU{PY`iCvJ#nC~p8syJ|fI5on zBRCg4FmX;=ao>0pO=*0eiqYpw_<4p|@cz>!=6qQ+f!Uz2t{*rkmLTF7Jh9~(TXm z1|$w2Y37hI-!-)mf7yGdMto*XK&sl$=hx)P9kg440OW^Tsr@e&0KW6uo2T{7sI@1N4K~FnrT*fT3-BXe+LPWA`H)v4!J2=08EAQET$ckPY!~M}LVKyA1xzL zUZM1M5)OHy=C!1!=Bq?8c6Fd*OZ#oslzVk#D{iZ!#fuN*`jLDf2H-xy<$ry{mLKQG zMVWXyxkWHtynUh@VLn@bVd3`GTWYVWSGYF`iXnV`+f(E9K&MN0an1ZLsxhgTUkT5O zG}DCD`9XW$>alp#nQ`GEh3M4`QNa8+ly=<%_uSZmt^o}w$>q=wT4az-b-udn>%Y;^ zfN9erL;*Uiaxyd;FKw4*@Vt{r_9 zm?bdTPD{22y z(8dIS2ig8gQ?H2DfBv7w2aV8?(*H*SSsgM9aM3eWejmtvJMcPn{|HNnNVtXS9rI&I zRfQS2b4xS}0kV+chhCJ`Vt#T3SjY0aARrgV5av=m5sV$P4!_H|{;VijgX1F={{5r% zISQu}8CbFA|EzC2Q`aMb>p8MA#%H!v*g?sc#ojPKj4Gvw4=DL~<8ccbVqww1(pCeE z>f;8t6H?pNDEb%)5=5`~&f}h>FsA|336XbHV}Msa8^`}h2MxmXfzdEFxYj&0l7)@zeWYiJX+ybRQU6wSAF6_Ic)_0n`9gkH2K~+Z|eNjSId(X z>Xm{XFZ21Eaa>+MPiryhYS+|mQ|doh57Q96?Z16S-U5Vzv`7Fp`W=ok#vq2iZN~D` zDt-9kerf%SV=vS1SLOy)$pH+ZfG%#!1dL)6M>TDbDXX&qwF>(+hyut51NSBq#t>4W z^tb+qVX+Yt*wV;EDMXl@5Y?{{EtRAkZiGo|%nt|n5Yi3F>yXn0Q zkF#aWmxmBmx?aWSJ50JsnDLaEa~-!uL+Zq`TV0zCEH`pds^;U_{DD!7(+{|24UCkC z&4ZIbF3*9!+>KW4I?uJ)o~4NaD~FW!J5H%FbUP&aFPDY-(3NevKL_Uo8q^53M>9H6 z7J3Tb`1JTdVuEhIE>__lXK)-m$+z5_9ncdnl=c!d-0ax$V*?c*&(9GYS|lh*OGBzP ziy^ijOHYGQa-h6M@&!e z#Pb1};q%>h$Ggoqr->@+6vP;x_DTT5?jvH&Frwi=+yMN<5Yo;%ZVl`;*_wwsO1xdvJcI1B~MCarjtZ7kdY2}#J z9{f7%fJ)ucHPD9@W2lduD82qQM#MzPT|g*@S*4bGdi_h3PG4=Aj4Trt2;_^><{6H| zba{fL-SXlJ-JBL5f3dm|>TIjYa=zd0UA*Gq1>fUE8Q1b0#Z+1~d0EJ2@Eo+Hl$(+8 zK~QNHFHEmNL}z;dpSt^4fE0Xseg2Wha!x?|n@ZZj&W0oP@gV>)NwC=CONoJR-_Zc> zAeP`Z^j^}hn?8NtmM7E&i0Fehifj6w%inq3U7tJNmBc;0y~`ERO|IVp%Xh~xTX=Nb z%|sfPCGu@LO*%LPjUipi1d_`29@9s|cP)>qbC$lO;+QARjQmM}W2H#qB=78VfDj1< zU|9YIe-1mf&^BH-F;#05JOVu{x7k!Z<(q&MmU`6 zUSTITm@nz!fBvS6DE=~#l7`fdpi4v6)P`l_r!){iWuyK8H@T=uz@NPyh?o^zhi)cq zOGe|dJh@7Sfk0Q3^fH(U9aiozJLYZdHH6pM|M%q zdv8yXBxvkAP;RA^-%~nQh4-;*UB314Iq#sJmd64RlxglWNi&R*f zInUcf%BmMh|9DPiXcA@jZyzP0oERr(7$Zb>euq3^*Svup^9zqm9Xom4z)kdYS7v@Z zCc-#U<)tXgRjXO*(SX-BFp??``=tN0C48umRvX|v`c_bn_313yZV~1no;J&(?05_5 zV{A^M4187dM7zFsuU4mkB@<9}Uf;Dmn@VzYpXgjdpu4S?l0e1(E(mHzSwF|~t@uF0f*mU8}het)8M_87e%I$E~sYki4t15w; zr9wfHuxwlGH|N_mGIR*7EYl&GbPO>wEzT%M(hEBiP4ApHVDE zm}bv#k~~6q3zPwS^O*K(JDnh2ohA!sdFj(x@j%Gl#pHSzHxZz%V^* zhMw@h-1k0vy1m0=wp$2~wQY7CIJ}Dt?mx0t`tfuAcg)1Z=H}&>G#&ut?c5A;svaB| zneFX~#KXo~6}Nr6yGT>@d8{&0fKu+}e1iJ@c?8o@cJ_k0%5K+A4rfHR&{REy8G3A5 zt$c4<_*>*c_rMp8MuVMnPy4fw%j=pncg~a}aO1Igg&<|j?jx?HB%UD%ZuC7++%ASlg*LJz*~;PF_=6_wIBMec}rYj?3@9e zXgKFQt_27X%DAkU>P*=IurlnBI2P5o)On2$wUXvBEt2NKJa#~Gh`aknGVNzw`$42w zy^seVxmt~4Rm{r&0?t2E)ifn^{sPWJ6-6?p83(Dy8nl7iVg+$qE+sXsBNvzv#n3I! z8;7MW+Dcc)t?tMp$y~DJv0BCK6Dls15m)cF!r1ga^$f%b*GF?U+z6Vz*uX#00x6%= znhUa!H@W2U6xno3nGMmfj;xTBPPt2-N%c21F$P!O8*8RfILG3fb}>vecFIVpQC8~? zbTiMBe^Z35AO&-T4O#+>8^H?{=TOalIWC%IaIfwNN}qN}O7>DPjI zj@maE$D;ryp!}!B-i`$1c+5sEAPGlc^#Ej-LM{Xjd3a#Nrh=97U3bQBm?BE$fa^FHv8)|qaJ9<9R zk(hBWb{sfpXDSmG1BAsXLaUPcH{qWO`9%KsHjU4hQMPw7m5GOvc*T)CaiQ<*FW5uy z46I!1Qro8t1NKdjoCzq|x6($wm#3vVZVNuOdI6$JVt4Wir?g}?7(Rd%BDP;8YqIsc zBNn)-mnRfzTEpeMw9u+oTD<5Sia3^YvrRJ^6jYNfVB36dZJ~MB;xE4YdNKJSuV8kW z^Gd5&Y?7b(vDcR0?Ij#@&*@&!R{jXMWCxtlH)m(MkCz>?-gNn+sWAyH&M)oGy4p+i zMyzSitB+ilHRJlENM`KXQc_ZSI+s|(zL#HG`T6!LiyzErEQDFN;=654HGbAEF&Eg(gKsHaDYOtLh;HJw+6Hr4JvC)!u=0WW&vP~vLYv6`jw>H z(n^&A$yrj$hw>jp)%HH&49AA*@p+Q3J(TkEerR-EiD*4ebAOl|Z7nxj5wbu2A76;; z#exYSqd)N87jf!89XG@3dR#5LR+`~2HG4hnlYQzY3)OXt<#eM=YdzgdUvB*Jxt00b z;({(;1GsY49%pU0=sSTo5f6p)-c82j9*0ai=Sx=pIRZjInd@~R$Hh4!;C6GfUNP08 z%dt2G+;?QM5_XrYKO4>s)bf6*F8!$_MF>|ZkkmZOB!r{8;wGunas3{vp$Yj%rbs8* zS+q%^raYpddncZ;Bmp+W$PSuBd~>HU!}?+O#>SgrWU0(bz$&vgdIUeUxpFOGwkHO4 zSMm+L>PneBYiFO6isNkqgq`DmVbL&F=6{}%kUoi zWx3vY0bnme=vg_67c5ZN-f%QjJ%zh9|BxXM5BWjMFl(~2%|$UGtS9O+=ZP#JyYokd znh8NoB=iSNcDrH0^GSE7$R{ge?&g*Srx4L!YVK3FDoDACg`J%XWaRlFe97$91J|lq z=oym%S8Fm`er?jK;QOMDBwXFX22d5*NXK)|N;;kVD7s5x#INRGe(Wn3N%c9x)^n0L zzH|I{wO-sTy+3sd-}Pe=N%YI}W`6x%GrVr~=njCx!b2SgUwHyN+uZ6V2vl6y5KlXl(W|bJP z!cL45O-6L|LOCr8FRgT_W+qGqGhyVNo8lReq7{Tyve&Jkn%|`!)gm{Q|2iR=#XXlg z@C+W%!)>+B&sCw&-4VwLeM&-ivQ|erF(g!xlKAl)Q<`Vx85g!nx8XRP=sqMe_-DEL zV?PnyMNwpg(ItCOo@b#|xM25oDfEW~p;#EuO%QoDXxMK}27@`=n-8%Z4Qxm#%S>*- z2Hea$5panUr0YTs$b%dynlwwx5x9iRb;iI4(;kTR&L4{n1we}{uP0c~R$Uak8$=SG zqdI^bwfE-G!OGF@_dyUNqtp24n|mY@cG8` zSE_1RNlU+j7^m(RZ@yrE(3f|aXLc{LH>wE9B9ru?FZog^8knO560VqM{JA zH`Lr2{GXbiXl2(*?XKcsC<98FH|b8@BtYhd=FA9V);}Vi*Jf`_N7OFtzJ;lbXkYd$ zajL)L>-rED@trHE4{ec!C>5k*5!QV1&b^q%aWUlZb3QC>GX7Gle{M&crZq&b^n2!d zG1d3pUv@}Vu+F8@67gchwj~OT+o>W?-1a8JucpfU-anG$tLC68g$s^2=Zor{Pb&5l zdekfAMw#mSz4wc!D1h01aQxwE!c2VKFr}iWa5-uh(9cmRmB4Y9_!U1j0al)?$HUea zNesL9=9#O~*tN#PA+vf+X_V$F+Wz3e&D5I>-xSCA$Z~;`p2_@`fs=tx}pF`^v zYc|AK@BDjM!QqUVIb3hyBofuWt#HBFJK`CQI^*K8QhlVYZ5I?SXw6B@2w}GMuBjqS z4uHRRF@R;Tc6Ia8v@Frzz!oD;PqO&suion{6%^um8Fs$$CC`z3Mr~5nm=?7|^-((H zhY`jE4LB{VlF=nT@v8Yto;3;6++6I|wT6-biIUi9Of)8U(oN!muAB;2@;LjRMWGW3 z?ynr*D4lm)R<=g-hSHYnA?>!KR2^|kAbI*(rGo-rK`!qG;d900i8C#&r7s<>n`v3PDuFHzkC?khj<7vf+1Tnz8uzKGi#8nK8k!Kfoac#51BL zO>hvq?s-^!YVnjhYUJB#;T7&4ep~G{JLZq)?hMg8Fn7Wd?>2(Z0}WG-vWmdt%{Y)M z;6NumOd~(+__1>#Qf1E@AKc#KpzcoGOnmsHi?$XgSF_6cpy(`8p1mT0ZpIKX{_%I* z9PaxXW%$SpMSS{1_?_9S9}XVVIvci{SSFHF<*ndgv>6WY|Cyea2=DhoL@B<0&O{)8I&LPE9LT@o+)j|qq!T|F{wvON0C_g(2d@y&D}(*rMpNRZb&Vo zTToRn!o&0mu)eXePW~6AiZ3!-hqM##u-Wl4{}Nl+RAl;D`B?VQpv^veY=CxH$voHN zw^Xp&1*`KX7r1UUQ|FM&tjT@r*sk(bjAtp7$PHNc#q=I7Nq(l$k4{+|d|tGn>f zL0^W@A&W0r#r=e&7TgvU3;doBsts+|zqquxGD{p_eYe{!iw5#7Zxvk8($USA8x3ap zKj^r=?;#41o`TvyQFG9J*x~$PoQvPW?wQ4z4HjWPd)$RY?SX$&(~+DE15H&|I)QP3 zuo=!_{6Qsq>UNTZ>%o=mzX?Q^HwIW^EcNhx9r#<%9hlcV&@X8ozpg%px(BIr9cZ)% zo+m|y>7~=X(5Zu2zv7>j>JXUMGfc#oG{nhJngmE7dtLI|80lfU^#i0_W$+;qG2!dv z{1|2APS)Wr9Y!CtqYHCXkabhhX`gXXt!%H0@??|sKbT4iGb48^T>&uq+Jf9A7< zBQl*hb!h|0hh4r%1tDHUU{kXzsMxQv71JR%M9&D3Sf(FF%#O||PGjB*vN&C-rT^W| zp8Kg;3-4f;ixk7wAfsr%0ABuk+uBDIrTI6^B**eGZx3gLEq?i8P(Qd}?8GGWlB9IN7 zW7fPTq!>dK^W*Yb0+uv9MhTl1OUVQK1Qp(h|sD%9A8MNDXoC8kY>!)vj1-c+Hv zdq(9JwInouGG%61qD4vFhpJ-+2n!3@b6FKEx}c}S0IuP+HIF~JBGaAF6`Gm#p|a~x zRVlAlXn@RVFRdhm15k4HzW{llFeNCE7`~N$!pXq$Gyj$1B8ZoT=3zqbNi(r;`8#Gs zh0vWrb=ufJxQ?zJ+z^kc z4|`rBKAU8F!Bz6EMC-ZveK*D4z`y}%Gz>1Prz6jAuIA!rj-+%qeQaVg7Y{;^x~O!d zs+ZUErL*RIpyI;7fRtv!)!7*xWnbydnpAj(XdWqjtON#gAPhz75ZRg+1Yg!Ct5 z*@5#ItOq#sE8D^kLQ#xNtAAqDSQJ=pq)nzPVj`^eU`5 z$avV&$OlK$LZxk3Y7LAt4OGEAWf(lkZnMn?A=^WLGg?MsuoWn4+WG%l;sXm@KmL?BENXp830;oqp zf!~oWLN!8CkENjAxX0FHv)<5J6ll2-vSJ-A(@@0?{P^KukI;PdQ`b9k2;&nrl9c~@ zGZu6Zv7H*aMfH3Gy8M z3dwM_7P&NqT9b`slVCzTt+C>J1SSkA=T(&|9?Ggxf^av70eBif%RG@_B2p<6KT^l0 zj!TXtkrv6UWnCpMRC=P9n}OEzuzim6(VSqyMStITwEiHYf|)ss3c}m2CgwFG0?Fp) z!gc^*6C*vY?sxCiW+0oJJ!QnQb=B7fI<*}>z}9`NlE5l8L!`aJcR$l4zfKLofC9?d zegtKr2B+aqLON9k?0z0$u0l!7G~-Iit-qc!TzpW*K)Y9NrrW&(8y!->ekId{N`~^B zrF-k%;5RMBm7<{TZ6TKas7#nUIH~+%-iv5=r*+R305Q6=v-9m}rjsl8KNs@aj^eeG zS}#n}CmI^WPUE8i4dG)iyEz;ZD7*{oyMw&5NgYzS!h%nmVPmL#*tyQOIPNhGPNRX{ z<9vBhz26X>GgrDC_U8FCT{2RUp=4OP^_*U)Z>LJphB1NQhW)wK;K=7$eO!hzTEgvS zZ?;nW1h|4r4my_QRnUy5$)`|;++xd!MOJmNI?r*J%6aS71L-41;?;Oa6`xkLu<^;^ zzgR$&c`p4G7<-$D*QuwIpVbiY*3{wM?Z(F>UqMHdl{*>qMowLfk|NjEyaYI@gdu@F zMJ{6okj@*ldiG#eSa=$H`~Ha-!-$=p$i{Dnbs?zHirVb9!J#~#KuhU!ITmIiO(_W{ zIoldRS?HX;ie(VXCP9?0S^|&TsY=$vKq@F4%*tjB_TscpJ#aJJrKE4w>=@Ei^RUQ7 z+v=%rksq!J;pRW|bD_Jt|DNtZLA?*DanUQtbU-M&Y$Ap$B&6|hhg1O(|-rHM2_dX?S@NGF7#NN*wnO7Fdh z^p>cghzOx~2+|UIODG8>XYqdD*?WvL_Bdl->~nM06{PX-WR+R|^EXis%HfY5Jq}_& z75r>3sBfK|@NFj9C}-p;UCxxUjNyBmm&j)+S;LcBEfQe`KzSMataVNE_xnvKvkd2x zSnFs$qg$pw<;MuMhp|w;3hLv(RkDKcL%MYR!N1hAT3_M8sD+l}LkPla9lnrxUNnBhLK9${dLGdz5ewx856ARgWKVD@n_9|we zpY!SO3x^Coe;w)AIwQ>LXKk*7G$tZgS4_nx{m4s`XRPk=WGKiTr>NZ26Da<=aTP>a zb|q%Ce*2E>WDJ{rh#K@P_1O!FF z3xQDK-bwELWQd65ZX65=sZ#TDPcYD%Z?B>9HQlRE62vnkmP#xTLv*KvLtMR0`56tMHmQXbHdqCy&uTd}7Oeg>b*X~~?l>)q(fNi-FZ_g+y1b2j7Gm-pkI8~1*p$cLCs zz?xQ$V@b=(GBA9nl^vVQQy6Y0yMhu$R#_m$#Cf29K2qK2w6)2<;&N#t7UMr~W2(68 zu58F;ixxSSl_cdWAS=q{I9P}`pkaB|y{(r*MIGPh{WHVsQ;5W8=e+-{;RolNxZ6=# zW23{LG3Qc#`oWwKMP8F8D^+uIP4gfD*@s2CB~KLEslLMhvQl0^2xY%6L^Td7RFLMP z>CXQBo`4QMW#A|>F$^!ifA5}5z`|@o`}wL*4yTHrO*YHKt@X={=TI zjv@Ay^723URdRk2`-R$9t0eP8i+7k-=X%3*gR557!G##*>rdp7!Qa1s=cRv}JNHDM zOvc@R?=KL-KB)(N>F2l5)9`1k9=6NFf%JR{UyEV(9k9CdTK{Q*4%DxhnwmQO#l?#k zTU*--2FU5m_XGvID~`hOJ3F6*@jn;xYid8Me-`@BTWw|SYoHeLAq)O!81}+a`}NHW zxW>>K_8@%mOdFxXwCMxG^<|uq=Z2+`m&m{hm<^v|l&v710_y_>&?<8ndAdy@Jv8=I zY9_1Rh2x_mM>E9Ahc5##3MzO2=Zy}c&Z|o_27@#I{$IW|P zzwTuPM3g?aEG7NNLG8L10S!N{_9QaAp4}M|o2%|*I6NS*&=739Q}PQrBk8M!9u0ct zY1&HptnemwX&N|I4pxz~0ac@h%7amG3IX$a?k-&&9T63^6LfrzmR)M9vopTZ$J~v! zbNoW>6P-^kzux@~r(G`!cbL1zVS#*2c^-mB{BVL)Y;%Mvg@)5^Xg!OeqjeMCp19W; z$5C%w0j8ev!B2`n>qiQ)><>w}#2*|t`G{hDJO7lF@-58xU+4t~Af`SOKoR6V(SVR`x98P*V}bET&3)ZNSH&Yleq zQ@$bBik7rpD6Z~`q{7_bF)l3$@Us_7qhp^kFxO|fd|BdEhhZwm+??hob?(y@v0W~? zSvO_RJ~uVx)QpU9KvydiRhWpGw?065+SNn~Fmu4|#zHYT*v zB?z4>*Ed^tFY;MJ_cye%ujis`aA!8)6}!a24fbGZlJ4+ zUGxs>{`Qjs|6O@bA8dCig5u&WYAKbMJW0hd8)*)qG)@z($jA5;4g_H*FOE$* zJXGbfLywHi&@M)P&1A8(;OLD^flk5myBEMNP3v~;!u*_;ZKhh5%p$i|MM``;>cc&8 z9?p(v{(9)%+C$H!^a+7KY!7|DCu$)tFJP&@qU0Y{KJR<}INRI!zc!iT5<1+F)&j>9 zTzi>M2F}wp1GQC0)x)yMk z;lSca6NhBm(__n9HWBb%v)y49{#)wqEzjfTA8?vE+dsfM{%qMUG@+ynVNRyG=p^nj zxA5Tur!hY#n1hlJ{%x`d4=l980wD;0Wj{Yog^xqE3urhoDbwLr+|O|@3)0{7kklTG z$xh!&Akf*%A0L>_F)8@$d|-v*))XLP6%4W&@-Em51$+>q{<@kG6Xqt4g8ZA&`fT$XeM_m zbANosb-v^%3~jMyMzdzR#E(%_Jo{pnjE754D!M}>bR@y8*-=?h@l5A-b5s;^w1S~Q zz+o7Bn70aPEC5PsoZpyLbw_6w$+HKN{CnflXQ4*^vYuzDA<(l=67*NQw$uGsenuuQ zmC;^-3N=&oTnbPLemfEulu9~vYA%UYk4Jy<+y`E|fQ_%fN;qe_$r!$6JT*VsDeQOyc%dWY=KpZh$l zJKrVepRIg9h}FI`siOQoZ1pnzV2z6JfKKcUAsH#|)Y=CrM{X?wcO`yJPPVLkh+~t& zVP{^+-=heY0(qoh=>exYEt$PvDEUBQPf8X)c@Z!r@k(ep*i-6keds0T!0X2v#`Wye z&FF^^0Y_EiSzgJWPu})fa&4-VT?=1|vI3`U>fa@=^iJB-Vkc8fGX>0~$aQwZYeo%; zeEB(>^%Dvj2D267=4BomEs!=<3{Z7#3-o+N3LI`vRv`>`=6x-K_B>iDIEgI zO6hRGRf^z|6CvkR{~bE{F{A}LHm&}CAZ{CD5y&OYg9*iCvIP#`9GH3&62d6EtLwg~ z&D~hD%A+N~$ET*Q&gnW4@z-DyZUGZ@)CY)G8Y-$1v4rG|7O&-nR<*Lac!5BN_MaLx z8>#!)c`d5byAx%CLYEZm7cq;SKRQwALZFPURtzMoXMGAaGesO`gSF&{bECH1FLVY$ zy*+8i8ImqX?>iY|LiH$G$PR}+K2Q5zEY^0sES%P;EGs=I&L}F?F=E&y+#QB~UBO+y0u+2_K>>E~By$jcm`Hmg zV!8n>8H#Bv=uSb)S%+m+L9{)DJU*P}HuH6j-L_@wm&DsLVEW5IOJW1+OlY+~HPALHGCe@0BfXw^7?f zI(AdSs&L!PBYWBXKZ71rRApqtJ0$$ZbTfY=8Nrhk0w7{0KHYHGTqxw?}2D&?PKK#omydLZbxl-0{Sx ziuQ_URUS?m)kt+HrX;5vx7Z7sF>k}T zo10hYo?~a+^rX+4doPQa3$}x5UfL%@RaNzPZWXcwRsd>dESBt2x_8OA0&@i2+!$|@ zK9G;H|B`7kcL2uA6#nT2JSH*P6JkOcv2qSoObP2j5r7F4s_7sJ*S-w0rx| zBRF}`W7{GK?`^4_uzOas+0CkxmJLV)5Qlp$rF^EHDn{X%vQR&N{|Y0b!KS&a3N1%w zdiq?tUgr{zo3#abbCS5xRHP8@K)6z6g7iR>C>M0lu1=l3(_SG&+vO@5RP+nONE=-7 z6=~mfc1f#72`)*l#j)!7-tYw*OjOpj>37)J7e!jmA}g9~(j7eMT%dB`e6RJX;K@iA z4v!`E!1jJP&=dcQ7f=`J;jtZU5vZyj(%-L*5;W(urpL6S0gJ#-D6=c&Q$f*8r z7CelOm4LlN7hQ9jlCo7s(9}6!mlvBOe=H{#bkuQ9<$ew_h2O$*(FM0hxx2|G?sGr( zMt5>mOk|vSE7md8RNHm&`YTCTSSN@7TjhR11R;#O*vR)!>t902+!Ua|V6vcL>`Ot% z$V~tf1AG<4E=5Y^!kRQ>e`hb>V`FYVKQ84_SHYKXj=CJ#!9CuFC!TSl5PzX|$@@k1MHHVbnAFMX+?T;`G zve)W_{{+8ybWO9@@)6h+($otDktBQ3N=(vo8T2~U^E$BY)qjBN)D_J9!cvD+_f=Und>zM+o}&X7~;? z?1MGR$fzmryYY#KgHe#OEGo<-HQPK0kH4d`F8Dx5s2apYaG#c43A$yZgIx?4G96yl z!);uVt7K5qh5Ko;8Sz8Qn2(m0mPI5R&EZTcC)8HLvuhiTO?952nYZ2k*%c+ce{*s` zHp3a@l(WKnCw#Gee1dXX^B0eQt;~D}ECpvstE2FwyjR@3BL!(^6@p!Fu|q{kr)`1H zkBa}@(_?bbcY3>`%oV?KoDU9|R-0v=PH6^d`w_P379UwQMvQ58T#r`Sa1ePOC6aTc zlVOM8EQl7h9~ndmAZmTngqdQ!*nkMn6WeF$@w?F@k9_P)mtj8td+__xnvt*Dk-ev- zWsawnzE6ETV}jdf!)A&z&b)HiP}{tAX&w`{c(C{0Sn8DxG?Ii?)m39*bCE9nDW=llGB@3~nhFXz_q!b~M$i{`g?JBE=ucgdS z!{{|L;r$jJPZqy!fP zCqFP1`%Dwu#XQKTpk>Hk5f2DL{c|51zdwHcxc2vGg)`$^YQ(P#EC9WN*m7H&z*iYp z&{9$5P;m+iAE9NT=`{{-L`Q3dJ|rb2rKF(6tOLkFu;rS?MMjtD`V7Rv)6PUl)c^&) zWfLB_d>wi|$?MpWC10=uHJYMt>xjRP%PlQj<+xjMYp`L+zoCA=G>#`<4MfqNsfXyA zyUiRCIj&I#PMi46jVlV#=-`;K-8mem+t{uT2kmdj_Nw$VExDK(O|A@Az|jbHc3z<; zn5D%538~yK@Ayisck|$>mBD^#l~>KG_tqy9)nE-uY4tA503unG7qB@$FPm+F)JSU> zYUG^$2uw^HBggT0`D0V>X3e51Hw4%eUGIzzGlO5kC)Q{BM4cR--g{At`)O4xbh z<6^0V85%ljdWgyWkp)x zukP~y4U^7nomjdx;2=W1k%eyXg&5zgGjhe*nzzP)?9?-36TE80?>lfz;QJls zM}P2cf2G=HB&Qudk9Hb{;rDm*3h<@zx%PVLqYNYU0r{GFD&s9LbGSUV1^BxLvZ|2B zz5ZbZaXs;*<Y%N2K;8zw1&0&G!CuUIaPPN$mt6~2Vl?TPN%asnjNShA(c-%K-%5r5^YrbwwQ2k9S7H-@_Rd!&F*{ z-@-k%sm_I1F1bEbjw_b?@aAOs*!GX+NBZ0xab+@sH3Am1a`=8{;qTTfdhQmnRf0UzEtZVeR>Lliq63T!k>oy_5>^tLGibuOo616GZ zrO(rGu#3J@+^;k%(@ghR#i=mLsh64KEl4sebD}7e6A^Vbpc+-ujym$|FI}Wb8)=@@(U9YZGusg;RLcyLc`D)sI`=dUpv8`6(THNT$?wabJ^)@zkRH2IV%n~$vTK<^_~VdseZKF1 zlH%n^>PP-7pOk6`3`Lbjs}K9-H4MHsah`mKdfoW^*W^ION#C_gG*nGNJNdJ-UtK4C z&0D{EkD5PRZg#c^T93uW#6)8P0vx-4cVD1*RQ3jhLEJE-i+rf`Na#*iC)P6S_>mYl zA=BTpDofaZn8Ue)A(GFal10wtZpGHy@NgQI2eVzB9I`$~ex^@OJ7#BR&w>q$rxZpc z9TVh^7v@V!O04L(>`|HPeYaBewI}MF#b0%7c~~n;bG*+?C>dPlF_YGAWmrAfrl+!} zXO!Q+Bs)`3CS|*mzmE!Z#w)e{Y#b)z<%=U2D9Y-)1V+ydz7_ zDscQ>Zs#iI^&em?;Ar1XSrO;6=g-n1ocg382sx*r!?JGnM)HjIo$%3ZMh))4EN3!e zwbIC1CBE=%+4g+rv)1=MvIwt@28@jU>|e}0*SGs zSk~)ImuKhaJ=aC)aZ<9dL%Zn1eLeV}=3O5`Ev|xH+~?1m^BeT^FP)560O5W)@*F|M z9@X6DS;QjM=5`LUM{(bVCJ*!Y46F-$c-|3S!LIwi>MKg93 z!#&uxyOCS=n0&*GtJw-!W$xLvSx?_XK2OJGopUM4bp$i0Erji~Hml`1gZBd!i@}w=o?pd+7apVG@$| zm=A*MBZtt*`}okjauodL4SLrA+ZPy?Z&32b5$qjg+~CzDoqMEhfJ{4g_WZOcu@|^m z@p9`4)}3<4YiirtX!2H~;_XketD5pEFTV0m)*o-Sa75&VV(LasJy&#Pt*?@{22oc_ z7J8-a35NuRYYe30eW?J3N9juPec>FiR^KXfkwkL!5^4-a1Xw2wBa-#d{}9qfB(*kq zV(lL*+~F_b57_3aewf3h22bj?eQ2*<KYpW^aVX@0UH?O^R{WAPZoJ-Y znzT>AixlWWjS0r4M~@t{zxd(?ZYjH@7*0ax)eedwgTK7{6 zMvdbK>fH&oj~~CTl~d->(=|QX=E)KS4)Vds@V0gwv$XQsLJhBR;Rf1jSrQbt|0@?xW<3=M~y>%J~Et!?Fl@c-Vsj(<|1dDv+Xi5@L=l-;0TufN8B6yQtm(%$KI}o z_0p4$t9q8#-$&_SZxBVZ)49nfM?5M6j849!XUVodD7Q-&UP)#w-niSE>cm$3`D$cJ zk-aMa=#V!^S&Z~;ATWrdH>WE`5=j+Kt{6qHJGAdeIbAMjXtNH!A z+xZgbuwB2^XbS}C-?`RQ*9IB(3kc9R`2x6^N%`aPy-xel7LpLXKVC zPSv`r(uxb~>l)P0MDIiDq{5?%HKsX=u^=e$;FAVUS7|u1uFb$(=GO56w$QEI@!&1k z>S2McYR{fIf6{~DgDkdJi5#%oaVi5hfu}qOY4ZPm;IX@ew%GqI(X``cfslq|jkoxs zcrpV3+ROB6Sp_L>kcE;L1T+zAl<=a0;p#_i)11w@^)4?xir2M=n+Nk%#st`V&8K>M zl8n5|#Ah0F;O#)x(J&QNd4pI~JEh(m9G_2%m!$0&t<2l5ATtHpqcuk}{zmZe9?bh}ZKpAe2 zHq+#V(OEQD-f_$bJPxE|m%DG?J@FjeIWAjUW=wBFvQ1z1Yn3=qY;*J_SCHaS%_Cs3 zoQz<*6IuB`qc8tILiM^eu?kqLc+9_^a5Z@tmS=b4Bl`(<0E}$OJne`27XUl-pW$YB z>)Km2W-RkRzh%dIbaLm(R07Z8WPA?Kvnxv$wGVx#lJKv0PqES{=O2og+_Yf|!~FX% zey#A>V7?HU_aN%_i3$VgDEOhAP9xilsShs?+`iT6!SwFmZ@wJ+L@e`8PpKqjPK5IY zpmm&#?8ttRr`^{1O``CdwVhUJ*S-HG!f>aza{cVj8Xc7?e_03T`q@Jax@z>#JA%bx z&vCH8pLXAkR8@_(oG&l``iv>r(nL}HqC(V-}nYT5_QpUG-nSyzFQUl$aoPTT*h;j^S(1EDM+x?!MXy zLP|hSI{lg*VHPn^Z zNg_c-`SUc=A4qZ+(hF{>s@juDL^_YIjcI5s8#ERPf2r!Y{a1nLyFRfLxD)(!wky%? z0(JDSKYzxKH}?)22fdH_Y1yQvy#`&qHe#4-{xIwkH`~lz_44whFChceiNy}7Q*j*f zUtA}-4a}ctYVJi`$>X}Zywt;bwKubA`6sisw$Kly(AAL&RrP0=vvOd%>c+-jcIm(g zV?9^*LZ;3fH%2Nu#u2iAuRUE%}6 z^Cd4jm-+a4?-xTeO61yz@$3)H%kb)&nuOzhiyk}r3<>+awrwi=8>1wAY4Kf?wt%zU zUtSbVt*v6RlDyr1L|K4DC)VoQ<8^M0LWm0|OxMclD9hPkyj(x@$>CPJR^_oXzBs|U zz5TWYeD|<)GcybRmF*$m!FlqTtpoYR3vM*iRhGihWw24!dQZxqMgd%eyV}*J`Vx~R zz}=wE(J9%;Ab&5$-+@4fo7#(p0A;c#E> z&$tV`ig@SVp$m!~26TrYvKPd{ztLoWism)h#n?jMSlITrhKL*pQv^UkK1`w!Ml5cF z`t~3sVB~RD9C@L#%E)}^00LQ_VU12oN)X^6^ zW15spbh&tVMvkur;8#rajhij!Edtttet%$fT$(~cP;nV1N0g^KV=_K_^7K@mhV4Ys z7iCI#h)7F68)|X=d4G4gYKcTcYMI$I3C<;;fol=9yfjVccFk9>kTU$HarhEkH z$6%3GUwF(7(R$CKO0VOF%`x?dd&S!G;ENc693&!!Wuf@!sHnmk5RPr9&dNP>U!Txh zUM>kaZ?TSHyvzYN=1$Z|^@0JF5a}4|yE&n6YN{CG5DcIF7E&!j(~HF90E67lKbI$mp_X3Bb@!qJu3!1g*75$fAoF_W08jFpy zx{>SSf_{Y?yA=>pm@=C^o1>uIcUh;Cqo=2*@C85xVqfT)3|kt+Khu0>iX>@mk2@S( zqK+@$=lU3O>jEvUkNyQc-uQymA#lHz@#>=RVXfms zTj8c*rxMqUdy#L=n&ucYK`KKnZ@Jb2d8*goS@*~Dv}*<84jwzpMHZ%YPRu>;a2trR z>G@7GHON-$Ru|7E2x#_;sAY-WycwbV4AbuTUdGGaVz{9fg(A;D+4F6`9uxQbS`Y{l zZ^6n?3H_q2Ee)Z>tVb7#&BSeapZ%^ZDR$?@Lf>i3uC;%YLelEahwCluc* z4V+Hd73g_trCv%LzsQ&7rc+yPpzlNCeH`-b5nRnyn9u^Vsnx9Vtf%JZ<-G=2^z|Pd z#~*NV#vqD=j?ed(=4n?k+T*d~lY@h5t*&*(KSZA1_@*Ub;82(O<40&+#lxhmES16R z-COM{F+KwN&zpMf65FmgE{A%LsM+qdiNS6j-z4BR7^eg}+S}(#%K&~Q?-dAY&bTsO zZFc`VD6$dd*he}#dszEH3Qkh6pVJ(`OTtzf`yDvJ46N+yzDp5IeE&2btZJ#yc?qfQ zBeuuxn)XJ?n8xU=oEMl%kW%&<6_CXR95;C~4`;?UjF-yPXxvU3CaXAqduE6zT%9t7 zQ`JQ}aU^CY>v+qS_12BM4E(n5?&&I|dU15YH{{xRT(DNWORM z%*>0eYSi!B0`RVVYQ-AF;*Uc^3%sZsiWV7uX$W&;_G<>jg)VEyDK9@D zMxf=Gu-NAi{S?E+r35y(VJF$hJXOq^(Z0{>NaBo~`^|cL(2&s|Ths3#W#4;LTue(7 zKqZS(oYKpdNW#Rz)2Dp>fg(5B<^aQm5?dyJR-yL5ghBShfCqpE&Fzc z$#QfR;!BH*z_f8UH~}Dkg@FYhwBMd0ZZ3cPsmNl%UrtnZv1JqKWq#LCPf6);Up~;k zJ$5k5e}&69Rt@!(xZO$aUp8rQyW}>y(Z$0sXgrP>lP%KBC?*tp!>ckSJ?rD+EmT#v zQ$>+H+&@Q&Dh)2Bsd?pH$;O8!lfU_r?qZ6%6GFWT7iA1=0Bgzob%tc^G7l^MhU*M?9# zxy&KuzjLrM+`SsKGcj@~+%JCEmvij7*_@U1xwgKEh-PaKi!0uI{Qb+Ky!^rJe8Wrt z<~iPT0!{HwDWCKa;L<6P$gmu9>JipsXXnmao~uu-M7}Nk z8PgV~zm6d8?K%xQ;?+xZS=eAG-TA*cg$4cRj&s2EzUte_Hm!WXgj>*Y6ftfyhue+< zXqz$H2|O{l(Ojp-@IIr~wexSPG-3e`MSdm5gt0S;MS;cfms&9|0m*n-ib}>}lB4glekQ~T_*_F8LKw7RMsDiRSA1Ox=Cg1odQ1O!YS1O${P0u*>pFV#6R1Ox_z zg0zIT_pkFDSaa-U;@&GNm9+(vik6hUOYn!CpbpZq4)T+Ox0wT<*PK6ecVKgq)R5kz zU}5to_v7bJrV*Z{Vknl6fN{l3*W1g@WVNj~GP#|#BykKFiT~LE)0x*BSzOza9vYgHQys7V zWnyAN)ju`mN2UrzlttaDL!g+m5HWE#Lv6V{4l#ZrF7|((K7{6f;st=ij8~-<$!i&b zdb#ljHHv>C4+m*za{bBcO3WWLrxPu$NO2VvO9#69y{`(~?gT!K<$Zo(oanwd3x)bP znZ(O^jgh7*WNh<$C}xb<&k>gg@oJvNll)bkl z$g~NB5)=quAr7H}tE+jsGPASOv(wcma7#nLm#Ph(D@$MC?oZlo&xR;U(|w8?k_Qqd zt9m|85=s5gFhQrpW^*lfm{Zew+{qHk2;qtR+Swft$6k!Pl^g7edGD1^pLg-op8XJ0ny zMd*o?3S&Q$+cT_`H{-_-Leg||OUnb3GPIQMzGC{Ofp4IyDPZJ20*Q2d#KRdU-<62} z`_bAYc+_Vdl>U<>iakS+EGB!$=JrOM(<>C%@mWo_pQoqj3Z`RXX-LrM=x3bQia9>~ zp3KRb{ZGog$b}VE6{wS1(^R$dWUu(By3#LFhg-_Q+(EIFy75V$0WY!OZ+n?1}_|lM*Lz3Ya z@cd94|9M`>SE6Y*+JJv2W)$!+oI~Dis_2I-k^Js@6Jl_OPwa{>n&)iqnEEpf%~Y1Jd6<5-pNy*LauJ|8Lb$G)@c<|6y z0fDu$QY!+T!yFRQn3%8^SXk+{x@YHV_?4BvYtdDN7q>CG@KZ99;V<5dN? zzp_rLzP!A&AF>fXOJ5^pX0sI*B6exlse~@W*hrY#ec)yZ2^nR_O*j;&%M+ld@_r8` z<|!MWTEg2h_mzcffJ;??O*V8Y@n`2dVaw; z*U<-5QF&=cf-c~+w6Zypw3j7_1}I}xy@naow0_A?bshbJdtUUgZ1#~t-Nb!xIXJG& zPLhR4M8YN7~e*Ql*`EusE*5sGAi3Sk%n0xHJmtp7Qp&?5fkdOfP(#kcI! zP&(_^?g_B5w@gm0J{LIkZvA+cLTCI zNWVWEVp{%u_v5;|KhYX~yjl$1djEFgx<&}=?c*Cq$kA$6dr7gTiMB#jaQ%1Q#t)rD z2nZ_75(d~=`~mJ~*A-gqVk)HtP76c4-XdB?B$5cu46Xi9uT&_+-_;2Gz4iU$QDig? zT8M8}ZkVp~*VAfxm{SDWIhW%e!P*5}`$6s}n)6C~sH$cQ>xEcu1vAR^*m1)SYvhwd zNjKZa>)Vk9rZmsM`g~}c&)XJsb^ZCDWE&fm>~D5DR5wC`YuE>p4gS3LbJy8cd9bf< z{bc3T#LegV=)9He?1UWO#A8e2p`hR2{X50QBcI4H39<3W5Yeiz%cjo+s~oEixys_5 zXB!a@kb$YcTrc-hw?Efxqc$Ks8uXr6lhz_0Ld)Am)QbwT#R6V8wxH!QX@2e9sU0@j zE&kg2)o~oVx3}k<4%}+9H1eKfT6R{;d5B&jw#&`9zQz5wwWlTq*kI3yxyVdShQ6}D z6YJQ;%s6_ySVUbT_r+FZ6W+1On;_(Kr5ENL<>+G(Oy7?d=U=~@-o75%oDBEr2n2zo z7_|idu$gw<9If$rOqTPi*h?Ytyw3Y@T{xz4c|<1UfWG`A1d5Ma$B!fs%@sJ>w-bd- zxIYX#mREaSicu*RWm3{j8<*qbT+dE0mjkHL#?kMy`Cg_>FJ% z9ZaXq?$3v(9<46+g40Aiwv735jk@Y4o<_1nvPenn7Q1y;Yo@X$g5H}vmW<6mj`nn^ z!30xX0e6?LET5tQuW^~haD5&<29E+=KBpT)580nON{^BAw2zmw;al%thtR>8fUg=f zw6&E$r~wz0wNhkH_k7EJ|B}89nt1=KuOOa0_ERAakQQIvlbUNtYvs9jAkThJB+}uD zgmNG13u6kIsf>r6hw(_o`GwnIWDB|-MN0jAz0pB;d3$+e9r={=o4L32{T)6gpVOw0 z+b(B8sH<0$yazOHKau!y!UZ#dtP^|V=i{=V$cXdG?bUGsbdISXjsS^jUakqfOziYN zpE*jtqOxKH+{bo&U~r?vRhV^`M(9pYz{}%8n^K-XsKgrRwA?;anb!kZ9npq86;f(* zQrN~-rk2rjtkfi7ARz&oN!HkwU=8cpX(18wr3XSpF-v%S==uLC_e)3HnP!fprsCtP zm8f;Px88&hE4`m-x6(Qc4PDe#p*&xzfCfS!pxz)x->Z=;lOdBCHZf#T#D!o$GN-afDb_L zb}CkYWrgQ2JU_C2H&0`x<>&CbnmC1wQ%>RvCnF1UF>&IzW|}hwMiUuac*97qhG8I7 zu=he#FooAvC-7=}^Q|DA#=S^a6xQ0>KF|m;1hXkp0B(JyUmP;^AU$t{kJrEoJT&AXlg`4 zt}DF(T!#UkM_W?9Vyy)kNBxJzH`J6q?jhq((hhh^_J7ON660b@yolCI?Rbk(&z6l| z7QYHMxmQZCeOL_rgarCl(m)fLstfG6{Ts|IG5hHK^;VflDZ1q$K3iY!-(U^LebtrI zE?$q%@^QFPQU$>n0oOs6p_Bo?zPi%ryon{6JRENf+GKk}Bcnlq$aQ~k zW-4&4xQMj2Q13ClX72fI*y42OIf-4sDVbTOV~cbw1?&!v1IcQ@&VDx_lhD=!zv}Vp$yR(dA;}H9&Fh3CH;_jVj=IxiFH4v z86fW4L^?b|64#j^i08do!n7nQS zxI@cB)lu;zI4A=K%Gd~n2! z&E9pn)tpv=${FlHI8%0>pM_&SYWCRX={30n>guyqS@=dKjAmiTA%VQs`N-$zXEadD z{#kSqEA~7 zXYtBI%N(~{<=oj7{g(?!ziSn|fHiLlPboi9Ro-DpV}2(P4RaD~bX2epmBqw4hS%iL zA})qBn@nr!(Wud@URl$helf2<1S$GZRcm*uK+m29=1nffX+p#0m>4bDR_ZH@i>-8Wr$;7$Wt2`u@cS7p!AY8w`<8j*AO$b~`F8Sz^dA zE_hWb2=}Y+tJ^GQze`a$AxA*yOlcw3^2czORZSRzf@{;+Je%uPQ+i7HfG*gmeX+n6 z35`3J4k@u5mCF90VRH%8d%agEM;2T2fhE)F`K#52h&~ zH!eFic4=U((E;v|>a=6YVw(c85wI1u6&A6-eh{1aKu*GDUccG>)>EFr?@LE=y}2Aq zO4^vX-O$_!b4U3gLQSpZ*7z||fqm`}kMY6avn)`o{BeW zG=Pigo%n($CY%`j-5)z4{QmFUZq~@Hq5%eN$|}t>V@G)tr6L;K`7bzKxg~cGpLm!N ze@>5ZQ+jiLq?Tkx=EDY)L75>TBS~z9Br5YLCq%3vlK*HGj5vT{8%&?7tqe$zxZ6xNdje_VYeo zVeodDIFpKo<-_kn6ufqD^u;t$li(o|0w5Hiw!X1#1OGDH?c9q3L36K@pQ8RG{lDs9 zhSSj;jW7kDy_d8Ljs4rcK^|;G_~GJJPOm{#_+~o&2^b60m^j#H7`u^n4G&@5g@r8{ zUXo0fxtJc+J`5~rwA*2{+baHX2vR$o_*O|XrWg3sw@ut)((c<(emptYIgA>5G@Yw@ zZ2ow!u+Uy=Xp5HH&OKRu7a?OEGMddEk}7 zNb#aJucxFE4##}*hadFo{S5Ov4p(VhS$fOjnn7q$gaBK(YnV`cQb8f#3XO!x-y^5qPdF_Fdl z;{-1uDgl!3`IlN+|N1vUtlfnDk<* zznB*;$!BZN*Lm(Y=lG!)7nk0>_#c~{U?Zk7G+KmGU?3n^1|2>d=)+XQ^OipiOCi8n zmd#Q61mR5*&kyV^Ld|b?>g+kZ1OApxd?@w229wpnFpe~%RvEU~8oD->#`obMVKEjE z`w0>ARQZ3^y!8^RX%Bc~wY9G{;x+0tR}IQzvml`06PNmYI=pbXVUy^O$laMwwBhtS zj_7;HC+7FK#PT=4n>xUr@IdH5Cy>#04tND#95(fE4pDmx9$%EBp||Ia9uw3-8F%}h ztxXC6w0Oj&LL<@)Wdwp`Y0fE`by!DVm^7K1y%JHpz@ zq$dDmP5ETmhfF?FfdJZSC2?Qg8c;YZ$-p>P$OzVd3XiD=A5{kI##zCvTXA zI1{AgzZS7~$kQrVDbS9^RR-oW|!I$r7VL@uf-uk7mK$-0e= z7%o|v^m)5kPYJ!fAcWzb@l>6{8_%icuu{Ljmr880`$JbZEy=_jKTp&L1&BgFTtpj4 z@K{)E1xMUhU}AJUszCAn(co6QL_n!dgl*dW>c858o);$B=Y&B5gM_Z78CDv4!_SqO zn4Fm-;n&b)e zf_6o685>sR3Zrrv2Y2+mn4vr{@)8j+zg-oO9|$`VyYym$6Z)3U$T-8v{JI@eL^rL0 z$I(>CX)$sAVC1O|ds0=B$0^kPmY6$X)EEcgqkvzchtE&LG3Q57ChLNWpUr|$w+ENO zcg;C(`AWBUkA@wm@Mg|xr*!4F<|B|UK2v2%eMyj67GORs(D+8o&^UsL&(8RBtjQ=V zJ$xBy`t~gEOhPIkH3g+Ronc=Vo%DYA9CoT5a{$g^qdvJD!m@0FmV1~3ba!RiQv}6^nlvX1YLcx ze;2e5W1X+KVW5@18ru5qydBJ$5-mI#N*Uuk=yaC=3Pw{{aX=x;1WJFj(a#4TtnZ?byC{_4VEyn8;xluy>{ zg#l$fY4+9oi!4SHKARaLDPjT97nGWxo1df>Wa?tAtQyIX!FKPBi4}xpS>mc8x59Sk zqt#~tJ>UJe!SHAiO~2>BZ0rgtWarSaBR(d@8#LD>fShL9?TdYxo0$qLQ@9b{H#>{w zuJqdQ{V&p5AQbKHMaLsDAy>WeFOn03A``(zOIk$VdED(!)rU(8s`ux&Q!vGXAxN;$ zga1KVBL;kwv~Fv6{or+jVnNvRfNspD%&%8|HfN;?>c=hbw???O&0__j@>_8*>gD7>S6!^JrCI;$0+!`2x4G-7$ ze7M@ovBNjJU3_YcfekwO*bjG(Nba+GAm7mR_no?SAahayn8GPGdT*rK&9d)Hy`R`)Q@_3?)nHnCq24FUylY%-<#fuDJ zpB4*SNkuS3)^%S@2%Sto26$gv$ZacLQEEQk`8zROotq%nH-H8~io)eZ`PI65mSb)O@b^A&gu4Q8k&!q7< zm7jC=p-qyMdA+QuOk%R>)ozydtgo+~#=zD;?O-;E*i4yOJGHk7%x`-+n)rcMOil7~ zvi0WWOnV+mu{BvH^f@=9mOjMQM1eJf%CJ3I$cmEF9GlU6ruL~|O2C@-<#uopz}bLBk9gsGc#`@g8pA5GCPY%7r2*Qy*>L?EOBbhS zIrry}){i(iWytBZ6{HY}e^Z2nPBN@~kG80JbV%MUl@s}%3Cr!muKC^hRU}KCOj_M@ ze5mM8#XfKN&dWp-^F-Y=VGpapv1!~DkS#EtBOr@|<53E{d#?Do)2_kcGEp!j1^<{M zYR9al^uD{i7~8Ec0qoVOIa``3ZgE`w_0rk%(Ijz^A-3#T;07V8&5qkt;p7Jr`i&QX zk_(&bv>KC_ z;g`kZlph#jQ=%UIGe{RDU4gMGamg@{Nu(+3KhQtRTz>fcMd`TvkYeCk!lkzHrFY`@ zmt2zSL&z;tkM2)jM7Ic7TVMA@Ban$g+^S-FLeeP&8euhLvCTK%)}G8%Vw|1YLx_bi z0sB04RLHMWXZgWSM)RlWn-j64^wuzeaytG{O%-tsq8SUOmXALS-(bnmC;y0;<5?l~ z;ruP}Gedh*X+-h%oNTq>p6JCzCld>g_}M%hoF#{gzx9x#U>Y^P3Ur_O z`n@A8Om9M^3Tc|I0xd{tKNeRkpy_rQ?ZPmayhx^D*ok$@Fp-3})l6D>qO{ss@*?f-rzZNKk9f{jCRVBbY4uTW&{hlU0yQy3L-e&gkL912{tj0>f^`(o1z+yFE;e zlH2)79X}}?Qms?tq~``VSC|m2U7=m8pKCV}!uSQ1qFM73g_irh7OIKI(2f}Qe{ZV` zZ1_`kz{osBS77Cst=lGa%r!@$sDF8sd#uDTvAr=%NKs=mtww1w@4+j&9-XsprQ z|5aoLnAsX7lL}(>v%9)3(g_eW9lB0RGNzigyM|`zB9;7cDcbR@&VIVw(otb2o$Eo@ zqGTo)@bSsoS7Op95cG2Bu)T}~1&CUq_ zaPYAGHs_<16rEeZ>*v9fyPet@?2+vX_kNTKNXeo&I~{}~!d*=S)GE|?5YON7F58Z! zDWV#<=}~{P=4PE53|(C?uyf&5pm~_hdRP{ZQ>>;1D@PAs2#RxR(^(zXiXpkAlXZ^~ znTp%T&iF$i|kuUuSi213E@ zk5JNJq8W1Qfd)y~h%iZ|#_12dO?mjji2#5)-M``mdKuzDzt7KO(`y6o2FP>XP;`(z z#98{4v>^{zQD#n9pS=%D7kd^^@A>?xP_wm8Q=e@g8VQk9IYME zi*+PJ^;A*r?n{8P4cEBLYCX|PZLD%YrgR{=BeQJDBg;&OU;02tOQZdKp|WEVVw5a0 zUSOB+hi++bM08%_a;L#O+WCt7G_n7YK%zG6F&Hx#xB3Isd-ynTi56;=5_@N3q7hqU z8Z_q@r_4G_A4c0j0ftX+Z(#YeLZ^S8OL3Ky&tcW)V3BXhR(W~>RefA3Z6 zpoD#~d>@7{78>I^JKs0n->U;ZYuZfBbagA#1t9}O4V^K|S!k;bHNknUj5__4N>1)y zuB#Sb@70ag1LjxJtz6|)ex~)AY~4+49g}Q0Mwb8yNVklGRy*<&F&Wcc4Xg^g5;D2z ztN`UostSs)xq@yk{VTf**6^v8bw0_mjlYmYKe>)rt8!qQbHSCq|I5X6m!JE|LJg01 zPo|V5eeccfMS32Gr6H$+3Y!`l#+U?mCV)50+OmpeQ;JDM*qH94`;SheQoe_weAxyZ|17F;2Yok^yngWEpz%_(2()F^}JYz z84Zn|jAmkvs2o!mg`%&IFFPHEq*4`A5WBl4eB!S#^nw`CwI-!E7wSmfZ!jUH@9~BT zOeWf$r{3ah`MeKkx$b=%eD+<&t=<=}7k&WTjpY%oyFGxA*4&ste!D-+;^G_+zmPzA ze0096j>hW`(x&+QM|0_D?ij{up5eb=Qa9y@V>K$O0mJdEP0SbM_PmucT4!JVu;x2kp zCKBT0y*x+|f#*iQT(4Sx7bvlwFzX61TF%l41615krvb+|p!vSS?ZEvFiUE9on}TLP zo0FITGe^yy`@Bt6ZTv!y7pw7G!U3=Tr|TCy1@RpoPZHa?CKW2SpyMEpmdjLk!^GUM z8SI<&COZWq)7_KHEkEktW<5SO@5g-ObtOa*=E%zPP)&fWDtf7oG+`k{TBoM#|`5=Z9L<90|FcN1+;d+LFQbtFO<| zPN&x@s*FYMF{YmWa{0!8z=U!L1?hG>aaV`H;$FNIJ~0m3@QX=l~<)ZWnvnSf6nB^;$=a zjoSF!z=hM^BEC}gTR4tGa4?ZdYm!C`+WMU=& zyBN=Eqi0{g+3d)~yi!@WAm4`1nH%a6H8T-lR1Ml=bMtqD*49In-RC45wqptMnK4xI zjXXjO@(fN^u8Fom^?w=r?g%=in9i8q66)0p zl5zpBH_V5$jqEAA%!0fLZ9N*H6yDX!*@mR(EwqA?W`IIoP+cVrZ;^|*NGH!U(za;>5`d6pV(_7+>%PQro`n8U>9 z6s#^;`|)uTIWIog*+-y-v{c{J15}G;NUz&@E83TD-JynLoRl5xMLIZQWh5=YncA`c>FLe_sb2j$jU~6tgB#yN zPLLVCzz~(>z6vb+1P`K*p?{kyY-?~Yc|5LYXng(VXw~+Zm^yFG_Kk7)&RWd|?Hehl zHIYU|MrLNs3-I9gOPwO?P!Hqi-Rj<9tGnFrU>sz3*LOo-PV8 zUk(GFm+d-WQH|*s>U!X+Dmn)8FZ7)m*lBVySo`{NR$wWXkDE41;H^JLMHsBSUrq>U@-Np<+y+*%+=eC^w>h|ZIAeMw6;fo)YzMa znk&}eV;hYr4Vadei}(~y5|oaNhjG|{A9l(R_4A)qD0*Zj9p`nV_YT`N3;aE zg8MjSAt7QeN2Te&yLs~H+MoAk{(nOfk0!IwKE9d+iW%DEueLo+*0%*#ucPRiwqHO{ zqJ-TVd0mdsbb?Jra3%}mcatHvpQM71s{ZN|iwGyjK48m5CMzs^s0I=xSLfEG&{O7e zaJSp!a3F-INRZMZIuZ$3DRSFW$pn&hR_NIC0&iaYeG^;saxREJ zVM@C684Ef^CS;pkqZ8G>O_c{9At7C($42?zKNF^~7H0i9>fHq!o93wV3jClWPctw> z&1gynhv}|g{F7FCHS#@|N&`J-r8dQhi>Zo(na}?Ho3Xj(FN8{|We>`HT-to=>hLO< zUaP-+nm=L<7xly3xtebA5+dA zSiOz1C*=4K3iJAL$WqT~BOxlThWL29%BZ5`j@N0^phQ8$^Od<816&FxZUutvK~3R3 z)26K`0Jw`AG9g#_KhjVvFR|D1HgqF@s3SRTW`aY?cmS9~V8;_(0V6dkHgkATTag(v#r<#aCkjW$;^K}xy^%f@=)21($PgA776ydo|y8`TW z&>@y!ggw@Ie?ytv%Y4ImHeK-bw=H@k^~DYpzJEcs#yHd%fb+aqx%BZa4sXjh-Gnev zD$y!uDS-eCvoZC+B+_&DED#mwupko9Cu{qXn~0*M=U1lK^Igp$w?K@iO*k(V?}GIt z5>k?UpYyGf(4~gnOXy~SSLt%w{Lvg?zum0?Iy}4(R?@Y2BEcw8f^WDp&e4G6JNW~! zUo;-_UEuTWHS$17REO(aV!TG+E;oOP!t~(@BsKD2d^eX$Jo6Nq4`G7rIBj`*tZ=-$ zUzV9|&*${^JWz3;J4L$YzSJzwalhJAyCvR<47T_7$+4&{&rYTUXnifE)Z+r{?0KMwYhHXc9`ts`1Sa*f1m(HFe~L8*tz#jX%T){AQ@ z$VWi=tKh>Dvs$NX4Ix*s+!aZ%k6uMpF+D)lD}&NZ-1GsskwHxLM-;^a0Qd?x1B1vD zPsWNcW2C{9dBlv0gvD-Q36rOTsY(^o3=!CiXUz3$coROc=j~)t`<>db zr65b#9|tN8r}`--T1lNt1&we9UX;&yJMTJJ6TZCN&=M@Ho&Xy)2(s50VLLt56K!8G z;x@5|QHtOtL+84fU?*Bl72iSiwrRBXa@4byJcbkV`)VfoT%@Zy1mKc{K zxSllBC$9hNc9+hu%!C6I7k~QmovH5C_5&B@82u=)ruif;p*9?Ad!35%ne76vKat&-h^;jogW zgi=;!!>a%8u{fy&=_2QbxOqAOo-A~M{N(O$oQYw^|1lq^rPt*1P#wQtiuiLGq2Je( z&)G+%hTh^vNR6nT5t+dkUQGDpRSglFPu-by-=>cX4(s$&2i`0uRZ97wgrjov5L~$DTi#FOYa(WYrF1!dHbFSoO zjhJK`xBq?8?EDxB2sLYKknoQ49#Bm;Wt_on$x1+!A!IXU;(vUsAjzg)CuaC8wpWdf zpV#`{k#``dD=WPd?NCdl_>M13sVChjlN7;0G3kGkf13ryLR!EBa5;cFUn*P}CYoINfdbLvLJUgK zgXgJM+0d@&r!*A@f&#eE^L?Yv8AFv15P?g01sv3>pEuDGH`-nYqZ4o-x})$QK2P_w zNSD)&gz`QQI9D+pRE+Bp@H(v`;^H9@n3z#;&`q13i+CH4Hn5AdgzE9SUE1WURkENz zeEIjX!=Y7?+WqkT;xGr2n$~Hho{L%v=)u6w_;Gz^CIS~%xuMUgj2GDE3YIq(zxIkE z`oK{k+g?ex{g*(Rx7fGwf0}1(+F=dj6t;wvS2B4+g-6EJsBqZ)<2AZsw_~AeG>@)W zswtcIv|8%e5!Lc@0h?|Sw;xPpA&0L~FN6-c9cP>zv6(3Orw;y>-g!ds!$dThOX8R> zkHWWt`mp@m#RkjfDsC~T3Khn%LlJqa5ErflJbI`sbj<#0i?s$rA@qlhIJFAq=8f3E!IJ<9UR3+G?A=^U!7 zH#dHn{Q8yu85}KU&h7ecwe&tDUGmk+Zuj!= z^z?j?XZP^nP{f}IFxnAtByNX(|J3xSxxT(0Cw9odV71~T(@HV&JJK^;l_Wf?oP~MA zst5G@H%^8Hwjuyh`uNSwt-F^OCSG(uwPhTs5yDgT{_f%4KZIH`f7?GUY?*z@jId4e z?!bt+vQ(tjrk=bOdHM)tMmN!hU~MpAWtRTdvT9N`p-8k8r8@2|?PE=LnFGi=eTgYjZcE_lPl{9cH1 z=KRIbR_Bx$sJ5fS)APeqD9#A|F9}NW3v2&$io0u>*^chqT}_l zYmI10&_>o95?adS6b*g9x&LOpizm{_xf?Zt!0lfP&pkw`0Ei0+_*|!a4eS)b^0b z;2{Jm;e7Ky1mVmPTYZDu>4FP6mp?n#S0oi;S|ErORzx-o}6V+e)4}>^5k{E zJ!EU)^31+ucoNREw7RNs@Pps$U0v}&SJ~ic^KiDgIkG;#JTgBu9PwHqPv7bGwEHr? z43?cA(C{zBF)V}A!dB~|Vz*TYWMO=qN5`ftmKZ*E_W-OsO0*J)NIr>u`J4#134g$0 z*4Q{YHpanAMDV{ogxVPYw?0AsZ#e%S?li;;UVvd{sh*rEM=7A^KouoN~K}Fvr{#gM7Ok=J_4FO4(vafB%1(g z86pgtGC@&$d#6)}`C=zP=xtjMc*B-vM?rmk{r;!U&W-W(sqI#ua3AqJQRr-LCnqaM z6JzFp8>;h_8)%=lwm~D9FU0bp`pmxHKhZRx?Pfc9VuvbI2H##_J$G168UDFiPA}%Yq#_|+-RuoD%+CIh_UNCQ zLc>bKAkWUpIoaD=D^WtXCtiPU&EQl68vrR+&;J!m@0mcePzp-|0E)K9*oCxQ?xWx# z#2VL#Lmj*(D^N<$1z!;l4^QZoj7=KA|X`U@r5)ST6k z6G?^S3E*rKRi{`_>?&kr!WVK|^a-2x&+uou{G5*!4qORuwk)*Y%oqq3bHT|fH>KQa zWnCzNOk{STXexo!NK%r_o-E&hFup`8k~K%rv!PBqrU)zg-+3FhUac_|Lipj2Nux$G z4u58UgPzOjj1_vQ3PdnJT209QU4NbTg3YY^n(AF3dkC=Bc+e{`pzpBMWS?FRz1DOA zGY0J->hHe2KHg!UZb~f|x7}!mdV>e~_wdT2>YFA*ur?f@v$2YTYD`Q5MI}KlqVwrL z^lC>(a*vet~wXzCq;QNipO^w|~tB-^6eS41{CQqhpe3q}@+vQeMd^r)y6(*QrKHfTd zf01nv!5u*#rBUHy3Z z^0YK*Khe;;l9!&oTyJFrC1I+Cy`KDpSa*&>%=`Lyapw1^omvt-Y3+5fw!hj?*5TzG zDoeq^vHN@zs#k03u-%Q!Ed+o^C+2Lg%iq~(!-ml;5=XaqfE5llANNVk=-n{TPqMMI zQvpWiZiU&q?!Mc->S4gPdt7PNnnK$WHJWm<*O|_jviJj`=eXwj&4eRgl~J=15fDz- zZFZ(ijI#NhP1st4GczHyXNu&|5C47BpmS}b%@OiI{!Q#)Z;ww*>~X$QkSp@q3j%&# zg!NdCsIoFL3i^AK`{(hWg#MmTmRCq5>k<*}yt{LA-e`3_nnD#sol<43Hf)JZk^iz6 z6@|ZCV^%1~otcqwf5J67E!Jyn{^RFQ_nW=8(NXrvvg5TTx-Z^f{TCjU8;H@D+<>10zdsG&3^s@v!eW}$OaPtMr^(Xa{~ZpE08R~__nC7WS=ggYb@Obo z%GuHq9~;|eZxFMeotc>#8alLQyA=zxN9Y#l@6Q|+WUlx{0gvap$w{4k1?t29{qEMS zsF14GtVdZ0y&^7rBJ!7)i%Z}O=i^MlXABau?-QYZPk>aLF`9tq^Jrj-t5R1}AZT@_ zY&vMi7gdI8_O5QGN{QeOexLA{!Ny?l?VDxRd?Aq864tNGJVF2&bkz$%1g_<2^t@no)!%@-?2czfv2HFBY zoeqDPfBV=8O=L9_HT%MNy4a9cT`gtoa331!WhYd#1mz5{Ug|bEYWFxZ$S{(X?rwE^ zf{S{%=jlUTtk>xX3Th7z59g~fo0?EA3k)&j^;@D_b~c3+LvMm7D^{4t3+M)^#}+{R z67;~qIypV%a2?b)67s|QnBj zFpDs6B=w71005^42P%4+g6R`1%!Z8)(5$aIRRL6tLAxSQdCsrTe+SfuTRFpxbJNlQ zSI6RxhX)5C9~@jiX*WH3w+Kyb)Py>ph?Omz%yh zLV-vyil=mP@N~APB3+$e#7!pB9=Jbj^~yYOH$SxnX-d+GBPr48pZfiF&_bU5{AxKc zDktaT(UJIsx-SrFYl_D{5Z;Zr2_686Bl&$rk_-27Z?&l^J>P(0tk5#D29=nn$?N;* z*hJ8`c_|Ns&5~GQMz69`)(r zCVi~9U^+Kf0QhT)72{Q?@rp#hg)DN?2kZ9ueKonkda}pwik5&h;{ws7*nq?{Tlaz# zTzarY&{Fa8t|i6IJxKD{ZFQ=Qr#X_4lDfIMeYLdgg_5MxLA4d3SiQXT`1#XRMz$*m zUuClSV;?HU}UoEr+Vir&m zzMZkCp+V_vi8TNDkHC3(`E|#W6UQ4S5xSQeHZ~cLe-K<}IH{0lxT5>XoPn>+zU*W^ zaHY4{nlHZc9|N0>@AaQOY8ya=pl=$Ir`IVqUFWmRy2G#L!m|ezfB| z0XIn&{{2&}kkuH9!k>SC=OamYI~=poc2?1mk;C)y+x{==-mQ6U3{kIwl9F#Ss5Dk*TSmD*AT6W9l~VE=o)mGFmiq>Am?S2XyH=<5Z7|}# z49FW#RJ*!-9XC4A^nyY{G+G_Hv5bD%x+CU=NSwn~6BnUm9JaNky}WdlYgqBvP0?^( z!G5(}Sl|VsX$WM1l@AM zMCnB~5yWD%Z6_nf`0}`l2r|-+YfEK)JjEZXQe>IwNO@w74Nd-U!l5AdR`)#pf#k*p z&VJj)e5urD^z%Rh%{?DNL08uZv}_Ia#N2RUD8e$sP8B;e_?xU}3+3s{tE-@zGayTx zng>kyqmzRX9_7bUW1=>HT`}--h5cYeNOpX?^GrOaIn$#@b(s|_J0xz+3#c__6N$O4 zPtM69evXfhW*iBK2g6T%XELADuhbZt{k4pykY_*%GzJJuZJi8>xJL+ss+ldc+Q4Ve z3F-}28QGMeR-aYh%+AQvzwi_uruAg=Txx5xE+ZB3xjSC-5qDO&y0)U+oIhDAXGSPEfuLj zHcpg6U0PN)Fz-*d&w10$P0=Fu1;wjWcyzFAVCRSLL)(|nx8&ny^~=rYDEpG9cQ;oK zvn8)%sfbBbrJptd`#zm&1}4cWghLVr`7u5SKQ_WQFG*Eac0ZkKKUfd;5hc;~So^fP zs+uxtvw1-33;${mbVNV2x{E&)R7i$-mp@7M@sETl?2+Y1{M(^6bbAICxJalJQ?^QPJ~o*gs~!jDqyxaOohPT z-%K){taxDPhEbTtk*&jbm${^#QCvxeh##ElBd%$o2zngwl`3V!Y7akz>UvVwu$`R> zxx&qJf8v2^bY2M%L8a3o&o_wtK_v&5wl6-y&BNo9&xtRIEa39XK-JOlfl-hS=>%bDd%J6_a{RZb&7TySj>uTjYn-u`r;QMWDL9;R@= z`Ic=LZh>0ss6 zAZjZ}OY^r`GaPFwRhH#T$%^=pNj`rq!XjKs-weiQ!+3>9fsLa0dOjsAVTE1*=h@fC zrb3Fy>I_uG=`GxJzuDMq7i+v9InjEIJ3QV0G@OgSh07rE%Epqge;-?@d5z(umvRxa zy%WA*&ldI~C?Y?2zdKvG1kE>Wbo;nmY$l+&$6|)!(Nhk8rKD7?F%VF-4-SQ#8R>a< z2YIx6ZjH;5`ypkoK3m|47ymcJwXZ#Mh8Ysp7pSBW(5p|RHE`>BDT(Jv^6 zJ9ap`4+94?jUU`P#`y!gva#u$QVJZJz0U)_Wg~OPnPkr z(x58Wwwtl*?-%n@8oC~XRtMa9Gf(fw@h2yvM!f1;qjrdS_^Z`9P4UG-%L#o;E34U_ z)U+izjC^qc$S9?G5zH=x2iLPzQf_r7+VCbzlkz`pHmF(-%rj3x0eOmxIg2%hqW)dd z(*=7oZ9blKJ<*G^($fuEo&lDB$=KpBPa1vD5K%H+nv5-B~4%acF`_Fp)_Y z1ng#F1@B=*qHJ~bX`eKdaUUN7Zzqnu1`@eQ7tp$At;&lTRFpoOBCfX4U_mOkb@6)g z2YW2_9ioAbR-8B|_x6|vXq#f#d?Y(+2#3UEEyTuFFDNXu z<})d7dbi+B1CbBrL#RBq)j^e!)EaYu`yIi#cWC-6R`@s;5!tZ(*1v$$DGfZN&N%Vx8+Sc5gUWf7y zCN`%)eJ#v+O2lfcK0i~~t5vmq==OKVv?4Bq55WsfWi5WHbjmwv?UsO|mEh~bC_nfH zW~y)T<(6wVG~!6i@vPu-yX5uixD{MV|6@qpqFld~>~x1di8HRj#R`oOmNsROL*UEI zRDyzBGt@qy`3pXphwoIIrJIoGGXG`UP}jsVmlg#T?v5A5`hibp1)a{`+CKkF(Gb&W zyGvo4?#bz)GWqppvoS)R%-Np`nl!UMxm*F-9@_;2$G~Z^ zCLj%!wvAN$BoChTKmiN&px`_GBn%*4M7$|on4cH~D(bUUc|kpj8|)_@4$lHEtBvRi zF`qk{lS^I-z1Et0sB^TdEf({lxAev#3-u7m{XfeLAJYxG6U_6*1CtBPY(qqr18qA7rb*>U<@bD1p z)EM@T^K1?HOnZ0bjKg0u;gIPm;p5>-G-PKRrccV*hCC^T5yg?Og}4UWN!SNlQE;m7 zzi`y)YY@tn?Ii@mSy)=`?Hov+sxe>q_dwQvnRd0Mz(PD<@7>)iF|Y|7+#QJO^m~RY zMuocAU6`oUb40>I%$*eT*@LNO?fd@r_KZI1Dt(nSvSSIwH5Fg33@W8OZUTn@PL$Rm0yLVT~*8y*jE-LORYGIHN8=Ma@$WhlU}~18yEO*5C9S-#d|)%DJ}LaI5~*fKK4p~Ja%t)FLf^>igu z<8w=mDAzPk{cG1>e@bF@&;AJAKiX4qwERFvWMRqB+bCeL+g@{KbI4enstzimIgaK` zO;7&48ann=wkIMe&>rkC)o8UL82ve9LjIN=(fcQmmFqQwb=V1Z_V6*;2PAVurnAgs zDX}8UzmRepu(7up`Jt<_6WbD7S1T+8Q_H`~7hGIjr3+k}561vUawU7xIR0B<*Q}wg zK=@3l>g(%T!AD4VI1b{Ny6+>y$M+~;aqY%;qWJvTQclsBON2W41H@>Dqm_ATO3k$= zpN>Vknvcu{It)A?H_!}cPi|+wmt$1@3Sgv)?&XRq7Vn=}#YdDBnRz66cxVYVzYKUD zd3yG*q;r1}>>dgo3hAw>IRkle$1UbQrf6$Rj1ro;B^hRyGs zySdzk-;|3m&K33ojL)j7=iNT@T3yPNkEUu;}bGF-y9hY=}-i+SUY z8yyboS~9U7VO*W(bdDCH3St%+u-lt6*X<5)D^Cn z!z3jRD7dyKp2Gx<;&W9Q=byMn%kYUX2wPx7a1j}7A zTRP$H>>jE5HOxv$Q91vlfrhp4_^g{jT|;ASO(YzRxZPv2H@qliOzn+{b?r~mftX-C z_yx%aCEc_7xI2(tPOr*KIb`5FUFD6}FbBXe4uw{FNkmFRJ;6l`&XaHHF>|(S+NYrH z5JXV=5TaQZg{mj9%o|%;S(KF81dxcNqeER5$#be99{n2VR%KV)Q!(%dA>=!YS_V<)+kx}(iWU8yfJ4P&nt_Ib;LsUtNk!e6ItTqQ{ zYn`~n#Hg?A))YW(ak}DokwE?2|D~wc^|0LRqNI-zsThz5tSURFXW3r;;xJ-7?S(`w zx$6`MOH(EE%|*A*&do*B$V=i2VpaVc-_3t1fmhoI56BQNIX#z~CH&W@^M_>}Ib^}r%m9EfCk z`?D~%G_Lji_Bb`oMJz`QJq#^UuT)fo8dVVI11()otTYlS3c*X__a;y1#KdIFQXvn= z6!VzFA3suWkNFzv>kGH*0Oo&5h@XI1m^2CF1It>c&+*gkSVwFSSt^kyqTp6&zRSmSP{XIza#c&n;Cq!xgk zi`ZXob)?7LgJesK$B=zLI@rnj{@t+WCq#Z~b91vf8{x^$fl_-$xugSIoe7B@lzLu` zVz-i#le>F0vLShvA3roki0k;0+-=Nm$V466{h z7=U%0FZ?K!YSb{I)|_5Y%@*<%j!I6&A$DPi57-ejE8ec_FudtSKpHX9$)(7n)s)vN zjGOZtZC?mQODrv+pHOYq3vk=`sR{=onS7{}qwObv@J!q7{inyziib67ARaD8-c$`z z6Vge7psg&rGs3Eoj8}W(>r&n^4t}IV- z2=D&mGc49G4NIQ6TgmjHKz*{N{zfZRqc#whax1`PFob6K`Kiy00gQ3QV+tv$p&5$q z_TmyA05g=e;E}g+0|=GI8N;)2VMmSvJQ->QG56Sb&Vbsy1}KDo4#Z=?Sq4Q|3q9@S z>3QiXJesZ)^AxUA;=VA9xxIPbV1YblDj~DZ$sbXtH`{vNCxe>Fc^_yGN2FX~3z^#k z(w!@mHGW#eo6rx@+KX)2e6GbxPVV)(*1jQY0cVGMOQ3#j@WgwPX2FXi7yJ3!h%m{( zLNkXgnZTYhT$`O*gjCOkcaU|VLQ4odFT}2zd$#sG#RQ3KH#JumBoAPv@vN>ltZn{z z;Jk#Ac@XoTKck;crusu8B2W*+<_vjE^*HbzoZX!-?=I(E@Ud&;qwuN^ zCxBXXAg4%Y*H{l#&>w>|gD=2E`9UR}*Tz*@AQ0_jE2L97lN;VJqn*5C#i;wKeA(e= z<&V^hXnTT5gz&(Qpj#uN&BpnUnJuJ;=Q&2AQ&Ax5&q?M!pE=1l_@Eu~>9Avq4@2)0`vq24`vFA#K_m(IV8r*rfB&=9j3MV7g^muN;hp1au_QKzgE+C> zQdxOioEUDn4!&E+5x}U=WLE1pN1#kMzwpVuh_3g{x#A;XOncvx_3d71N{Di1a@+U& z+3$w2PY~ed%B!1EApj|=Tfu$bcEIIN4*)bRHSD;to66(3C1rVPJ!4R& zFM*Os7ERde|J7oNYx8Q|+=JL`Bis>@VCr zVZqcW3;J`Sh{QV?%I7v3v3ao4_2}5Vb^lbaj;hDyl!39!)!M*mot3Hfh={5)gl1Si z6vXL!yx`_Ur}2&4vl|8~XpH!pgmAOArmx9nCFt$!;NawRJGH8nv8AaY_$-oa1C8@? zHhzeO7a7LvwOBC#H)Lja_w;z@req#m{VmRZxQ#$+`JE4*_z&R z5Tw5ShJOgc_H?eAw-Ez2xDLyu!S;1CJhY{UO!JGyTa%}Y5Qwz8y4pC!Xta_9K->Z< z5e5*iSr{fmkJd0{+(mF@t2w_#{j@o>Q6?0DJ=g;Bbw(Bz6C!&?PvN%)_8tKF!g8^C ze+%gsB*Dxlaemnu8D!O8;1HV=p8QT9H{daf`r9rb9*p3PP{|Gyh2~}j;ppVH=bHrB z*dvxM!6C#g;CVbEg$me_$w{Fve}^x&%)`-0-OkrZr(^^@j_~e;HIKn|US0yFD&)s2 zRuseilerO2{3igXH($Ped*7jDQ9{Ei^3OlyVfeRLe^N`B8Xvz|ZXTmb`1+G`InQZx z^9(?ll+s_eLf~7YzJ2SvL_o!8t{@QB1gO@R7w8&;_W0~eL8v823}yt}hw&{5#~ zGbx?=L*05jCAE}PZt&NIE-%HBYHlf>^D@U=$oumRL}Z`0ox{Rk${8_aB5$L(iw|uC zfF`BJ&_^<;%X6!0LH~Hg#gWg@mWKohQ{IE4->vM0K8iW?)?q^{=QT%WwJoZ^JfB_^Bj@qEJ1sGTtdiGaR6KG*DMlPDVoWZLp9fL1c-qA z{J8!owC28<0Zu`><+ZiK5U?o(crRHkEU3XRaN?uJ`V7*l&TLRE{HsVXC;!_Hdfs#3 zU&x%nOYG9lj&?Cmr`AuiDdSN9RwHG=vfzTR39Dgb7GUiSKlNPTRkfV{(P%?(J=`udPQ0-O?05`RW| zdVY%}!HD{iKM*d;K&In&5#klJWf=Z@#qOmO=d`ht9A=>>ZHq3mP|~h1im7}&_2z(`4WnVB;oGM2;asFdtWb` zE6o(u(9faqECHZ#wRJ3z(yiA^rD9V5s{dnW{RU{D~AYe!OVkzWqm(FARcEEwn z=V(+)-`sb|Mc+n8w>c1&qQ-!gBkG2tCoh5obcaS8R-%@P4r@gumHGVnAz$hl2bX{x zYCpIEe!w^)*08fu#^GeW5!IXeDX{Cs;6uX^(K=(YLaXD3u-%gN1vam31xY$xf*ci< zG0B4?zKH$y(R`FiB*Y43@ zVj&g`F3te%-#C?}lKxsXSHRNLbh^i|R3*D(Z-19mxUChCEX=vmRUfe9W(fP=KqQi0 z-PFHn{XSdkg{DGVGLS%ly1fMyapM-X^AH?BMTH9J3!%t~9vJe*Hv*g@wC5yX!%Cr% zVR6jc4%a$cb3~BSdF@jc6xX`ElE-*#mYW1{p!xX)1OQOC@&IsvUwHx0=&zUzxJBEG z*ZVObs?jo%Efo5k`}@HM>83nzn%)}sPT`$Z{GTNE^w0BoUeI6$uPq`RMH#FbCMHnP z$;U`>ReYeK`4M8+H#jJzF<#AbbAuqMhS>z?Rlxw-+rvwmt@_lM0@m@wx&rk|9&BmT zi;AZBAEMpGveh^M;-Xpa4+mwku_y2!%^vDFp3zTw-j}EHkO!}?7piSH-w#*Ig@jr1 z8w~88t*CLsVNlA4V#!XHyYUZxfr=#Jl|2aoW-(yzaOHiC+iq!TdgvI+|oIyP{05VS<9!FYu zr7hz3^l-8Dp7&*TVUgMWd|A|U%BfA~{%q7U&E!W=2r5aPy=PQ@uNKhe9FCf9$G6sK zVY4qHHkyt17mzndH*yA3wA${AWea)`i<+@`()ByGpZ6N-E$Y!|GXIb(PIOrF*04v` z48o>IBPJV({PRXJkNxj#h6)jFZC=iF%|!-MWJP9ppNQZWUlB;jv@~RvKB6MU=vn8k z_OQZ24d*N7tz(G9I;+dSwR!2xt@ROBQ16#ba8>)44-~RT-z;kJXO@@Ylm;4*V1%c> z<;-DK;JRHqU%Q6cPr!q_Y5(^Y<0CqgT z@4$FeNo@1UcB$@g(U_f1lM+dS6xTKm-)vwz6YPhF#FkpR#}oX*&cg7)2`i~)+{%R? z#O}Cg$4o{vb zrWAIg-kdvQW*MHeP;lbcuy%`R=K_Th>O~FKKc%)Ok(37`3?l4Fh&Wef^-F!zzcVUv zh+_L$!&9^`XT#nV6_p?FHzq>Ca~uTx+{`8d9XUt^p9u3;Eb5oS0eIN%_wO1fCp~aO z&;m@xw^lK^O3>e(&H!^8WJLtc3kq96YeF&?7t914wi`nk0BO{RnG>)P|1bThy+K z*LA8GShbi@CMyBa!9OK(v`CJ8&Y7DKDkDnY&oL`}--aK9gNiP!F$&T;aiA|YD05&HbPdAR$eWXkwWwINh4@6VL67nWtUDGhD3bZp?vU1YxAXkn#fqx9Avzjo^D8$^ z?|K|9gU;^mgkt{7K+tno??BRq3+U}rG3pGfko4| zj*Nb1*MDC+BC<0Y)K>k#ChY{pGy{S^!G8je6!BR4qQjlW#-Q6KEG4|TpGTz>THB2D{ zVa{8IJ&dIF9lV=oMiHMA7)M5))~IPl8d_W7J0E#4e)^3M)b(k{-ZB@q_J<6-l_15zeL=(<*qDxF$Wb#*FJ0;nJ_Aeql5`A5w370JZLLaSF|d#6Z8 z%dZgW@7dPj&Q3;hvV#Md zNyPeoi*QY8bdqbyla>Tp)<>K+}$faBawlJtG4Ev?hTp`?WFxK~HqBwm*V z0Cg#uaP{>&?@omxdOdSeaw?9~=wdSEEoREf5Dej)tf%Rin1qFdkb8W5o?C1eAKzZ} zLruri9=}NVzX3Yzuiv%~Hs@;3^MIZXnzu1ba@~1Pv9T%%38vB(o6B@%e_kl*_I$y! zJvsEgzVJ={5Bmv4nUikWyIWg4zCQ?%T@>qeVZ5&n?@9uy z9yQ3cT|k#P@_DLn3t%KE^WzX3B*8@T4dEzcg!tI3EKC(M#9Hy3V_j;l_1{_1lXRUG zc{6nxGj!!;J25vxUDz!fZgYWaX^)zz{NG@IC{8T zof_A=QKD=+cQ=`P-`*bfvvq?_gIVc+fZad)xs$XQFkW%uz!0o;=C+jIWFq+W?Y_;g zBt2oWlVJ&+X5X$|k30d9`9E6H2f{3p?4w0TVcWZxwa%8X2(d+%&J@<=z%-*nbxmi5 znfnd)#ei`@UUDRsr1o5j3YLeRpuUY z#Fs}p@Us_Q$%GiGHzjZT2Lb(S7T!!CoJ0WXnl`U%qWmjPT{>;<8LUn> zi%-~56Q-9F*Jcjj*Jhj96hKW~Uq4M@frEqd|;UmyR z=c+V}tgf!ci5aD3WN?1@+XSF$LFvHEE0=*#8vxQR({DBUtl6eZ(@Ea<$$AMoBcFd#Wfvfgtf-Fc@BQEfWG%c~vmGLZB7+STUxO^lBt>~3oH z12Vp^NVC@30s3M()nfGmb~WubGpwxsa!#FqROxj%;Z<_iL=DQ92T;TTgTnTl+Q7By zUQsg2@aTkpYR^sBRX4_~P(1ODs8CJcH!)wI(NR%Q6adB3x~BU%Qd zQ~NZblFlbU4UjKii{;{+Y>9wxwMfXdzuw^%BM70FEZPk(>UAYMsOo9UUYKy#CthXQ*&D9!U;o1b zMB++ad4qNT6v?|>h>-}l4Ndm44gamXUYZV}m=;C%X))i|PIi@yCFL3!SjFP4HSV0N zeK)}{*!<82H5}FB9|RO>{#Or}nG+4I9@dLJvp@;1k|VfWY1Pa4-D+4qgah zM#czx-$3{VL9-bNmmdFd6Jhxpy*;D~t!z2_i>IDLEide7)To&_$lU#PEB^>T9FALWl)zm1vTqkr6&y77EzSgUfi7j<`*~?fi@5pnB+XlZ< zoX3)gveE2J@3GqUN8WEmxJ-=0%>CR#6-7s0Yq2*sHSPZqNe&X~^f=w!CvS7yh!yE< zmB2$K$z$C?AgMPQt{REqXLai?&s5`AEOq4?PIn3ncR~dP%?cM z)igB~%~GeqG3s!2xX%eNMHePzPyxUI;4IKlpOw=-;5KKRYDyc(|zRd`_P8ozBnia%X zP{~q0Z#_8HRS)y&fcWvqvE*DwR@O=m81e;ECa;{?oaa7ZzFbKb6g!olT*?m)wH`Kt-rreD+>#$;ZR8&Y;6_S5KL^15Eos!O#kp`m@mJ%FA|-J8(!@_GDG zMiPRv2?5E&$flCsvgXqgVt~dM95rTfHC{EIl%y{8d2LmsFY4iSbXPM8^Md%{5snib z0}r>jm~lUxmbSN`z{A?lT?e198Xftsp||%kM{-GOTAGQu)rRyC0&_F5M90zGr9u1E z7#9Agdh5|0bO4=_uvnhZU+m+N&L$wmJ>1CF&0d~v(p z>y1Dq7IE1#06`57loYA@}yqxTPy`bz>&sgD7XT+elSwYF-s%a zx1*4dCE|Il0mqmI1jkxb+ni-}dTCO#ru z6n1FsXOtDWa2Il`HuFBpa4c!Z*=l<-k#=@s)?B--s_2F$4VJpt*k?|yzZQh)hagu| z)uC9!em9v3i?z=4jdxbVt*x{2N`J4pOIP>c|c2O)U)fZN$@Rt;46l7Zw&}74q5&=-Dr=`i!)8jkw1`lE!4R1+3_oWtBvS3CN ztpSBBH1ImKU2Z-Jh77e`?0WwWQ@?z9)Mz)rySl;jgpDH+q4<6jZBumj%}q<@Xw!TM zT>_HYwbgS9L0mLT*eeZ$t2R87!n!hbZ?<4j{m=97>N?x-{Sg76MN!Zk#M`GNS_7_d z^fX~VOM1%?-F8sQyRC`~J+*T?8Yu3&mW3VdjN~I!dI$#C+@Wn0y1~jh^EJ z1>&&J$1naw8@~(}YyYxH4cbAK_yP(hEs(B>)p_6crT8df%?83<2zlJtuAl&aaT1!q z_R-Pr{cE2)rUYDHzZVnqTT&ce1^1#6U(y{Z7QDS4vMYEtG`5J!^u(kjoTDgBCgn-F z=K4lz0;H|RsZt6nfHOk)jE|8vGoy~M*4ExpT3j;4L2pd=qgONh^IT+&)#7d4@84`m zUngRg?4rh?7(0+Hfg|6{1Wwb7Z*CAQw}lex#Bt)tVb-&#AJiFjG>;4qpM z!8f)Vur*-wnnB|(mXVHDS zIUEEmEK=Io9bcS4DKTwJiHnB1nx20jC?(l92O;S*%fZoX`QJE>kM>fG(`uWe1%%oe zqJA7>aE?3kxE@%e-wV#&pJY{?uJCc3o7FNk#dFO8{VQ!4869dIk|c*&;HK3uMZ6h! zOpP}Jhljs*bqR|Z2TVCuS6AiB1;PQ(BZqBM4yUn%QRJufjekGvhisZoi){y9 z^SRk`_QI0?vX=Lx-&#ej$wcD$`?<1fBNY50Y<1X4%EMueK+KeB>0G=SV1K&7b%H?@ z!_V-*G~p{FN~O4kLX2;m1&AS#TQV$oR-cQ6IvL<0gQLRnASFaGzDa|JGzQFaz?R4t zC!dA^bw~Mm24|?Gsq~zufJXXU8xBI7*!vCYaGqO8u+QCz$L;9HWyk`=ui>2OqC>$% z1~r`g-}V}hy>&4T`wXRW*^;}zCzePojjHn^MzSB(#ki;rc&1d{ZN+iy% zlBFDvVWf5du${eZA-+wnFMn@%?S;Thp(JRnuA>$tT5R>36AI0}pTb_0?--4&b^{@$ zhIoL}BrLU1bSuTHE z9CJCqgzo_^oW>rroU~}UEF}v2+@J7=r>77&A+rZvZ@}IS z-NX=DbjN&95uGF{2aM+-f~*G(^YIj`!jBPM|8-qB^gjRh0pOyiXFk-TvJTwb52b=c zqRA(6W5J3{+htv=n`+|*jUyo;_1_cZVj#Ph0m{5U+z@e$o$i0zThz*Rwyw-{CR-+{ ziPK?P2n$FO0CRLtrwdi>qB%Dey!E~*F(C(BD%n5K!j3T2&(hP=@8RzlcY7b64t24! zut2yWi=q;76LDEyrMu9VZJ8E@_rY@lPUqLHLk6pi;^gE`+w6T)%8r_j4PAiSKs~LB2DXPGlzfo$Ov(b5><5WNA=3jf95c=RYM67smVZ=m+>Ecz0bu zXa)aqSh-W4yJXdCy1cw(l`ql+Y^l%>;(s3Orc;fKM3YLCQ(ZFVnlwE>U3_DlBxXRl z4IvH>3sVmgX~H5UB_zBJMw^q)_dQ%5CE$IueYLLK| z4b)pc7dm1Acu@XsK=+WHZkUi!ieaBHe&)CNWFEbM;Mcz=souYNu+vul2#F5?=Kf3@ z4J(PE_lNc8dVytk()wC4fUnpom2$;wGtqyC)L1mX;ef z2T)!Nm`?Xo#2$$UGE!`y6j)sNEqYmewTp*`v$XuDzcA<>GB#gE5)5T#k@{PtTiCD7 zz^Kzbd(Ajbtcu?Y(I)OqwC0w6DL1#wub&&E|9H_Bp`+u-F9A>$*)O~1r3#z&i~>Bi z+u2;G%d{!#ITP5P7@flvLbYMCKCSv#ZwT3n8%_vsL_wl zgoU&P)s&8+3WY*b z2%uI#k^>5~)xCw3_4bz08k)q#Zt(^PQ_a~gTjA2Y)b|)Va zGBO}MI0Tw%?2o@m=NFP3{S&Y9YZ6CxfI{*uNWz4yU^^J-BY=j05DbPga=p0sZ!nPu z@1qD4I^Ifv+p^#9}*q{^Yn^@ z?Etd|e7pN$Vr{_tD}4NNtAoB6_{YQh>p5|JBXuFw-Ps=%O_5agY+?cD)rNOJ9TwK# zV-sty%VVzFzjlqf;aCiA03`DzPL!ai186J251xA{et!<98NRrv!=@j-$T;$LuSRDt zRS^0Q3&_k$@;sPY2ecsQgq&EMJAjR;+lLq@P=bo_&Fc7Rx{$69Rzf}cE5b@|Kx<7% zAr*C!ZL=)XE3{cX0~d24AnnN~>O^2d;$IMbwj(sy0VJez!yL*E8y#mR=lotrN;!z= zS$Y8Xo(D~-ofUc=*Jw<``;r&;krcisrk$aaLX7hD@RtTaAxRd0vs-#tJedA2ei`Ud z1Y$T_?e;pH@jBAks|4mM$2KQVv|Vb9Pq1=w&fcH>NvOKQ5cPK5SYCF&If$$y7yi}h zZVx?StV)GWwk%}!qYa?eEv&58I=q@)@Qa_1y>!ba&3ofS^$FHI9$fjU&-ad>R&-Sf#FR}O&_>tm&EXm8b80wS%!g8d z#xefwb`}YB2YmsZ^vZ96A5hvhhZDB4C!q$-(n7*wn>+Ser5dpiZ?M`jEYqe&xBPj-JCT}ESV$f zuv|oo<1_qAV$_06P-+WuTha*xeFV{5G9EvQ;HGU$4FT(Uy~lHV#N+hEt=D(AOO6I; zbkeX;1or!9BX%%~T}T{(S@jqhk;j8_@y?)nAiF<%ZPhu49E^y8EWKP4;;S|6tV%~w zWj~+9V=d$ZF!_1NWiWU z77SQ%e&~l5hAP5)&L#5_T||%Ia*71O&rjW+bdcc*!|f7>A|i%-9@*tAfC|ij&%p#A z3W(I{+R61=mUK&rXG1T#dV6PBqLp;DT9>OHT+V#*ui8a%&o3 zETDZ^ZoUQY_SDvJT7tn$INwG&u!z})cb#s8SjT?!HArjbjraPwc6r?D*qK;p+QRf@ zAD^E3yuXo-$^iOO6F^IIy!c*z?D{_#ff45z3{#jcO^r!qFtCZB09oh1{8utG!>{Xo zKQyvo06GNFX?FHp5ubNI%)4ly30HYdp5#}Ge?F*g6ZqWmW!n<)>I(=6U}8?r@)RIZ zBLr(QgD9mjfZ6u{{FE1)fC&Hh?-KEP3c#+Ftkx}$7e!N^Tq-y76}AfVzrR)>B8lI| zk-Ta72ds?oq{{QCBTxqDvfu%8vuoa?{ypVCXWgb?Jjat)CyJG1OYMJm_PYJzt%V#` zCyg8W4#NHKygMSW2F4`8wDkR^43%_X(`FC?qdi7MIMx4LA;Q&qVHRnr3HF@zjK@Uh z{cr+I#SIj^-Yq$7S;+f8*BeYBcPd$q-LUkdbI*_|bd&m{J-~oWF&($eEL(<43{C};y3g*79+kBqzaCc;CU~Tor8^I^@e-4cy-F^ji8SCkxM;M;?-%(-IKBiKxW1icYdPlQ<`|p?q zMR-i12UP#v0~pHna_|kBNa+7fa*k3K{_k4=EO{p^&`a&VOVC65Z~cJ}G~@g~y~+OX zBdD*h4cz>@{)1n>INvyVaJ_f}^IiG)@C(fPNGD*S40{JuQ&RAN@);qR0L~i$Sm1b~ z!C_PA9=)f-w2{VthC6t)*v$JLcWe+fpRwWLWB;oAyd?|F^PO7-1{b^ud5{A=vy1ms zfk8TOaT!R)f^OE=gWqU*7ZxtM^YSh_bKiwsZ*cO7`E>SUDg*8=??TT<%5AvBo`k8G zGQ10Ljak%%+7|Qbfi{?6A!VgD>X$N$z#;8z0kXm{}G9wcwp{%t(gDe z{wdHXV(f=buHWTaA$|V8-~8Zz?oAP0zgfMufAi;qC7WLr0U3cv_ZRi*7S7If#!r7% z&ja3=X<12mv5n{~EiEVYZ)n|?_{Q+E1p;=Tsyi=4#CiikK8#_lKEYh68-qb?>~eA zKZC|K74rV%(RA@(i{Z0sZnx&ppxcir$PKo47Anqgb(Tlpx7djQQAAYzwg=mCrMKw_ z4c_-x{mGz0F3tuZeqq2&qyD?Ta;SrgL)Y8s8Uc?kq-q>S($a(QJ)S=ELA`lbABTR; z_T~r#q(5E#w7Sme{hG%Ygm$W5Vr9wt*zhClVX)RV_;~(wtuTr$Yzi0?J~nGU0Piw8 zo1C%kA@=s=^M+W^o9NtHR21!alx?x$F&_31N`w$Fe4U7wy3m}>u)F;^x4HMN&hDbj zCNGKCUUOq$z;`PVx82Ljsn6>DdGB}f_iBA^^A3!Ef`T-57LZXO1kcgT%y9KSZpATT z%koqis~$sAN1!e>yG>Lp%Ids78NPZ4&HXkK={Q+x$Pk4y@U@CNTj0W0728xBA00P6 zd-UwbuUxdBcKcskeRWh^OA}{;1b5fq?gR@CLvR}gcXuaPaJS$Z2<|QcLa^X2!JXg) z3GM{;Lf+f=?Q-TE{+N5GyQ{0KOMX?=0%FiS`E!)W4Nui3?>Y0Mj|4`6)XC)BmRj+e%VT6TM zxu7Txck##O;lTSY#)r*Z*~x0}R@fKlnGo;?ic^J0-Tan~w7L+4r#hKN1aWV@ffH znrO~>V8R;-csYMysmf5G?WRlixe3f`U6$l5Dm{GM?V=a{Q*(J7&Iyz^2dWVRJsjfU zE{^0O+>O7N9$f|^3O^D7@j@J(IG{7vLanCp_Dd)txA+2Ya7b0-&Li%~-QgI@)SSZi z(V6X&L+A6I%?{sbg*69WA0a)AIQCEKkzvs}iC_nOmfQ73nvZ+07azw4#2+sYW-Qm@ z?SsTH{HN%?dLdKfjyzKRET?X+1Px-^m@yTTKBfs1c%!tEYIqbPvjYg>}YYF62db+u_aYY*P z2W@Qz8#$%y)##D3D-s~G(ydtnSt*k&dUUPKVmJQ*6mSMK-~W!*$Ia^!=jCr-9({V& z_0=khE>lnCe%iac3=gG9Pi5*q>O2C?xz2RrnRX16xr(db0Hw6`8~6GKHY5YYQa^tN zdRQnpoMPi`(P5uP4fBW!0e&g&s=LF7F51<{+D{%BC6LPlkDv#l;1CJFK}q{s4JHQ! z0Q?&8aKj|WckZ&+&5lMIdVjsBwdV&kd`07#C77i7dt2|qJxluh!qe%-cGAW6!)8|D zJZ_+^rrFM>fG06x8^0A=XjcxTy|a@7j$hl^KRde`om*EG3u(xA_0gLUgPBtYYZz~1_eDh%YgPT+dh@Z1TI&3Y97|3$SU9J#>T)Ss+_6& zOH$qjleIpeEju^YR-?tjZfCf+aR^$-pD{g9_5113UUu@O=vfO~PcRjb5B8Ri#FgHbd z+qG4iZry0JdiLi0=F@8X{&cHzoYL3#cPrF@r6go}0Ei}B7X#j(*T0Z--!`PbafJ~e z%eR{Q=@Z+6RWt^KNyy-H@y^koZgH;EGM?WD4@bv9(f*PVVKw{BGzZ^5K23pvB<`?} z^Ea!P%XfypMn)4r^QBMx9~iFZ^{5NVS5NJ`Zbd;J{?<-%yhABW*S{87zc0hJebxPd zR-x4_J$pj32sj8!3B35K&qmnyxgU{Af?3s)lS8lUOi5R*$LFqtK-;8ga@_W7r>2oR zzD@3J6%{Cl#`)?(0?xlNQ@U(f`BPf=l z!9Wi%`MWbtEM#E`gAKLBl)_kP_+wcaf!K$LAd@j16|$NrG%o?YamMVG^(!NMTHC z5AYn=uCA+PI3$Q`L`bpkFmmn*)`S5!uW9p2^+-$@9mreZVg73VA^K!c0wJhSi|Jbt z;c}2Ix|I7(g!?|IFFCcWa9HQ(!GQ>DYL0J#dfjtzSn(&A3a}(As(C)XDVS%?hMs4d zu^2hDtG>2&=A!K5N5TKAfCWG>k9x&-57xO&PmzJd@+wu~_7zD~BB5`9{U|`n`b%~F zWFr(4bA5MXv0;CzP25}-NtP`{ug&WLfTlmI2v0H>9?ZVtAIes&_h-ItJE0uh2pB7x z%h^9ELm%%pu3iJt_*cmIGU(K0uND{PulDAT9$K{lr>wk1e*%ybp{b?u1Za{%0VyM{ zj|d0-kUb-R{!z!32C-e`KIf%5dRhhIHn$y`)`X|phd*|HAEpB?NFcn<@8yscn5UBW zPQF2f4~RMY78rJ=U%!xe6QT1pnt!+Khk{db@wk>TX*?Biy5 z-g(DQ2AkPR&L=8P5z`5xxwMZq#OI7+`uZqb8}vE@Dus&pYnH z4~aCTJvcZxlYtu@9T}*pt1t2a+i|$`EG-e^Mw%KMr)OrWZ6`GqCPQhvbsA_eHIyV_ z4Z5kcyZ`G8Ag?Db*d3(!$iN3LCK}JgJvIe05GSjOlj+XAcfkTXmb@3@g+?;9Qo)gm zb3zb2m3YZMB7QKMVz01SH%H?o8%lUcOC=qhRu9nWTO8viVwzk|ta#na^}c-W3=RA_!728z<#*XJYJoTuxRtX}|UN{DQYD25sNyeLl}Y|s$gWDWNy{4dm&7j1 zr3>EzRR@>1=fSiI^qW3n8#5$?NkQc?lEa@~7PE-ItI}y3uajJjiaO6YD6d@hjJKHq z0DvX>K&<$!#{{wxJY{Fs9*@OTwoe`kk^0lMB2^DZwUCJl7K)_fm>XtRtrn0o6*8;c z6~pyEWfRLgg%yjlFr@x7S2p*V6>m^#ykwXc(V_OU!j2b_sf?QX=U6?5t**DZ{t6Rs z*wdIi6Z$)6>+?zq4j_^DzsLjBDRBK6ohjf?8;Wre$sK@H2yOm50#0G0xbpM!FVJ_c z;UMZd7-QqyQGH%Pi=Pf{C6h_zYyM}m&@WBPZc*h zPy@@6tWl|frE~tv+2>1a;F>E2j0b-#)NDw1v)|B~)%*Y#J5S^A=;-);IF;rlDk@Nw zIH*Ho4lrA8(%xCbKoK?Ow!8_J6C9yv6{%mL|0k%7W>e9CMe z>b!6bqx|g5DcaLU9{75{Y_o`vS39x}$^$w`609mZ?p=hIrZflj0uBR9iNo5^-3m^z z+MAMhK4OEG^Rqeb-$Hq1!LG$r+&RaX{mn}a+0zD4CEkG6n!iqA$$m*wj}(5syh|3M zu;uCb(xN zYbr0~ks*JD^pWgPH(>M^kE-X*tK#WA@K85 z7<%2@r=Ka0I9UXm;PhE|L@d~iu}Ex9P3j#hl?Rk)2Bhvn__Q&|lq%Px4a$Paq;LQ* zc>vioyw*#|fHM{BMHPUhuT_QV*=GxUvDiI4T;En;&!73RZ{Iq{;o>qIa zB1S-<+E3wliJc226Vv{Ed>lkU@QXGUtE!Kv43YfovC)<&V9`&sJEV+`+X)B9Ar4Nv zAD)=^Bbn5+if(Qcpol495*EKm7GiO|*qeLhp>g+ih>F9b<_$juJ3j^e$*D?#Lq2~9 z8A)JAp{Q^3JMlTQnNy=29fnsUu{~ROZHV6Qf15Y)qKR|0kkIY)`qH-ykYeV@ktY)+ zt|m^pT+hsZ z&U+Y26Bn6h4?;mFZ1i{~ZJ|dN&+0aULy|yd`d~?_lCiAf8Rp1n{W$Sn%qI~z;2Jv0(aHgbfSyqwPZ&6(R#-@&OSn76o$uPc zJ+uUZnD@S-h66S+n0+nIZvoO0%{fSppvdWwE%JLbC4AGwot!_{FNxB$mGmk!;?Etb zbP!&=k^nN_Oi+P)>=2j>wY$_)g4D|+IY^@FUZI0xlW{*Hy1;RU_INcw0bjGffs;HQ z^zJk0SR>-DNroIneFKD^Uuu}H!HaPO#XXe$a|p-Z+zi$?GPo%u%gD&8L`OusAVR zlPray(nLfVIjv*)XDGe(Q?-9r@0p(GQqpcwTt_u0wmpyn3%^NXTad66xv$-Cx8*=x zwPUKv^I2&+9oxVj%tv=9$fR^AAh%_TDsaiJGK|I<$hyCW$9IN34^D2_kAvl?jVqYy zf;e)fYe%n(<0yM7)UBK)315Cm`HlfLF2wlch+Q<<6c@|{CAe0PiK_TQ7XgrWC{6dv z=Y;r>SE8|Q1N(8gpynf*!3b)n@bY1m@4gm?8-SrrA}xstA-JyCzgSCM9EuEnr9(cH zcZ3JYC(JKZZlx^1+P30u9FdO%k@NTKYR{FPls5|4hIT%edQ= zFC&O4Ads-!1LlikFx!&VlyGv*iAth0?7^)ekvE~$#zuBvD(potF3-Lqa&Kn@8tKX> z3YiFo@;R9Pj0`y)A5gQh&+U)f!*b$?m_M{F7%9((>=JZlyR;x^!@?$-Q_~@+UA!d( zDkv!wIa4`p3|h1SV8t>mFW7ofKsMqNb3~B!^Gj?YoW7p-H;Hq~$%(YX!3*<(t5^7T z(n80V&M}cdHPm3}Tf#P$eTEaUzJ#6rPP~@%HAs1%IUCZ7L%r~ri)SW~+^J87hThLK z7XiXx%#Hm_MPUka^Oh+e6ARpy=ZN7BIep)M4u<8gjZIx6=QgFkEI3vo_ln6GM8nSN zxEev@V(e!!!?B3u-7{pk;~(7=qf!4JKX!?^BQwP9VIDRRypmfJlaw4*n;sBflb(MR z%RR~r%_K;Iq*5n;$aH3;l3*9oljTwT-i!lh!CjWZH`ZCS5~OUSVL^=)ZKTh^Q;`~s zf(_UQFQ1IsQw3#oGZya3!@YpjvEs7`4kyClU3mn@<4|L>F7`SuuOL#o*ljx$$B$Kw zA%h>)*BP+~!p6NWJFpXMtK7-jXyKDOIO))kwthoD6w^Wxv=)~qy#hzhV=YAzY0A*k zq5yuMVqzzlUH?D7)67d&8c@{jdS8X+EJftGZ~=ndCv5$-!Up|j#P3TmG%axep@ng7 zu|4_h>#+tCZboSh&$+8Pfw5Lu=*nPAQKw`b3dX_yjYVzG3+t7WSxX&fA<=okFU$7{ zJB86Sueg+;8kqEDh^5$-gVg|puxKgalc(FmTnV$KK8tZ|A%hJw81j3rJH^XZWM}cN zpUM;sw;)co#|!Q?lt*(cRV!(J^5-e5gyRj7J|p$E77;Tgv?S%>TfzP?^_VG0qK{ax zhQcr^*J6T+6OUxcnWGeiUjN(pUR$@xdm|dj`TXiI+!=7(u#!Zn#=BRD`KH7b;8U2& z2HW<}bfb*%;Zze8HDZNYhe`}8uR1WL;Zd81vN%f$<#L#}*d-3A*SzdG(;k$ovUi;{ zHlxrvVYPSxn6vg1N>g-&IvX0e?C?3>D#Ld>t3md_nCOH-_nLmY5r(#Vf zQbd)x2Xj+Az#A3aR3~v71UiIT(tRFRO0%} z<@@A_2!xF|JlP5_nRO~W^L7>mf3s<3z;#7J&#lutagdd-N4EC!$J6+E3MCGc&KO zUr2r2uBX*h#$4TcpMjTaqQjA+lb*R*hoV{_rx9U;AREIIO?a^6c{3NVDfcG@Dl3PL zv+CqtG(SXl{XLV(y5`H~9o@*grh~VIPRF=!Pec$B{e@dHgXMC`+GE!Jxx>oxfisXS ziUw#hN5~BCpa;wbsJG&_MejM3C2mycHglu!xmUs9Sj(SHOe~6tgra2A$mssiBU==d zU3Oh3FQNEEe|nWIbnagmNqO#b$;j_XlP)eI+sM7r4EwoREtYg!tn!kw0(v>SBn~@K zkVBJ!$<<=Fr;ZzMXR}OSg3AHIMXz+Co$<@wF(@$H>AgHrr9fdAiC%v$Zh_ycxFI|# zP)JXo7!YeiehRz+lpN3xmY$yh$|%KCHD+-`><2-2Ija%8DghUKcRiiC{B~er^PN)y;&{_9bxOJR&QockPOTiQ)60yP2O>3pV9 zrc=M+4W4}H4ez008h<67^;&B1Utd5CagQa4>r^Et6@%gU4Wsd9jbDX;fhutplk#DZI1pRIhewlzN6B%(%vWr5@^uvXWPz1HY{0M zUGti+UTuMj=WQO~X>1Y*S?OyN(95Ck5}Ousni0oPs&0kf@gnH7Va8WISJ{yoVt+Bn zy3#l?RvOJg`}?TIe7a7-N|GZaE(qx{+v+P)i53AodHFPcPpbA2er%(iZ~f%(%HXle zoIo^0hxld{1=d*CgJI6#Gw1Y`upH(5R8AK{rOld<|Hzar`~22j=`TV0zrBrZQBdt_p}2&c7%FJ)91U^$JW zk$5<1SKeEaHXjKh&F{ZVzhx~Z?)lCE(K9;xS~^W%P1h!iZSP(u3V2DO8^VLJPFM_x zu*llHj)xOZO>5X@k5%O4n5oyr#0*V>_%{UusXPkR-X#&|FDd1w%4smA^dAs;3?Xa# zb}o(lUMNa8HA3jnIEeHjanVb}O~Y~xp{!c`ezZE1*AZ`%T>J{T6e#y$p!n`(2 zncGqjH?kJT)F20T1*w1-zpP7+44K3sdPW!S;dSYGa`|;(iKqv(O52LF`0}*AdRGfD z%rdmD`|uBba80+bhPIhUJ@U!WRbCH3xe< zV3_F)@CU7sslS-Y_H#M5ApACERUD(({9PEq9wYzbKBicl{srQ?QQ<&A`GjzZ51SGr znW|QI1*@B@E9-Ye1dGosW12-apTev<(mh3K!w8KO>CP*^C>!T2xW^xuFj|^X zTN{u&Zx6_8fK15wsj|G8Bmca~5@$9Hrv#w~A$Y>e`jd)*fd8v3Re~QPT6PmV^O;ut zDQWB61gaCIE-@-wlHH&Gs4is5= ztS*_#m&T+BlYFs0M(EY-Mv)(f4I;N%h*YI0yX57lvFC^<`2p^&E`kN9gvwfw!*DsO z0WaL6<*I^wrTl55o^_{T;tH0rRk8s*3>>P4TlahuPG3u$3s`ZZA`ebzX!!85pX=J#3rP;QK^;@2#L?mG_^nO8&oo@=fMk zt?H(3Dda8Ku+=oc9L`oh5J~sOO$|fO|qTUS%OEHX=KI zWeugVI{pEZN)9)k>??w4kWK-?r)um%rr<=f^+=7#t`Yd$x1a9*MqLa@$YQUcRmtnD ze@@>Hu}l2oa|)N^Gtq_p0%1Z;Hy&($xDoFBl7WaT$+8;IvmJp{7ltUp6YotDLED7a ze1hWv&ec$m*@B6YPp!eRiRVojywP{@MSFC4>Vx)*PS%AZX=S7r_ zHu^5;A63h#x>2ytE_lQ+_qol>B5f ziMlB-Y~^JLN_xe$C`ny8a*SSIpLCNt$QdCmnMolj7J^qBUFc9{;#^ko9_T7}g|+}U zZJ_qLh{nfEVdf^B019t0cj_Qt+mN+Ep8>&9wPr#C0ddXxql6?%AP3cl@ zFyh*WaIH8_w>n1mq>_swCl%4)SVGw&8w&o3TzH*b53u(ya)e+;eMgRuXX*m_hMOV_ zR(G~#Bz-9<2!h1o#~_NrM*)@&;F~blLz*Xa^~(DbK#iL5>vyh#D+ij2M?;Xy8vR7X z0Nzcx@VdLSk*oY`-q043T?*-xPAe6o#9%jZXv5MucxT_z)a5QwPc9WF)|fI@4fXfI zkP%+e1E!?@k1{}rM)Q;{MmAR{jO?#u8O0S-dLfBRkD&qJsW+PMFI8N|UM zl8wAm{QZ3x1CCJ4&e1+4;J}V0g#8Ogxgs7vUy{pzYo+G;NBnI7g&1+u0GGmm~UixtCzLL8I1hinlSbyVN z9`-p%V`=P)aSy!f{(NaA35hRTvCbyST#ec^ukHoOju`Gs&EwpfnT8~YzAQ&-X(@94xY4(23nad8lQZ#V2^k({ql5q2l*hVU4D za(z4rPj%AztMGX2AifXgL*imx_kUX2Z&y3H+I@Lk2Z^7Xygf(t+KXFAd|~lgv6Gq4 zx1=<@NvO>^zplZSf7+%Iwc?Y(dWU44mf!N*R;jp!xwuEqY7SpU>NrxY0XjyCm3y5D zD_QJaY%qr!*A+Tgr~N(8TN>0hVqPyOFPG+q${Vka40~T42?#g1WcPWI-509LAhe;P?eYCZQ>3~f7ICU<`@>6th|y@&5MItWH`LL;OsN8i zZc}K$&I)K zem@};qAC6lE?22>fre-n@ryi_-`oy2(kMlM5opWnh zUT>f;AMg8CQ>vDM)_UoU7W4B+>h1zq2+*!}#1*#wdD3qOZsE_S4G;tKQ>Lfog9 zs#Eh+9p(w0r+Y4HIuh|=x|Dq~jM&0GsVf!-FqWN|jzaH8%@6KW$%0zYvoEqR;8YNi7?X zu-l21jMiH(Pc%a9Uq>FJw6g6w0*{MZS!oVUW1SQVnN99Lk55|eq5k9AvEz9}K)zUt z-GDm-S!3wCG2Tdh^k-PDktp{=j3TUGM_&tOsx2I$fb{x;`EG@LJM`BX6>w8NyW_Qa zO;L&te9HrkbWJ6)vT@2Pj!{8KJ6qwRe+V!8UMD({22}P7z@B$$Hy+uqR`0FjjK6-J z+3tekmaOhj@YW}1R$E&r{u|UaBhTa>=6L5u=K<9g_9uT ztBH~!C|hvt$?f(&!zicD4wT_f1(oj)&XJ=7wwgJV*IxNHJkF z46llPQz}=2#(*B`S4T_ZmGY63J1iJLr0W@>vgsWRLw_Sp+ME*jW|i#^suv0_F@=Ig z+8XYPJ;V-IbtRs9vPnt?r;n_{9>Wx!3Krwj*l0qcED4NbEJSBa4fpO7zP|d!^Wg6JF)lk7F>PbKNFXGv$qRvMh%$#IcCgcSMrQeqEFg#D+j|mD*bHSZ zU5gV@rn$t;!%W7gbK~@yGYwYqw*dtjTgv_Vtj#?&GC#V6INurw zYsSVqia8lW9f^`kQom=+b3>P!&eIZCs^MKa&`Q5=feKzqwQ<`RdW_f#-^$OWk(@2S z7EsNhZ2QO&e8jEe;U|K6nH>h@8h3wD#Wn8TdfRyPvr38y)#Gw~<}=&}G*Wp5`8=O* zKKm<=Z4ZXMQzxrTcTD`WJL_9SL+y7-x4$|ItXn_0%(wljLP>wrZM<34f7E(hJ{|KA z$kZVhc3aCuudV#o7qFLQ-H4tP3Zfw4+{(*pzgzF_I1FM4r8G9`Q%I8^fq+VdpL$-O zu>5rT3c@k||E&aa#+@*vO+kMEk>|N--irlqLU5B0`L&!8XcGe@ZmB^FYp+%0X<33l%te#0vE)0(`)&C=RAsglG za40>&ypA(zto<4p4xV%*%Aq_HGxHZK{hw6{Lt8GHbMx~i<~yCFxmIj=@B)I%m{=-K zXAZEPr(|($L?kewPoXxlBW7v^=-}KPk~$IRG{1_;i=EyS-4TdYV8oRKQzb(pv{+K4 z-;ur6&rUeqfX}Sd)Kw(1J&2eBHC3Pkn&gT4j^O+~*zCTG!)ztfi;u*=(Ub*Va}xCV z9kVG2@U;kyzrs2CAxohz&L&g^$JySvd)hN#!m-^9RQ9NRYl}`t$K-V>09D^X_>?vY zpqlTH3lcVE=Dm+TMdAbz$l47krxu%-;^wd!GnVt!zVNpZhdN5p5GrLc&C_bKwa%l+ zky9p{rk4;2SxlqzDc8=&>_gaO7D*RoF-`M29R^||s4K#{aL!UqjcErmH0Bd3FW)+0 zawg>A>B-iPtZM9Xa$4$HgQ* zJyn+B)Wj57W;lf*rDxxUu_>{qV<5;rG~PWOv1h}}wUHlp=sx(+Hs2;5eMl-trwJyK zAf3&W@hcNWWh~cWxMfC|yg`j*tk!ev5Lhc3HC@+Z--LS~WDTEb?$m4E8ZL$5fNZck z#9%P_>V4v-Gzy_%A|*NPH$45(_0sJkjDBsu=thJrdkXoAxl=MEO+e2|DVfkt)s$
ST56{muoc?e;0NU6RiEn-w;rhZ#I^_c8)&#fTKsq48UCsvGb! zy_kvEBeOmAG_WwG;s5+F1w<_PlYzfe$Qs?=r=k4b?)?6gc>nFw=6Rdp?M5`gGNoUn zMU=K}D>1Av9*r&BWAwD1biP_yy5~z!wi;~A4lFo^u zi^tHx)g04VnZo=Oy&h#W6NRpllTk2%=xF1c0U#>7HUy*7ysHd!n+yr!ISFsKAK+2&In4xSd?18LVCYY)u(P6nud=SAHTYdXl? zDQVHA8iZG^L;yHh-VyM=69w%559zprIG_L}{r+_*f*XM>!FgSqjyS+=m4fX5?sGM` zMWgXscm%0N(^5?v)eMe=C#Dk&@YzLI%H!?>WdB zjaOzdg20wR{e$vnM=!ATkrx=7>CVnu9>H$!JuTt&2hNg=)K<$(Qziu#Sec4P3`-iL zZpR&G(IIh$&PJ61@86f<2e}{VVw`D14=unU3FALN4fGvUt*H_QHX<;h5T!_c&*LrV z460u}Hb93E>{_*1JfjcbU;m_$gyqDRQs?=Gaks{H)J*1J)k+6*Dz@EF+beP?a&aRU z@W6=DB~CsHono3}rxa<@#RCm!Z0sq?$E7t&{!J~35~+7jHBuFL^e{YuIx>orYm%f= zM}^s_5qYF6Z;F5&lY716tjEgm@ySIhQ#_MhQt(C^qELh#Bmbr*QFm-RS4`s{9wZCH zh%LRK^>ZemndRVHCNUA+1`BiY@j}_}ICLV-7fsS@%KIBjV89qceGI^n>@df|l;BH$ zl@n!SLQm-+F(UJCN<~!Q#F@xG7DNArrOq{*uc~Dr!2R&K;S?fdLn+Da@6iXs-WHaA z<$!5mPkQw%xB#w!TTs=odhIV8L=K;e1y`v<#rSBU=)k-!EzLsDs&)%sQ&K|AR*d}2 zd0Rn{P?0jh0J3j>GvsRh+*N+!5q)3>X(c30Dh2X#f&Y?l@&{j1v*gB_D!&XKPpk-I zGEzZ!K1`eN=zYkz(^Ie?)36^#~t56yft|e%s-QQ@&gDma2*@H4V4@WcZy@B z&fKm;2!?_yQm6PYF9f_rw?n{)U50fX?G<2qU#!Rvc!Y&^JYK@}MXIs<_uT2_HO10Y z*Dx<_4xo+g(D_8HZN#4^q5d~m3l3yxT4!uKLw1KZtshxADaN?K>J&`Ito(MCt|fyGJ{-D zp-$4lt>L*M@6Nw6(n< zn$Oekp}x4~8XZV*%- zv4BvwItXJ_0MLiFO&Rb829CwlOkOriDKN9;DBYkc6O}WRMvC`RvbdJ>!X=Hs;EYhvBvO1tr%dnW}ib3cMzh>yE;* zolYxNf=r1ZP-L+676s@Z=k{N%jo9JKV$ zQ<{20hU^+;YB*84r(ngFqdYT0zN>icTw!uF%q2}&Tr8FLjc|%2Sv?VMC7_EFNh`)-dYcyq=+jXE|cz& z$i0{AG?1$xjzn_CcB1#6GQy^^v^%tc42Ys5VsTnGL713vF{KWBrzWIzQY&JKYPxX) zUJz&c^{itfIRkU*!kUwcl;8j`5&etVp7cQsR|Z%6Je2?l?0mFH@ilu)ctrGzs@k%2 zNjVpjG&~5MiIXO~AunRE4vTU>6%06?T+?8Nr*e}FK3EJDmzqf+td_2y%0dqmB|*O9 zuZ+2f4)SkBIK?f|Lz?8Vo5<+nZ2GYC%6SJ!9Y}$vT5Q#nUUtJ-HpQyxX|z3p~wU{ zc&`&uLi2#M?7v(l$`FJh{!Y8m>f$C>v5A;^)_w&{hwc0Duz!lPD&Tn^RE!MdTLKAC zu6LIx^?w@eb@AYbX%#XQ=+yHKdb&Y}^R1u%Jqa+P3?Vs)ISF|iwFQxlhjSjg$8oml zJRO-|udj`16}@?#dxtfh!v=r;==yN{<90ImtcX?r?at4iiJw1HZ z8xs0evP6qm@6Zz7@Zh^Z5UFD#)tv%gf7MdM&X7GR%DVxVS)fEt~R+ z=S-JpQsx&L7K`DNkZkPi4;dxBZ3E~P6*YV?8z}Xxiab3NOBj@cxRwxG#KdjA#lr{% z4edcA@B!{lPtbGtkHuxu{_Ow!+$*EqS}a)DU|?R?>0A1hw-@GCF=9EqE}L_OBihf& zYk$XmWz_vPs7kN((_7=hva+)FDtiSx1rvyivNFd1EI=+wJC#%l41^p1N_HSAgP0N< zK%^~xkxo!!{*?YHjs5EJ`i%8|K>U+z5TJH!?Zbx)1|p*L)YQ4*a;AUf84y@s!!Jd5 z{k8O?v(soV3eB;F?%!YTzw-yUS%l1;kTBvoffS5^jTpSAkJ$$HEoR>GeV#(gdw`_k zS%Uj>IndAZ+zJ50O2==e@~6Iim8#Ly5Aw(z)=<{U7rRM^@_7QK0B(b3V8{?jkdV*4 z$y$J-LZ;B?w77wLdM@UV&wIK|$0d|RG~RH6E}P}bl-VT4LaAxfBuBwG?B%h}2%D45 z_3nB8v5mnwh=x~{y01`oCB2pJO57g(|H*IsA{xp)$>hB#g;Zyn>*XoTi|pNJ7dC$v zr=xZ62~YpLM0*5roX&lqk#=A2Ggli?$sCbMFT5eN`8BDgRZLo>rKbaOxLN7)jhW|{ z^uK-1p>5Fx)?25lv^sASt;3>X(EbtF3jCme9vqF?LvrScHmf)t3ZoO4WdCI+{fy92 zSjmr1W@Nv1py_pv4$D=brONC>(_VL6pk8Js>F?i2+asK7X-|k__*^2TBlpV{o6M$6 zW1x)d_lg`D93U~|i9J?G0S3=Tc{pBfBw46$eZPZRPs_^ME*f~tS)@q$L{!jgs<8!m z7K}YUF*PQmGiG=Q+!MoY9vT7ctixIy9opBTi1*O!tge>qbYNkp_gmlk;rDzif8+?w z)BAQ*BESFj1%TiX17bIeYIy|&`>N&K2M2-%1098+o`}{Wh^z)57^oi3BKI5xIwjl3oM3nT`{9H0P=6nMoHpxDO7#{QI}`hF2KG!!Ri-L3)W z^NL~PV{uLKx|dDUt$4`dZCFChpx|8jf)C)jQfKBAz>jHbZh5=EXZ>Tj2VR|`k2T*X$J#=eq&Us1WQ63H zI9&eaNpFC{v{6qotNA|NK0yWi6Q~jyKo-VM_g<`^NJ%*nn?VkhdA^Jef${H24rswb z{w$uFk-WFI_AE=JJjjzgv%nWBE!!h3PiItg1r}gA9{Uvq1!>apD%METUHWHoZXNhwhubOoKsaQ0B($p6 z#KRjWCLHrie03wR+o3U})KA>}{E7V(l0z=;?d=VRvfFMlpAw17IxsjGB3NQ+Y5kxl z>^TMhKT^;im=U+-(yqr`WbC-o_9$G;YfWvSd78%m+(zWjhka)%Tahesmv>S|zXYKg8sxxn21V)Yl|NC;mt}>U+3r zU(%hBb)vdzU2VP^&Di3Le)0S#BD-M`%*HBg|6~0s>%(>l0jJqlt0{f2ON6nr0s4|G zCQY%w4`#z!iLmC~p^;`Z{){7CF7HyXc)tLsCJD5PL{YDx{yxO4BTFNi?8yXX1!Nl-WFpE*?o=lQeo^M^hl z@2G3{*3D1kNjBO7ec%5MZQ)PIhPt-IcO{=D!1tEB!8(NnF*so3Yagya zxw7?4rMBe1Q#2&eVWps0Paa6nbKjaa#{)W+X@GV%H@AIGw$vw&S_1z*D1x+UL3S`* z*B88-TDLQ*`q|@tN*c}0O;Hi;Ut)dClYdvQRiSkVwly;l5oyxX2W!3icNoVnV($0X z0}X3ys&IvFZq8Yp)-&^*<{r_Mf1eni;MF1@9vajBa{Xr~)5EGailCD40x!$E^UTXh zW{*}j`M;Ho5A8k#!tF4-=S%Vc>(~0s%_8)$IWYMu&;NfWOU!TPqGa$>amaVz|J~G# zyjk>bR&V~?UXuSawiYiTJQ}gU=Jwt~jj`mvgbo1Kv-kx;YOLSg_s3$|73^wJ7f~>iIM4wuU{=@ zfxbS`q!ceu$l+)78>Bi zw5fPH9N{Ds^{8=dU}y+~wxz|JCaT@?D$}+vh9sxyq?gD(JTjbxx!Ft;TT8G57tZq3 z8R2)ysb$~b^mc8=X@-W1VX1*`nA|f3Pu6{vRT9?e_W=YF86wRgpco z#{Q0dmAW)_ze@UECrITJo0F?c<|*%N?aX|oqsZ+Fx!6D8i6If>HR+dv`uW0oQHk(E z!}6LW3lWMn6{TffWarFG@Xz6bAuprb!2DIxS9R#8?cpVogwDyTQWP^Xs^8eYzJ|fW z!>f;!wbZ0;iwpb?&k=)N=E&|h8n*Q6dgTcF)=W~&(JNGHmQ1fVcI&>0b7e)y#Zp>2&U++2 z-YD5rP_Xs%)M?*aqeLvv>5LZDm>RR7H$7W#N(=vePieN@7v%o*bFNAQde)OY4_U`y z+%zy{QdxJ}I!-9cg7XVdIcQX7>pUUU&)Hbx)^=SSzD+*c5MRy8c9tRz!@=&nEXdUK z{AZAebq?zY`A8BVd6=1Phh{2X`<{1Hwg1VCPcHmn;puuLbW=zqe7lL|D=hpiVBQF5 z6-TAM*L{))^x>Y@Byf34v)%ID8R##Q<6o^)mFJ8tW2$L-kUx3u)}(jT-*t+cH$U#z z!Vu?~{!y#b8AeM;YkZq8aIi{{QFR6v{+AYJLHh8>D)}=Hh7EOz$elelEfgzN1j7!7Q0qdiT9)0P?$KzF+g$7V2Zi4-sD1^J0Fx z=8p<3rxQr3JMDLu+Up=7i#wn%i%i|>>bI?`eQ#eM3ZZ};MI1H`juY-^*D_~&E$Qys zn&tTZd#z9*MvqY&c0Ozpi8o>nW|V(k`zI?N2tx`_Huk;5`NT_xr&o0{EZJ2_zp*o+ z1V0lP-1Mn2wff9Y44ug=@cYKZD+_!r^8EG_0w{2BNLM4m8(c}~wKixttb~MpjShpa zr_T7`5|7vRXcEz66cMD*Ce7*2UU2mGyd4=b-cyQ7ky$} z2Qd`<)Q4|ce`X^V3rF172ZFyaNR_?qizZoUvrYI)$474nl=y30@@fVbOeF1d+gvZ| zNx>8oiQ|jgwUG0Ok}QOl{fb9U>!tPXnuze<|9cvoEm{1>EjS)7htLp%Cx(U?X=y{- zzi6klesDclY#!>H}s0|B`s z73*n^Kch4zGVni!l$Xo36)Fh9`b!c9F&Y6cewZzi_@QlL#oClEOc}=@5QyKYuk%<) ze)gqaAneVPtYOYFkgyRJ>#=v7xb%g|rioUA7ZenP$f%mB;l8v`#2;QV*(bDhpEUQY+L`4= z!wbAL?`1n79lV$8Gd|%)!bF zL91%m4Uy(~-ryG&I_vKPWx|iWg4ngR6`KnBnNH$Q&_W4>rIjc6DX;>73*+n2_m1vO zlfRew2(!ZHq-QA00mGX*p4;hP&A80Ur)wJGb3@k!2!@$LqJ{4@H`LWpPG-Adewn|Neo6=4e4DB#MQ?IVy;kz7vKgMi_d)ba_!keHY__vfB?tCsso+@ZFeSH+p5 zaFF9ObRYlrf@4N*i`vdk`B?DKLTk5j!u_gq>nnQrIzwd!h0SuEe&+A{>>KNz714tT z7A7nQd?k~VR3GE=I*Qn*nDQnEGn7&G#E=oJ={W4jWvzRkYens_n z<};|&?fHs0LxD14F(lu~PRUG1#5vO)={ty9WqzFR#{2YY9mJzEO9m3RMJ<7elAKA zP^81(>cCDSD!fF@q|;rH3gdx#F%b^k94;wAfCw#7k!E;2AeO4UPydzicOMZX{1wYQ z@mEyiDUQCyL4vg1*0DucfXVGO49Huhi(nsi-H+mLQ=5jyJ^R^}`_SFMT>$G~h~*rj z_qQ%4k=&}P>OC~zYJM;*fE$a`d-RHI3YSJ!mF+QfyZ{UuRI5%UU2;%r&5Bns(yaxT zuG;%ZA18C%m5Jh$Uw6ZQ|EfUhZhgE2^i^z{bpf<~U_Vot5XrrP(aG*7>GlZDa?-DH zREO1a$o>AndP#F3{-64yG$YLf_ez}NHTnvHwAk9Up=^s364$2RQ{3 z7llM|vw_#}sxO9PvMUhtoK(@tFzE&`m8GQnIz?IQGJ)Syk`zaW3FgelP?r+M;JTnc ze^m&u-X`(-i6tFyexdMk(;-P>wc7YKWpiWWeB)v)FIR^G1dpV7j)i0Q>oXrf1_YgmQI8Juz=a@ukX zxwyCzg#T?eWx)?e4iL>&iH`y7A8#td1LF}_c;#eQ|L31)3KdV1)=KX{pcGjBhw}ZS zeJoJCHDG4#UnUmFM+r1JRgYx`bS&nrB*moo0@7^+_G|FmLhl!pJFB+po|eEoRGm3I*NySlhk9#UFVo?pCq zMfQptd~!Xy-gMz;#RmB9)}gwu>R!6}&;YU*LiwE5Y39o|Z#p`0)tg>eCD^kAm~9`M zYRnF3fsG$@mH(X?I02_K*)v8OZ3s0PX7YDNHI8Eu);Qw7i@QFgPp8_F{z3=>$;9!hgbnK zof8#I+{mA9ZBmaOKHnM0q0dRV2L(S%`MSB{kO528&=s;E^IL@XSx`v+b z8<4hXk%>WmWntF#grWgT>T1dq^6{~;F0O!&f*mp{Auit3+=vD8?lovGpnt53FC7Mcxj{cbD-r9k(;a_J_&5(1K(QbenB{F0JMeq_0kj&g!P!Ut=B zt?a(_S=RjUF!5$`PpkPx7I3+|c$_}dGc)l!-&RZ*0K^+UL|MQ`|8=G=MptdN!gGTH zssK&Dnc>ur#>3V{z-&^Mdt&xo=0=4+6tLvDx3+|i=Mie9Xv51|2b$o$v+#>&pF-NB z*%bkR)CDC$8%nwFt%_=0pshZ#l-%A0K1*~!q1ePCdI&Udm*%WExO6V9^ww}I(&7{Y zDyp!|jJGfiqNu3`LF%v1IlyhMRGQ(wBBpO?Ii8G5<*>l7so5n_9~?9neS8r@*giz) z;Av|^oFGRWfj)c3?NUYb(pRywZC5iuLm53l+~74jIVPy}a928q;`RVG<2UY$A>c{+&> ziz@gX;k(;T_DB@`xkwNczYwsbNzlnqnZ zKV-6TQ6nE&Fgs_P>4p&9-sC{{<{X?u>RQmcA@ONlgY~}VxL;-Os@mQFdE{F3o zhz>PXZV_iuvK7J0_|pvmYOmocqk_Ctv+m$BZ6Pj_w_VZhgjoO5weF03VrMgB*I*c& zREs+EC}2|rR(Ljgq-8m?Pg*PL%(|8-We#^*4#s1_Ul{=)MI|gnPvdvcC{%8J?Cf+r zT9D3F3YTCvUh|g31;*U2k@_5%FP3>MIvzbetTNt=dHv`~_Zn-0ytNUWB<1E|J)78c zUeTL`bjKN`ZjNkkNYUqr7^B++@qj;QCeq%hPA{lO`C?CFKDSLV8AqZ3JSZ^R=S%*S zz};mv85x<3(8Kh|$dzADsJ7c(?i0dWhKHT^(Ba0Laj9v<#F=CwwvFaPO{>k05jpIT zvDCZk;xRcwA=jw|9kfHWoAoSJXqd+%r@hts+4yuO!Ip>B^!B^Uy!X8!> zfhdPJH$@~WFV~Dv#)){9XU4f9_=tYv)*CFCtNHXQMnf>~S`H`cvh5Q?KV@upm{4_iAw_v*p#%=(#lh@rj74W_96+HL45hO=;R91m|4Gn z#U=uC%IOyAjDNktGDytlm*+*T#e1gM!Hvz$YCy$`g4_3jDw5sDHxo8K z7LP__2@;U0WX#~L9RZ+BZ8JB#(=Y>vefX-4~IbF$MUI)A%55rxh%)od*_AIzm&b+GMH? z4F&Qkybp6zK-(gDu;}ju_~0W8X<6AL!ucq8V8E3nI(^7SmW{r|?VU=PawZqvFL)?W zQ4yls3_UDdYCsA@)K8CKWLa@241rK+kTCsMFH-lTzwx!a%V|GEVj`2qUXU^dGmU41BM&+=KQTX zP28=IH$tfJ|{8v_1bjirdTb_F-^!xTk z=H_l=+ZlB(uaj}7rdC}c%TEXsJ;wUnt{WVv{to3+g<}3nnJjqW_5j^E{xp7g8l&I6 z+_7jz4h=DR`%X1m@e5V-x%(50#XstaaIvZab9N?K2Vq|-K&mY!ECi#HeZWL!qZMDt zDEMUTq{|hLi-3%bkC#Sg$;|KKV~BZ8E#@B}NnZIY?wh4t3if}EJG(?ZAMc|haa#ZL zcJG&>v-x8<15Ib=AZo)qEiLY8n}{6Lee!9+5y3hyvtzc2hf4toa0+c-G)!A9txtT& z%UiEHVC;@vZo+Ymb*(S1moKUQ*&xVxyaH?dR!NE3ai6oLS6m{ya3QPQ>-H*0N2N%a zjum;KRI5Gl*}0)P{$NNhFGL2l{%BzlC^-7p?v*cj7N&`nYFDqD_MGYi>WmGx{`fo{ z*%3SM>OznB&CAj^7;-XI-G~5E9{7+JQKRqdKuWcnBZMy;uia%;9Etzi#!pvhU02uM z*;P-O43`I^uK&UyW@;A9V7Yc@)(`Q9;RjUUvmQniO!w|fYS-VHq^sQnWNHBaOJ#}7Ua>};D)@qhu zb&{Vq&S}F)_Rq2$Yyd6wE`VcDoVUxWfSTbMw%9l~N0jp;A`#f$6$ILU{!5r#n6xD# zW~Gdn2%5*Q@%XQEnw8f9>Ee}N0p>rqAiOC;+Tw|K;|LWuI0SbfjFqjKkbd>H;n%;; zdjbsT^()a(52k10lmYNbQcD1PY@7_IbhpSq%Nxt?8eY0)?#ZXC4odGwE``N>b1q=i zfcf_*#-9q%wS`2b^x~$))lXcN0$MmI4F9%nqFxgHR*BBall|Kx%**40{Bnbqn^JeLm$YhD=;l$h1Z*)oirnx{SOY$7lZ4Lmw9wLM9!e5dHyTh9YOY z@Xo3kU@N18q&|r#CnP6VyG*%0jAV5}=PFUg4l>|WOO!Iv(~~nsbM7FCnwTEXG^WEv z9_~z#eIeSVJQY=xb?s=J?%WNX;pqR@?FLwz8I7==Io-EKT-m%%a&j_xCcos1Im@Jt6g1MconN37Wa)r!}n;R}<@5@4Xjbe!nfPU+*SLmHLLD)xjxYDXx zPu$w}Y#QwRwRFb*c;N}?Au=tE_VA{VHu;06#UI9Pk$*%ys=uWJDZyj`vzHc zAF_D5N3B}Dyu8fvVG&}E|A(SLf4&(rVT&p#jLqMSG&{30Qu5G}l72~1S4*&BOSHAK z>y1cG(D@ag5(5OuYz|7D@^YwRqdu?s+Wkj zL7{YJIE5#EH0twQPL4lz`-|+qFI?Ma+5dB#L))KI1z*Y>_1WFyDKrE*y!Za$H%6mt z(B!HioJqH?=~%}3Bp3{|LdBc;OP^Ch8Vln1rL@_Vc=ZGLVcRA)9bJEh?DVv4_mRtwqJzJ~6<8yj{uA0nSlFSXUr9wdO|_*mj(fto)0myB`Z* z2Krr`jo=RTViebAY`%1^FC)$eEI}Uv}a--#jC^W3rQTXSK(@EZQ|Vz7&F%s`wY->ew^jK6>wCJzL9sa*REz> z?64*PliC0GKp)i1Jggi!-X-EklNP%Db+67T=bIf{7B&AS3T*Yu8(2|*G{wJXD`VfcL+FCKJX&XHV07&_y2mBCn$| zMbILpiq3Xs`0A7W6Eblj8=M%$j>;`YVHEI{5_1vZ?I=B9&W`}so#bs%V%*ydbnsSn zd&4ry09b!dyc^taL3o;3d}`g3VUXL_K^`g%dJMPH55RSEiU4=OR>M!D7UobwdfGC{ z1>I)@Y|F>Df?E-LXTzA(KK@)?;##dRM#KfUdD9?a#BmM^I<)GK?Z*zVKJi8TKyM6K zYuB1?Ng+7y&KBbQ)7XovvN6mK9a?_uBRvb|kaJQHX!@-WdIpjj+TmqS3Am@i0rja< zI;YpyfN%^7gdv%M(ZAzW(OW0g%SOuBt$NgVa^3+HwXBJxA*z~uw&IB0C zIi2_5yXE|8@gMNvZ$KcRH5@cZ?$cUh_bQ~m^@;qeN!g#oVy54X_a;*jxM|vE0!dvM zL?C57^K?XCi}AZzEbwROv#7Hug0k+;UK91`@0xUT&uFjoL+JeqMf0NuSUVGybDgok zEspE1SN@l@W~52GX!xhQ1Y6qq+x_!IMvdd>44`0m9TRwOw;(SD#{vql|GbXzb(_<}u8pK} ze#iT|B*Uz+d+KquyJ7Zi^LkHOzkg3%J=@rq4!7cVFMWR9Uq-Gj5E2Q8J-G7aKuvOh zUomfVcX6?lx(TwhM>?hWR>uCi4R0J+V7sPz^!EW_dmZZBqXvZ*j2I5ZR3|tO~ zvx*pomj5qKYhgqzYy{Uk(`{mcuTe8d?3r@|rrWH!9?+m9&WCm|!Sz$~pv7 zCcujx_;EI84(!-w<%bC-)2AJ90X;^eTBMBOS2PnUTET^`ki z@X~Cvcq3GvGlh?@9?Ac7zG7h}i9&Qc7FfE0e0giK+DU@Y;uxQqad_Tb?7oEgBB)H5 ztZT*VB1g#YE}L>{?ew1AwjXc<6fs?0U-IJ;)jRYNPb!-u%tDcqC?ecc?LM0s3?*M( z*5@6Os_Z4zZt8;-5aoI_`m5kTO1ZK6+rcQ15!kwDj+t3Q7qnDgjH)n9$>UrIUebFK zB(5UL$9=+pqi+`YX|7gTZJ zqQQfLj56o7JGY6pdUCuThpGPD7a#`bN@T=Q&1OQ~o^d9U+O=gl0dB_T{d+IB*h3!r z2lOII(w44r4LCFEj+W{T=5IF0hYTjEarjx>cR?gp<<$`V)dyoKJzH`0Na9R^r@58^ zu5btgpe$k8HyZH!b9X zLMibjOhSM9U&mFlL!-j*-(H)xjely{rV!~PW{hGO`+{EM4d+ewN6}$JOfl{kb2^%nE6n zyD*Li1hKyoTCP{+E)}ACKM;$WbRw{@QL?@_1qa0rE`NZJ&rheo8$lt6@<4}c3T<1` zCMnzj=lWxP=8;s!octR*bq6@`9Cb@VJ#qvf127x&iGr3pB583k=YhE*icLOGS;e}8 zHmD{$sGU8;%J_NFq1(}A_iXH?!EARDy$mca5@f<;OB+T>MCPvNE&E|AA)~e+VSi!7 zwHCpbb3RuJ7Rbupn4sW`o@?)ph1;{*g@UPuI5n$j?GV;jtGxbu0pp($;j6WK3Y&vi z&DTU4kG*dezdCv~+_Jm=ba)mu|DNnU&z?7-{L%9eu~y?^IMhm=VaDfBj%w=n3*)`P zUcCkMBefhMmqYK;n~9lYQ~C7PUuWOtzh|`LAu@M&GvDn?9(=N2N{AABxY|=!Aicay z5V#oC@qDp6h!B85xKux%;8SQn{ZacPcQ(>_iNeXMD;cG>v@7wFZ5AY|A-~9GOj&|( z!zM8|uA^!GLz=OqvdZh7{x;>GBk-+&-=HOT2@^yeLg2Z4U4G_ABbM}x%m+1N7gp@w zE;M(m)s)4Pet1c+X*_<-k$;c>MlMXe z;T)1|q?Zq$;3XTc`I((Eui4XbCjf)ventZ^msvEh!&TQHF<@Sc=A&;<5VrSMz(BEf zPo{$j#{ zryd9N<-g4HXtSuLNP3^E6t~&9@XR~*GV4|N=ORaNs7+J*YNriG(rC0nj4C>5B;4-T z4*~y~-1I0n|LG>=F4L(o;A@ZbMvH$nSsY&$^5D) z`#XXkqwIx3yeDIu#@nvBv-c7-K8cSK$X0qhqMCj*+l+v!!BIWZ1{oD>Z3c23RKY$} ztxl`%WHbe+PwJ05Z+a+mgCf#0E?&SFC)3GREmu3iWdw50YW#_c&)`~i&!JD!-&g^HKT6Nq;V8CKch>f!L; z`Eq}MT7i+&c02Gx*jc&&WWadm+Rjn)o}Au6rv1w1So`A=)*@HQZ|Ph>Ep-0gF(skd zjuR>Hw9i*A^m1^cW6ZICc{%J7DHtCeuTa+iro8RpnUC2@fKZ#1S{6@4xl*V3g;hzuN3bF&8TA(FnQND6c zeYYE#@TwpJGvK>QH0*oD^1d<2UX!{wWE*t!uuG8UR^8cc^o;)F-0zI5LesuqfC6G9 z$bNe$pX4h`M?RZa>_qdVr84eKHZ+|kvYp%bkH71G9v^OmL|KA~k0mwA$%(1@=~8bz zfVc*lz%{>Cky2qdjT$N=9UwC0^7db}!C+;EE>njl(+f(>U?5+AQ;ai4qVz42C02UE zhil`7emcp9y0o88NQZfsCXz-3$iD~KM<+KwX?n3!!HQDnU!yQF58&KN%nMnJ2zs3q zGD1r9QMjF{lVJ1G?JFIsCI@f6-J{KD(UlzJ%ZzIccw>E=z-7!$@u2G6zSpv+{@PT9 znOK>pD}O+Sn98wqK!l=E1{K~O!|qB_jnTj{g$XWxtiNP_hhx_33eGrb8?1W2M|QzbJT^>E+*HIk<2t|YhhHtUjIp+t&Fo0pGBHM!>Lq#hV?8P?nr%4*Ndb{YO@F+61bcxTG?@9Vp z43!nVfm_;fh)TT#bY6A=Ezd`=paVOTM97lAFaP)sQBo{aKwLT~W^RjYJCw1cmX9h^Pb&J-%(Y3Gt@puH=0+MRm@%FM66!w|D+P_+GGAFNh5b9^ z&D-Qn%BYWZ0yTQf7DWOGfA9h4Qw{OP+O(Vl0Wm1apPOKCpRmVIm#Q`Mc^^Ie2bA== zR@1qwyNYsHpkxByA|8U94be1^gbGU=5Mz67j(;H&$kz4e*C58l=gsdWUN$28g(Vz-j= z2{Jdn(tH~5np8L13Xah|4?{sMu+b2qB!dySwP|bG-aAicf2Y@2afU32!lOtRJ}cwe zU>mSQZ%+UI@B_bHXD(YbN4va)#kcue+r4|j`Oj7|iXIA6a)U=qztR|@7wD{V?kZH_ z3`|#&pkg&@2ov=Ue2BR&GZW0P?L37{)Fa1sDfJ1VtP4)OIv%Vqkm3&DV zu_|6>%dUx#+{-CE8m{|}?gsvdN3Lgu%s&j{7?ERjqSrn;y)nJHN0Vx9fs=J(mQFtJ zBLU#IUt`==hBsfOTIhZFSeg2kj-~oC!zH9$-K?;-dEz_ejY){RQ|4pRO*Cz9$2HxM z>he&iY?zNxc#DU>y&1#wtLN-!$8>MK_lPED@kErs=Q|PO52wsSpD)>?8*2?m@uy#{ zi>nRXLJVepRLt5XQ~r~ddl4HpZ5moM9EE8@;v|z0LSNq6tZ0Rwu4|wZfpuYF_YhKX zoS5Cp*nUZO6Qhwo#1NLI&u>VfG_p$YeIN1 zAG1C`d07w;h#2Yz_l*`V(7bmCTKJx7ym`11&wZln-x;{}on(BM7HPwy%y1EVp_F$Y zoona5^|;42u!%dZS6Qu|OUJqse)Um+NVnk&743HS$EF!9xQW=ri2boN*HU*jNX(Q#(%)O?R=NmKY>17Om7t+isRT|?*_WC;3bb3QO+20{6 zL_ka;0R2$!7@pCFIGbBiL zo44BnLl?5KZF>t^40~>&(6ojSNl$X{Hd2A-OpK>o=4*~qWp8BPThm^=N*$Q zB}2rd(ahxeyk7Ae`D-rC#fZB7YvuxU-u%2G^tv3MJploakdSjC9TJ$SE^JdIn5j@z z^-HJv77UaAlBeI9*0`24CJs&7{JWx5EypngB(4-j0_CuyrHTl7?aH+h-bY{tsclT9 zdFs9^#P1tD-e>`M@GcWDYLX;N;+?B*8W!vCPqL*74u|uCinljMU0r8Njt`Jbd{I=+ zsAVYL=xcR9aEUs$p8M<-SLg&vrk8|Z^O`n%p-i{@uJ3?p)=`?p_Bm-)d}l`M=Aeuh zjy@rvX@#KHUo)9pCo$J}nqKieE_|MoJYGT@lV4C69=?Qhf<`&5mjqj&-O??aJRvf|$lQ52wT-P4BgViOsEY;g)i+zcN^P#hsZUpSe0RE-!#BFl5$vRfUIkfXtPhA!8NvkX|)-bDS4@hE~d5mHVl{RtVC zG=l@2sO#_A)})JAO-&;VkTpSkjMriGEJR-z1y@9R9jp%X>@7|jWlDfZDN0^vf*bu? zjn;#9G&eHYrKLQ#IuOmN^5fJXI~y5Y=AUkg*lNj8$|3gBaLgWJM$Ih*$P$B^}pe7Sr!YDZ}njQ*(TxtbB)(1qf!?>k#VYcJD9J)gtQxv9b~OF{t2LF zL(MT>&b8Z=I4r&*m%u{2xC++!vih31v546Ihxof@EFY+*=FNrqE1N$d9_^F~%hVWry z3obkd|0WaxsO^Z;=a>xK=0y@iJ1E7Hn@mY9+}%>H7~-mg)Ru@BB#3uc@}qMRV0avD zI$+`=+!HESL67)TL1}HnfZva_8fPdKYJsQk?p&^UTwGj^iHVg8r}a#HL$9EcV9uly z=4=c7OJ8(+z@nIg>b^GaP;0KbRbgQ%nUDmyzMxwmTfIbOvL<`r+rtzM+(*WHO90Hi zFekn#*zz$U9T-;lz(JN^GVANnp0t#d4AQ?UDGgVZvQgS3d@A}y z%OPjqhM9y0R602WV<~==v~W{vtL8xVwG?TpSqfcM_V$aXoiQ3l)q-Wwoqft*!hL2& zt}U`0e_LTyixkt3%4VjsV=5h=wkbsJ>DOLXtCXH0Q+)~6HBtKUtxT}{?kC^mg83e- zj?6OApL-hr@=^NT;%$uawptY%_23lGSVZVHIgQATL+(j1=Z&db!p{TI3a8c11OB#U zyD-g-|MLYrj}dsmd9_<`6Z+MS7tntIecYe{5?aSgc5`zRD=sH3YxeLFgy7WUiqz7n z^Xd18kawTOKx;*uH^`ZPE*cx>6;qoRAfLn?Mg~VIK=PxG-Qw*SB>efK?5t^Q@;xfo z!IH4NL>X&iCiC@}U^+KRQXqs8TuYDhf!|eId`w{=!Bsl@68u-SFvK2ah+`J&ETeF) z>8$ctXiCBK8b_+tknoA=;`f<-I+DtSR%i=Oeod~(!3~ikvA3BxP})N+C(_9~9v&m- zDY6odWi>D`LPy82yOKTIkD;$xcAs1FZWG%yuSRqrGcD5;vXST{fyO@@Q@rEbcP}b!^Lukb zMXn^lV)Kbfm)mSL%2z`=EkdQ|pd>X^o6Vf}Pp zxr79+B(mH}+wv`x_mFb8Q13I`U**0a!naRlyL{7tqZ>;DMDR_8vw?lNUT$i_t;b37` z>`n-?2X623NzjcQ;L-3|?dLvosn&07WN9_m5%Ib1o;y4~Lgy)@?;Sw;SY$;+o(E%x z*x4tMgyn!Y-gbM#>*6_XbPxCip}aiZq>rWZF^-z!PORr8Tb|umWpnM0r3l`QHMJww zm~t(#H0r#FAP00Bu%+_59Hu)IBVhy`caMU?<^Ds-j~zUbQqh)Im*;m!z(AlBfxpG7 zI;vdtx3>wV2lo5Z+bfB7JZaTx43ck2NXMI;T^^PMHrxCVZXWOYPLLb*whC;h6$rC~ z&cDq@NgGBA-rN6B*%`zhiX>bh8K*Ly?zefkIVmGFz+V%(?FPB)3Hyzgr#1f=HeGGY z1~Umc9m@hZkUCKHTfLg3e`w_C4svpFZYvQ+7@eFPogOcf5t~Jw7lcA|ce5bGWv&*V z8q4E)=^U{Mv|K`TgopHA@S7mZPaDLfbsSApkL88GfLjuhQer9$hXSFV5T3G5b=utT=?pv8muO@Rz-gJ6NdU`y(9XAH6maBhLj`2G0p1|sh zl*jYJ@|h{ttWR@uJmI$6`6MNU@#L{{8GL=j_cIj0jZM2Os_df=3qW-<6-9Rl7jS#o z>Go#M&JS5x+22$j?)t2Lf5nuHAk=jDadX@ng!*18R3;$cj?c-m+4>p9Cyw_2USi#E zdSGDq!}H=Vq-KqLg(Goz=0%?_=~my!yXZe;k@^`a;Ch+i5ri>ul4l2ze}B2XA?F%l zvi66Jm^iJcM;v7+0-xPy2cb5Umz>E3I0XJ>oPia?I)A=ZBmKjVx? zF#yOd;}Q~TA4JK_U|C&+{7w^}WX_w=Z#H#8K|*Riy58e+w0t_8x``wdy<=@rMvY0*|~oUCLKIB`gh@D1+-S-sb= zUB{%E;* z^-o6^zd-BW4~T)0O-g|98;aNgcvJ$%!#Q2opH!n~m*>Z`R%@fFi^DTB1o8dM{X&kH z)}rpR!otb#spJJ6?i=RKdq+W>oz(EJpf#P|wSkPKWAnxC34OP|@2MFce&Nk0V!sY# zH&WQ%`SBh#P%*pL;dp6eWVy@WqlSV`R(saW22lUrr$9){NQ*C|`x{K9ASd@>4nIZn zs{`HUC!g_EAyOI|pjAfKH4nK&gj85$t1;h%lHA`ezU&F-%iYY$-+DSa?HFBs z9W$qH(|gqyEi#C|+I9Qkq*P!B z@(j||Ja_ddyBZhGE1k_p&wLRvD z1@*7}`U1_E7#ixj($d#b1g?!qjt2zHJ7hRuA-RrVXkGi^#>>#cXetEF%2sGl48Jr_I zwzAxGJy)}*qxW9w@Yron$_$KzmY1;M69|b5V!pLN8r8m;O z9wiiN&^0&B@3o%KURF2>=rzvA_uYZqOTm{*7`^vmj+0xOK$*N$J|oUWTWho9$%=?I zwpFOO0t5}8A6%%kH|XHzXd&e8EL6!k;Cwz1;C6GkR2|gTID=&iSZ<*~772r$oxZ5= zXL@_p$~k-fA&J1~E^2tN57(-z*Y==Jj~#+?UxXqcH|K{IlZm&}1?)_Z7KFOB0%oG} zPik*uZnwG-)L~SKJNmVwUUnevXMb@R9|I>@7iadb&C#1L41BwY?{M4 zwA`ws?;4w$Z1}0u0xPp~?~25_wIj-e+5JXwNHcS@Px2zm%bm9+{wDsthRPibNaF*V z9mKB^@xe@^@&#i7BnuX>vRVcDO$-?#1Sru8SpJ^x%k!J@@#Jo3ZjO(MnRZ4DN$o)T zKe;u9dC30q9B;|~(Op2w|GyG!{?mW{zh>(&K!L3FEEZ}M&8A@qXt=qJ4-+PGcuWG5 z4Y*9VegZ`jlh#M%ADSGFM3ag!x`Jv$jU|&3={;+SAS7!i>+6aR@`N5Q^Xw-_^VJ9@q>13`$7-PY5Y;4j+3hB*t<^QEc&hO0mXO&?66b}bZ8*o}UzO&9 zJ$MG~OUG~-A+<*)ts-Tb?qm*_m%7J3t9VAO=S?Rarn?$BVQ=pCoRZ7={B(N<2XY`K z6BWn;;?AZfcFR#X(N3;2cY7Sj*8y4x0hPw8IaN6iSm@RAlqh2;i zSxE_#ghWB3cD*Z5!wZc_I!QTxR1s)`^!LYPm0i9((GizP?e#z($#R3q(D1{K;N7L2 z@1K^ZnFExraq&i;#tBwd_B@6E^ZL|e>gDNnoz<~8hNq(&Y^ed-eED?G%_DIL^QTqL zCwb^`^Dtu0{nM!IoNxN-j=O6jUZvUx5D4zXbg2Tfl`X&Ja6%Cm0vf?PlJthh+0~Gc z5KK(0c#n(l^l~62cYR}$Trwh@kTpe+9H?x|%J_;8vgYjyd@EqT)!!i&bY$lo@!-%C z(wwhwbA3)n!$47|^hb_M{%|2z2510Rxk;^Ba_?6r1`&Gh_wC%+NLflL{Ay{VSUB@k z{anszX0+r0X{gjk!ni&WIz74(LiQD&1)tCNwLjjw0j2%nTq}R<9@n!ay`GMoQp-}9 zTus(5CMqk7XG^eOT;oSmafiKj=OH?He2$kR(Uv)O+V0Ql*%OajQ)2{j5{a^okebDZ ztD}ScA|9KK>A`PaJKK6}sYfa$Q*ec6XJTn%e2)83q82zK)MzloXd- zjshJgrv>C#K(pQg7cv$8Xwg%&BH;FxVD4phRn>g1p>e-!=Iq@k5fKqI0q%OSPoMUC zYIHm`ZLSY_crOBxHiJ-zYxc+5cZQSpwzoOWM!~K>Wx33|f9eXZteBYW70;H{w%^(h zL55T21+%nlOa|JX5O#Zima|t9rA894e&FZ7T)KO6=^*H}4sG)DY_snz2^+_UEDLwh z&N3}Gr^5w99S(#3wfDV|2jhYtXlY&cRCUJOz8?eC_goe$N=F!At>qm7XRmrDt=&bJ z^D(~7U~r~f4;w8JSnoYCbBXlYH~Q(KR#}ATowYVq~EietLf?e z_E02hPH%|xxEcYPl%E16X-B)7EuLL@VnLC==sju+fwELvl)Tmd`2r@1j^lN_HeIw+ zx}+(xCEl*K8lC)H@9O3WmXgYsPpw;}q-WQ3&)<#sL_L}+R6bF_=V(lV^&V)urSUeU z4;qw@g+uJR4kU&pPE4>ew_i6Uz%SOCJq!*pOREiVt$KLf_E$=At+2dwG;Wv*wylW4 zqvDO1lL#maSf8E(nd&XqgEB`8bx+sLCr-x;4i^SZ$FMdd8Grupoit=Sl@qr=o<9I( z*g>fHk6T!g!J&_j?#s*1%`=}1mB+=r?p@sPFB8M-)I_~r+VpZ0tH<|(AOhTu!*~Qd zo;sCufE^gG;|>&%DK9z)faOWn_o&k{6z%t_%1wNeT*ks^>z5dqP=W3f+qnUkEH}kg zZK0tGYAjx9PoQ|qtrE8r#~bTxo8$A@oQQnkf!wM-Vwc;w34Nib^J3}k^T|#KzktA= zJC~|zQVeklA7r<3Q|RgOO)C4-sL9lEg9T)CStgDt`%Hh`Zp=QU=402z>FH|^E)Krd z>f6>@uNU^c-e9LyPA&J-33AEFqcEAD*{osk=1or>A|lWo(C#v zHGP_(PD@Gouu`&FTXY1a1qN$-Z1?o{_X`0QTKzGJhS$SstCNZFhpXdtKZFQk&fjSj za}9=ekb-_b%XT^NG$0Z##`rB*@Y){tlM@=|An9(so|R@q$*T1jU4dyrUYQLI&Z|Ct zVnJ`qnhGM8BinQ^jus|~f-V2A^4=?|sV$24Mo~l&vCyQdC`gwY1R*LY5Gm3-97Rg# zNG~CB5K-xZ^dh|%C4?SS1f-YHOK4I;3nc^yC3oTZzHy)K%N=*zm%CrcAbV#dd#%0Z zTyy^CZ_-S%)~dQU{YqVKH+Qrn;1EQnmC1v~XSaYZC+oQYyc$5k(1>$gIe!{=^RKnK zO5XVhaogTn3J5N>0#OV=M$jNL=S>IpJpmEMSn5>uCcT~xrSwSwpc2pNjXB~V79{5Qkhcs zVs73`^OMqI=0?g**EhNXpx*je4_SH)4BwD}B_)8)CJ{=4DDifI!j*6 zCmmbI+vV!)o7P%dTH2q{NT7BFd>a}#)MBbU>asz2QG7fkH9al*5mGdP@*KE{@Y46; zp-}%{W6PpKk_i}flnk#^d1E@XWXmIBS&8Es8#An_Eva9f3JE!{+VaCqosCHJorHPC2~Jr3(4IZUOq|jR(iA!&4~F=c-ld zxsS+7^$qp5dU5(a|B4jZ$sWJ!XA?)Lu0ibmN|wDJ)-EJgEdG*InSZs|nM`@lI=zBx znwy(#mQ?9e{-POzIqzQUIKI)~w>wUFVwuq3VdV%`7l5{`sm}jJo0=M<2H7tuSF%VMbdwESVtOrSorusa)~_-{}%Jf_Xm z<4UX1D)uz8P0`BA%Ivj-9G^_SFXWvE-Pb#{&KPp$a_2W~$8bR#Ya6}Ee_l4Uh?e&b z(UKy@opPY@4_>>J8;;(yAiEb9%<=cac&sdR%}cA|M++O207875+xN}Wv%aj3l5B8o zreQT0?05s2u9hynhUCT==gruw@>A0G#mhzarW zRkd5&glZa<4Pzbeiims_&8*4HZtp7Gwm{64+^?{10A`p7=aTnKiBp81hx{T<4|+00 zKg}|_hcqwalD7p6dUS7|&%SI#V}|5ac0rx%1xJ;IbEkrWf=Fvn?|q)(x)6I0*#mz> zO{GGMcyVMEM7YGDfknRPWh1&A-NoaURHkq{6Qur<8?$Egt2=hV^vt$J*Me@6|Z4SslavvtN}DkB-uPJTo%d z{azu|!+3+W9|32C5p1noL?j!{>JxCs-;OWBVlPE!FeF*L}>!);n6q1tE}l z$*9Okufb&E!^2-~A77geqHdTE>1EX)9)@*xhHwSM%4{vx*SiTs4mZ>{yf1m9ntrKE zB0bT8lWhXD27@aF!ZYOPIzFdjW-7zPo?i7p5_=P7Mv{DQF z`FK;&frVZ_P5Dci380~HP1b&}Qqx{sSEfi#Qu(=Ovp#dBZV#UYfqJshQa|N6z!oDrIT%t1Nkauj%Qz`1uQd8Bg3@ z9l-=4D$$NjA9nIbYIFaw4^p=0oXz}|oh^Op6h?t^C(iTAnypVo3)G+z8itcE8BhOY z=hzKfF`lgTtQ5`ci)WU%TQP3>kvCmE{rw?%xx{Q&)8GGr{Hu6aj$>2PuaIob60>m^ z+33W?{l#S|uyx2vOu?zaRIi-L@Y|)JM|-BqwsQyuoS$NJY!PK%krdpIG%{@!3VP8QMlES40**ui=?GJc34nRU2@%#wPTqcCxgoFNU52e z;AfaCX6%XjiZM~uY?Mo*Cp$YYZ@2WdXwt;$qBt;cvP8mEJm!-m3Au&8y17^rEZPsV z9Aq?W{cj8plH8k?7W9ONV`z@IudGc}%H6%U5vK|{duDHKq~>6Y$tguR1Mlui}auL3}zQ?0^F)=aCT(H^M*|92T z<)XVFo^f5we`KUDYt>6t{n@ac2SmO8Q9bht5ijx8byKtv6~#s9S5JL_&yB?SL4<@* zEl?2daH%~_FPxW}vYQ5`z202Q@2ZTxLK@94Am*p|3crMB09M@YH*;`i8-qN5A?@|(x3 zIZtLt^@};)^MXDkeBLcN&o6`n9Zd!4awKIy*FLP~oBUy9MgwBO*;xkQ*T@`4NTV#7H7b&Hb9o z(41WqfY(fyupveBD0j5OiiA0P%cc%}7`ULWKlQRQ_|E>~26k23_ZyGQy}LXTp`iHn zuC4e5Dvc0*6&!++6~a}I97ByScBd*j$iMGB9gN7YZBmXD+21PShiKNWbj#UZltGYI zZJuFVUOlh0(=Rq0EHR1KgdRuDRVO4QXhJF4^~)c4d)|AlS>q79DWervuU<7c+#I%~ zTcAUyC@FRj1ois`eo++s2@!HOtYIx)I0WIVGFQNHjg9SSd3HAOfE$Ucb4G?E*JFC( zI@5f5g4`B+{a^2G^no3`c?uGA{K)CRV0^Z+>j+62g3iQV3X&UXM_~KYp7!sY5Oaqg zxBuQ=79D2q@S5*s3Rbg250a~tr#%(yV=1XzF=k$_}K) zRN9g(nKa}sJNb@2S-+GV(@lAa9|lgk^=6$jR(TwQSZrjM?v2%6dx3Xz-Js72Y#-}; zc0l{p?B(CpCgH|H1PE~c#|ubLrvMN#%C}}OwF+Mp{V9nwTU7R8bf=xg2qi!@{qf7*g36KCg*~^7n1pq9@skH3>2U_hjeuC`M@W1$moAh44`-bP zToqq6McG^8VZc@;V3eT3lxAFOCsX~Hp5rJ6(fj38a+On~|CR7q;%?NBpruFzZ%F`h z_bBq0XNeFZHIf_p*SMjSUheBET8+qUKnl=^Ngq;Ga%sVY=}#JYl(~IH>ju#emN90&Y@OL0e!KH`A}e^ zj`p%)2F~d!rBGP_B5KaayX1{+c5#f9IQwS(-=99pd;Im}&|yE0;8qdz27 z91LdllCQC`i6e2;AS&44a+CM2jLX39Xt1@i=}s$IyVBm?ZfD}|!&dBfRLJ3@h_v`> zx0zXGs};+$?Ls(4I#Z=Vdy>6fsvTa4h;~(L6MLWV8D<=&K&n0M#$mGDjk33*F1=OuA9a~?+Pt< z1qJw3>%5~S`Hl{D{uby2l|k^PTwrSI??n4^Jw8{QJKOS}K>H7cy>`wmt0mdooo?n1uB8c~|=X5(jUj zIHZg!i}?lLKKS_d?+Mcrw)bwS=e3=_*>vy0Nm1#^p%L!Z{ZPdrSc?(Vf`H>Y+=xdj zCZe%7HZY3${QZcgH~-(oZ*3XBZE(P~)vwb&G1X~2DRl)$cv$WuV=f-`5j8)YZC*Sn{fq^zn2OQNTDG-04L|yaf0JaT(pf;s^~s5@|9Zny zcB-UkKT0I)eTP2{?ClvkTAh|;dE>vY`Q|C8tNwe{|5laq-z)!5l|27H z|6zt;9=Mc%bPzr3y|*unJFC=Qy#l?hdcS;$5;60QS}ArthFKniNo-qlq7j-N+a{4x zWp^|_WWSga7ZSt|R+K%RgRxDPRN-dU)~<~fYyzz7cuGfe zp6!ZR^6u^JF)^`lK|Q^D(l|-yc6aC!+_ACXClZOMh6B$QrS!77NVM~m>+(R|Qa>JR z!=~O6so};=51evnnJva|-^4>0xbH-(R5?wuvU|N)F2RnY z!|w|HwrC6hsf6M^+8IY4kdj(M-W4cZp})pgS5pgYaQ*&NeZA=c4e-ijh3d`p^;$)3vf%!T8+ak5N?T(Ua_QpHJ179;n3%YgO> zIIUJ<$xIjKEhFXY*3$=hl*7)SzvQ<#X;6p(jX*HFoz>2}>uc^6<7a3z)7d(0Xt9b5`J#{hMm z&2N4Qzjj##GCi@6O@@@@1-iAz!V^cY1M%#sQ-RNNA6Y-jq+5K&sB{QJJx;85I(TUP+W&&^Fy8qf##bxPTI1(FQYqE=T z97kt`W^3auBcaOMGqoPboLpW84w(TU69Qca)OhRgxh9Vv|GdPlQ0~iq1NRpG@~%KFhWRT8a%5Q3D?!I5nL+>Z@S+O}PrFDOBA>pi?3& zc(!UPAbaPIuY{oBQ8NqfBAtD5Z0tj4ws7)Pz`-Na3g@XuA_DO}$fMs5(*4P!Mn;nS zecn1z=?C3%W$$N%^$@;*XRoLOMPC47!z zbWYcR!2mYPrU8kP27(gQcdJHKiBGIC%K@mv_S|Gf;g!#^x~6vw&CNBC&#c7Bau14A z_46vL=!Eb7X2IUDlDrNdy?5_kfgqB}kxuS<$#xq(%@;2S z(wi6Pw(ZBQ1t5@++(!py0Ge-QAzb!4Ih@9xr+=1$Hsbqi)L4n<#IyHRC==cQ(Ta!& zHO<8W@?$F-8@+~6-g`@%t?7BL25Lb=tvww%IRif1Tn(2IaZWw^Zh1St@?ded2{wOv ziYg^u>DIiE>EL~i*1kk0$eAAs=~hN*b0WyM5z$(j4>tXE^Gan?*Jqnic9AV6LV8@I zx1!sZDOr{O+!3{}e{Kp^`?Ke4v{6H;y z;U|)vR~OM|H)_3Nr?)`oUm0F(lRP;opYHu~Ch8{7l0;W~dv-?mD`~x&-8)|KZLbzZ zQ}Devzgf2Cx6dC|O^KLc4oy<^LExd9#lp7(158!QWuQ@bvkmJ}p}TMa&A=@h-*Mzm z-H#PD$MDkop;lYIUER#pE?CCqH9jt7mH9^n$gV)>(091B1V!^XE=_yK{N{@p5`KF6 zp0T6bpGcShr9Fo|ij8?_C?JOUcAXbt@JrtF&X3=D+5^7j4l-!#!qlP(+6xY*e2t@hgad7#nKFEzxt zEu!nzlh)SObncs!d(ffCK@7Nv&~Pqke*kX+BLe)f>3WeAC|`V=bIrUb1j3NS3&hhCSZ;gMK|Dx$B;Lz$e48J`e zqo~-VConkuqX}fD0}e-KYAN|YzK9A4EDbfp7ZtI8mBOb;ZU5*_t?R?{C|~ki8H-W4 zgpg!Yx9Le6MmrwR$Op6aL9cKyYI`7zuh!d>)wsE#WHtPDPU1b;h5G7KKqtC{^`b)L zy5oZw6kZR9je0=69moAY!#ZG6Z0bsNB<%u8@o#RxkgOOcE*3|7bi~oa3sfa+x^n_+ z%<4!4%zW`2G7vCo3aU>CH1>)0C6bJax1&Us9!JvUh0zpocGy7qNR3 zPog_pBmmaQ;}Frj30cP(w=%4qshx3qBs4&A(Rm8P|3y97Cfz4-sQnV>1af^Mv2fRz zLRbcEy#UFYg^8)u?7+s~U#$V@K8_X!`VXHMGwh-m;_Z@$@2VW*lj79oY+t^dXxHmF zX=c(P?3K`&TVbQ|A+=KGn9+`r?W*;R&~-H1m%Ds&E02jatz3R_)_KL9Ve^gt2G(ns z;9t#8D3!kR`C4MQ|(48+@NkGMJA`TihYv0v6Z18?kM4{+WHU-|iZlxVJaL z>uP|-<~C2{1H@wLu876)WQ#gwBH*CfThW+`>QvAPJMdznn!x3mi5jd)mBS6-0~DXg z&49<)4YXMsr(2DLc1l$(Ki#Iui8Kv)<-Mi1Kt-E&B<_3Q{{%(vU3{IZk~J@)h8NCrH}vl G_kRFzc_;Az diff --git a/frontend/public/benchmarks/tts/indextts.mp3 b/frontend/public/benchmarks/tts/indextts.mp3 new file mode 100644 index 0000000000000000000000000000000000000000..92c5c4679fe6b2d2389e73d45e2fe467741cb076 GIT binary patch literal 19555 zcmZ77bx<4M7bx%q4G`Rd2M_MSUsHV%Q;ng2B!xPkvWsWf0P7AO0^#sA;; zKmQ1T8jt5Y+%OV19J{w=~r+XAD1{Sp%s6Zz@ISdxD6ps%m*VJccNwfoMJDz`>iOmQ>+@{S%ioii%SYU3f^Vk(ejXnyq-T!Y zAe$yalgReXTQ~67e28QWeTAy3?N@1TB5lUHTaIM3-d8^zwX?ue+kz|zD}cjTzXC^j zaL!=DP`$85dx`~S>zwo9T}d}K>4l05aSk+3xn<@k%VMwwKup6tDs^jUrd>L+b<=K? zbdU|6elrpGy=vN^tkT{iFUnF+RIxrxP`SeHTkj1oSGF@Ob1oZAxwMC*6lg2)6~YIQEx-XBDX<*Uq){VX@c3Sp8MtdLLV7p zeVPMw$-SlX+}H`M(DLe6qAuV2y1zDuP(N(U^Ig4lfix`ThXi8P5gMsNTW81W&pY*< zb}moGMYFo5x<0-a;_JDs;P)+^b&Y|I!55Bok=%6|%w#x4d*T=7jOLIlqCwth6^2Sn zl|6nLvOUp|VK9I!_D9uj=0ZOuW~H?c_k${mv~YEL%c*`5K7HJ&3a!AJ>UHsHnU2u^`cX0R1?)r!^i6#pFTyvsylDd|y_VeU2?MB)( zAvOMYH7ORj&bQe$uumldvXbiqqN_k(NZF|p$+EL(?21XDGL&Lm#_*)lT%Gk}Bj@8r>!a2G{LhBIZYz4rSb(m>8)I4 z3AXFI_5TKLcH0pz-24^!a7MgCiZqh=D^9Zkz5`wh9}t32J!d&@+p zXSr;3%gXkS9HvZ6ltgM zidB@$(tnK39~>>+7OCiH+eJr@mI~J>88F|)LPZcq_qN!!PcfS&LgpO*b!-fL^^>tV zQQ4X$`vJI3f(Jv8;I573^*KI0B>(~a=%WzcnF_a;8?ZL&E&@>hEf3}xF2&0W!ho&> z(3~1c{CziUq`*I0<>lohrHcTmY)uTA=(l(-m(_W4Pf zMnZ9ErP2RZyqws-D2i1O@+XsPuy8JP?{d#(U!O9We|lZ_@?za5oXXan{7WVHQ3Xtc zEUhCoer2k}Sy_gMYL=^@Cd+=}R1}TP)_wEp$6>bIPqZ2K19BK#x%Kn)eYrX8v~hMf z>7ux(K5G*BJSx>;;T#aBB4?>i8F7a!82DW%3raQXC8AwnE*cw(MQ?7z47MNM%1`{! z%>~X;w{s76BeDUxM^@RL1zK`K3wTpHTes_VI_t9&qSZ4%*3x-FUOxi%SQn8bgxv^{ zG(!BcxY(8PtV0w3XnGLms8hPqTX-`ojwhr2cGqO2D&K-^T{sk95ogK2q#%K>A)Ugw-B#k%l_gA*uf+`>$C#S2Y%O!zaqvVdzB`a6%{zJYuIr^C#(O z4|;yD^ZKsV&agXCott#fHXJ zD3(^I_{1Q{V!)iAd-eBFW}r{ZWa&=G2z-JiO93W)tndQ`I01$f;{NI3r>F#Pu58@kqR~vaT(-Ykg7Kzcw$SgdwgB#o zOKqe2AXP4v8v*>Xe4!8(IV$&QmJd%8{0MKjjdq!li=BmPjnQSLDuWd-8Y6)pJP8@@ zTFfJszl1?)XsSdh%r8k1LV!n&j>g#o1EF`+_CxRqU6;Dj9jWj@i`0sCCq^V)x}QU2 zQF7e-&oQhQf>!in?sM>lXa96xRR~PI`iWtvO=ZyOg+e(;tB#aL%%L>0GDne@#{Sgm zpz1pMb}@TBW!JjAR5-j^YsnK`)n^o1KoO}b;M*zxVkVNV^gM@7K^ix(0w&Fl&uf!b zk`h6rtla{~W3PSy^B$3`|H()dFmGU?9n`QX94Rf1r$~j4TGZH5%Qz7e+i6M7;M~PT zY|@}puw93>CiU(&V^VTJWSIJZS=w%;L{rsxwm2lh(PGyyR457#lf)^O@wSheDY_#0 zvm=k?=BuBXUZb~&=GZjQHe0E!!Zme$%Ucaos{25|W-WF7k7um26O1k@C{1`qfpS!e z*^=FXQQmxu2>wT*{W4ADvj{rYjg1+B@5T45@Wxh{#b^+s{0{JwTCAZbLhC{MYCu&v zN?5buY-q@r#!^Xn5jiaBCgJ<6Zc*8Ww~R3%#0>Th91b^)D}Iz0ZWc<<|`7xq|R{Sx#zA9Kbwx^pB zO4RlQGHAN&Z}W!O&7<0y*BS~M_D?gx<~gOm@3DAj=WBDXaHk(?(^J!Il{jgSW3#{C ztg*?}f?XvCDEC6d&=*vRL%vX0U%dMD=oybcs>*MFE$M}K&mtc^e4ua)ab=Xtn<5m1 zws5sn%>W^MSht*V{lfdjW1&-73e?7%mV7Wgr*EE-sz`7i91|;JP4)@(5hh=F%vn0+ zvQ{1}fdXOiv@3Lp<4VMScl`ZyGt^xjq+|-(>Nr@lX&~kdcB|zi*^~-Y&IbNB8~(gU zFHQxFEaN->p8q(N1WYgU6F2H3655ahwYD zQ#j+*??Sh5IrLY4`$=Fu8rscVI~AsxV%sYMIV?7dDqxTi=8^6X%1&=Y7mN=RoxywO z%48+!8yz^Mlbt}D*bW26=-Kp0NQ+-TfGK|QH`zqzcQQw&?2qaq(NPi`&51|+fZ$N2 z5qulN_>u6L;n2%ny2&+WU>To^CqYw8ue(o`XGD{5iK@MYy?wD$AVuSGre1)&0)@zj z9S?;MQNWUN1avJGzwv$En6rv#1;V@guGL`FgqZ0M!xtD@`>Q{KUh=Z-hy3=|LMzQs zv7BYBc2q$_hFnkUf$|qKg_!$ih1^=JDXS6ye}i2#oNTuo8wmj5L)(mn=a7>Qj4&c( z6Vtk`_l#>=yy71jRrt0Y$0Zz=KP~touu?o_L+9yx^8R$^PVZu_RrsP-FN^9O^{7k) z-)VU~v#6pA+4WLL(p;z(M}@YMgM^^q<8ASx^c*~yjtiTXzV@c2#95-%;ukd~V&J9^)NX7`kcYHbD&_HB#bFKwnSU>r^qthdW2Bl2D0!x1gKr`IhGLbg1ZzVtd zU;o=iS2t999H74qg(Ir;ENs)_fR2+m#u|^~Rl;NNgkt^5qScwk3YUj<{LmPd{lTWS zlK<nU~9G++-F zquVW$Ti45G;7NMEMg8`>oh9-_TbHM&y|S=zDNFCD$zJc-7v@h-y(yW9EHqr`;#Anf zZ^T%aY}8~|*L$m}51Qrcz#Su!s@;es#_5Vt(lf@`AEWkO<3B}rW?O}HnP8eL8Ji=1 zhvV}$fR2+nQ8OpA&MuMF{#z$6T%D&K`S*f(rIqFl?4OARUZC6;UqZ{%LN;*b+jaB0 z*r8pKP|aV@1@V%B2K||7RrNtq6zv#(5EyYld{CTe<{%?ucnf<55fCy!iZTNa2OIc?H3P;X zMKlYC!usQ%hGqaG!ZQ#_Fxjhrk2Y^Cr)m>h6#(L=0uI=lv;)lJIpp!&&4D4I(A-LV z=7F_zNIsDStO7znXspffSH?{K$! z9*y<5KZOfBdZzHy@xvHX0*x^#8O^6D_2F^M?agy2t$WRvbp4~dqLj!eZ2~NBQ~eeC3Y8#p=>*(VB&dAhjL4Iq9$vTsmNHsX7Y~S>Xf6evTsGCR84r zgBPL@x{Huw8I2uMaZQvLJ|CIFoEEwlW{;Ia5~2}(U0HyS?hrOd1wjva+-3(Mq$8Ce zv*WYhp$-UhPhqN<6IP{JYm%>xNMn&vYaT6G8nM$U_gObXVn`T3w1@*u@+l>unw$6z zmYiAlzYc;{^uBD~3R8&>B-l&)nmTRe^!`RobjwnRc=eY^KVB*W?sJ|1FLf4D)`$~f zu40XIcDr~i}G*|^qiSQ8%mY-lugG`=cr?G1Jz89TMZGjLG1AF%+g;CSr3})?@jXE zyoU2+I+19oJIC@>_jqY4S?&I}$yHh;3MqoKow zG9v(2*TjJ}_vZTe$FF`dE~+kLz&+9v!Y-=h-TbMofk7*?kByw&X7@k^rgB+cRIe4M z|5<9MNrF%`kCp%3-->Zs{f?f~jl1sxze33HxU+=Vvvedw2`?ehn_*Jo2bQKVjF<+l z{2F8!ChsE?oT8Ps`d{ zWe#^f^Q?soWM?0u6jI)H35Dj&;Ur!#)a4*xCW53R*si6ocVxmgj0>gl?O!A0cki>Y4nr|G(@7pW6ZyYEhd=P?XqL4Uev zGz8e1Z1Lm)L;V`MY4vw+>V7e_lWVxgFuE9n!U3% z*pB#WhDr~`?bDP{3NdJV+f-P9pHmyl5A6BhIE3=$@HctP5WVil54>>sqydyULbGDxYcp|{r&7uw~OA#DM0F(KoPgm zgKsy}@7@iYU+;eVbVB#=(q&3ynau&}I3b}S%fE>kLnHyu%hEy%ant|)u2sEaII&y&l#f)mzUml!x0xRJ z)XqS-P{Hns6Snq^R+*57OAPaqJUxe2%Y_AZmHG%W2!I5@g#fGU2hsH1S01}LPSi#< z=<1qD%zz$M1=>9mP8f!{eRG2MQ|ClR@#N#gF88D^c1QFl!6&ypYRP4@uYOIKXO}zN zaZaKf2HpyrrPWWq`#dpLk;XrADC86}=I(HXu~M<~5&iOFWDL{Whhc_oBh=42Q;+JG zLWBBSxee1l&hOV}`md&me6rYPKlI+cKW}LzqvlTp#CR#olyNnPfAv7`*N4H4(ko3^ z(Tqs~Hb%;rdp89l#OHoAB4;iJE*5=rVI{wRk6w~QLqmwHmeAi(qs761AFke%%1)W5 zJ@e71s>S@BnySwjaV?}O&XbEI$sh_svmE{Ew-U>{_mbV@0`GG`V%xj6N%KFgD}J(e zDg<232dqOtA@+>cvEkzI60CwSMsfR+l0D^7C;kvw;AUwlwgg^SA7_zJ6*&S3nqa=K z$viAf9QqY2@mgMf!8eBq`DY{IH&)PB*jx(Wz8kI}Z9N{1YW?q=71ZbPHon>H`0|9|k zcaqxbY>$aFvZ(UgN}j{)>h@B-7X4%_cf>=`^Me<8bVv`SIfIfIAcBhNr+&JG7-w~@ z#-jEO@4b$0^A*LvWW=2Fqo}_&cNIS%bgrjga;hf7l5rCO!85HZ@xS~Vgs%6vZuZ~v z=Mej4>AH+Ce)kz~v+@LGF@wVLQac;-;{|zVCjTHKph&qES0fuk9V)TC;Iq}hW@4&4 zP7O?eLTiOYJ zE0YzE+YvZU;+s5UUlt{Ovb9U0*J*2ekP7C=?W)Zc2S$h?RKPi{yV_k!{5YPJX0M3d zH?CKzR?k|8FTMV=KGFHj)2O>+@%}#R_55L{GdynMRVe!V8O`UH!k5)eiL?BX zxiT+!rsI$nx_DAJtY%cwQyquhN^ar|+7Bfy`8Udy>yJ*f%9+fblY^2qcG`5wNHRE~ z5{Skovj?Q|td+V0{tUw4EXJB8Ch*5R+epTIC+cW!Fw)~6hkHyLkgWGg^o!~QaY zi@=XtdxSE{lr%R5l3qZ&H@2Q!Jn7Ai?;d8PLrVPHr&oGyj{XA_wOmf+b8Z5B@<$<( zg<@K@iU-nm3?M23O4bnO^GGExpDp`v+sVB?-`1Pfw2d9P;pdh9`P8DM zoBKoek?XD}aD6)1cA4`}RyG9MSpQi>zfjrRx65t{N3%-tqmFWplfbLro;GZ(ZO&?O z53a_lp0cGPcR6WkVox0v0~c-8`Q$_r2xJSNV2JRkFuiTkdc; zUSjd8aZh6LG2OlU-3U&gY6gzBwFs2!=W#2aypo!SmN|N!L zvaMCg3fwcTD(kl4a9(_MFZ67jU^nA!buPtrao$`WH91srn16V#q~pVCjsAz5+|krsR(T>*hv z9Y1s9lB9GD6Gf<6Cmt7v`;lBej@3Lzj^Q7`lsxuHA33IX*#{r7LRgc%CywC$UDYbI zxE4uXy>e}aWcdcUhWjI4BZoDS7!D-sxR>(lOD&M2=`$T8ra#Y{(j0jA)75(Smp(Q^ zXo4T=mc2N+I6Naxr}Kq=K_#N}S~R#YVJk(5@Y=j^b7<&HA!!+yAsOUGP$uwfpyLO1X) zAiUc`U{58oHGv*uVgdrr`0NTsQJ+t%)7nBj_qS&=JN!2Q;ZiC(x~l?m=DK^>I{VF( zMr4wvQBd%5@Ioe+S@!QwyiY6dejT(0;x+3!B*|1R56DyLF<)oE-Nnx&XPSzJS$xnB z*z{YqlQBBr3D^DOB%Q%WxF4pm^6H<2;T&hsEUpzw;v!rZ#Rp`dl$VP5+B;ZhxX`x5 zw_37N9hc!yKMMG06Kj#$8@jX!TAKr~Q~}*AWm)J0rWE6_9l zoW~V(HzRm(HbRrM-Fv%mAEMIUMH~g7sDw+?R=&|-C~tMof>MxHi7C3b`~my#r!s7A z#O@`|>r6#nb`Zq1xF1`^X4qho+PuBme<1Mvz;xtn{JzR@i-Jq*Gz5c)y_HImQnS8* zrRf7frj`U7+`MmQsSfS?;**6}|2m!LSl2Gsx&tlI0dJqCk>C+qjqylmu<(!1b9mn3 zeP(*MXcgJ-D;|Pi(=hk=}mr#2iFEd0|{-pYF_aL=i%pMG18DHEhkIv^uSQ$8Ta z55M|vS>d*;ptcka;zxoM;yPGKaqrzM=&8L?q@*(r`4bOvDM0TH&unm?$9l?U8l*BD zbU0&r8`@`03UUfgSjrcgie$X}x`UX`Wt-y*0%h;}P&->OQccj?9Si zs8W`~gmjI&&)S&smN=-DDwzWY{VpO6r+I=JnN{Y0xoZvrK%|D|E$s6wxG|6!ZA$hM z_n3BbB3A6E7|7D^ks5F$(E_^=&6s$!G`OH*bcE;>1cJQ!FIdi2a(Ac6bN8^|iJU-C z!#$8>sfhy|8btii5||U>C0?%BEx}PUXq%t`pLZvZVaV7Z>uCbRPGW{zjHX&3V1Am? zQeiY5xBrmH^4%4!^rOd$FkZi&(D5+YJ;8$BZ46JOYO1{8H1>?O<7qXexc2!5%{H^| zaev&H8OK}3-`hJZ!D2nWH-CLrgY%Dn|NiUSwtdc=5bt7wNMh7vRzg0}emJIcgLy6_taS7b#5d>sjCj+IO)yjYU z?+_j5$BMCOG=i#JbKZRp^jf9E-e?&RV%U@lb*&jn-jBC9emxsqtW_V|fmfrA7db1- z)ch5twK?n4=8yD3%rw{pit&yyu3+kCeN!1r<3SLrcmfg!pDr;KRf!E7dnfJR_gj;^ zku2yEcDbgqWF*WdGYcl6DcHBvQWY_ET!20e5Ga-pd+t+>5(n4Au(^jiLmpL03dmf1 zV1!ySk|3U*3uM7F2!!WD&_Y5(MZ*2u5C=dTIplOydX4`UeeA~=q;1m#KRNoJ3?B;a zJ7r{|SIWO=ul(A1ez|(Nbke!9GBuL?Hu0^n;?AbUaj;{#Em~IekQtZFuzCNFCd-5u z!iV+U2!xrf)B*hF4e740m+vF#b|nj+1DFE-=q7ZY>F6OG@dX9oKc$VgAgF7b3NGtF z@FVy!Ibeap-<&{X%5ap4aZ1Y4@Cl-*Gtse8v0T7`1GF`2#34}2b&1sk9zI}%HZ^^K z0ZZBpFw}YGHYgrwIh5S}>c3`;8}dRr7#-;5)mLR_un;J3z%|If#W!Ki_qACKT(L8# zblck!nvOpX^6jkgZ=;&(l^x*-DusqI9cB+&d)W%&H{9f6Xbzb>g zUM;-%Hn_yTzt=)LySZCImKH6!DRUUyc;VTKlq6 z&H%O|WrXS&KA(6w@%@~1+uxG)^n3eP5e!#C844tdXX4mnV2HM$440bhj4O?RPKLvO ztHdIWj6;M9{i@#P`_1Lg|SV&PVPc1O5aPaHHE$M+}oI7xqlH#*iP!m~I zemA+?YK$<5ts|Y3ZoA>gC(72YVrC2J^A8H_I>5D~o3{G=&U?ah92s#K)axH49g>o@ z*A6Zv-zp}0QzZA__{)Vb)xAE#CFD7G?Ou_V@tOUM`)>_&NFxkYHi@KkVzjayS9=-^ z!tl3d>kC5>2fyhrqD+;Qm1gk_<)YR~hw!{J&yuF*H)6pE-O3W@UWlEPDt$L1Mp2!t z*jiTkzL!zeBqz&a3KX0>{as^EPWeXGMNCbKaO@p0L+ky@QQ}z+cT^9s_x9d#pmGXE z08t3Flqe1BV+K^thgxFk-mWXNLzS`j7(+gYSH+SVTD^fwQ~9Y4sWIrPje0$QnkC>K z?ct8wBv8U2sK_0tq=w`!`cIsh_K*oYYz};CWd{&{FPcxt&+DhQF4Kcy*FJemtayEc zY-YLCA8%>FKndu#g*=9h1j6Sjf2F^hTrs#pZinrfGe@`?_jbU6zl~7oMcqM8gZBN8 z;$aI4UQKxd#e*_1Q0t}{Uj3$`Omu-ri<}?!qM#q}03VXiWprXt zs02nV9cYaCyuf~~=qiF}-5cEc%Qs>0#8SJ8$2yUc31#td2sRkDpb>`rLo|LYBK!1F z|KYlAbxldpnvQntXCY$Hb!lkZVd%S^aZevu<iW1NWz$B(f3j!}1l-XcO@d7+Equcdm;tv(2W{{vccnN0jW9EaoXwJ6dZh z!B^j9>yprIW@&5dZm!1TPmK-AR`)O<%E`E*>XDPmt=6r5kGOc`Oh-Rw7fZBdk8y#2 z|H4d+1&?AOf-c`eJr76foef#EZS-IrR-lhZQ*XNeYG$*ZuS@>)$8gj3HU12=jK|KO z)yx4-rmHS*T?3C@e0!0*I#qBkDjORbd%mj>QI?^^zf%zTZcD$?7SW$pHaDp2RH>N) zPf(eDa5-GRjKY5?qys9r(SIWT>eO;4srid64sm>IapqF;xTsLHv$|pZ`+T$idRt`^ zqmp-*QSwt}$62SLm|gbQgZXuq62!~DnYA%;sP)kf@qYtbw`?tRqDR9>IiO{gJ;ptU zTD-1_%=mC+SVyF-K-;wDpQ?aK=P-m?%WUYYUrrF7xXCPM06t(j<8$@M!gGEJH?Db= zY6x%bOHS7I1$EZy-VeO+;#M4`5;JwQFNxM2?;R-l9J-#A*u*n!&Mu`ifd>;?MNM-) zMlwQlSsR_qzDF;De+KSJPsz2H3n2}7yGui9%Z{BMRYk|%7t>7B2fjDi)3@2PU&c3A zhbj=ou%k*fL`^>ERQ0)8# z1cZdm5xZz}XTSP!n5h&y%yI{SFbGE*KoLjL%aSLLnEmU8C+Rc}o@qp~SDbua7!ZHZ zB`${Sr*q0=dIP@i*9Vxv0CrT^PpKpt!i->IMTURbr*~!&Y_W?Y-?j5=d8EYNC1Fz( z77f3xsA|o^=jkgbXpZb%z0&%+d|L3cuW)Ir*TAXU$tZG8`?B8Y<<|$DKZYm4CMR#7 zpI!_vE=FeD2YRw%++Loef91d2vOqI!nBSo5CaMuFVEqtDL$Ao-zCaxR@A(st$?{_Z z=0*w}E(;vdz@$Dl(1Ftle&nnDA{8JMf8WnwfF9UhHe>>sBO95tt|Nhg`Mam#zZo2g zf97bAz@SN4W2&j(IBmp%lg`BL{;E*#3y;rNKRwc7I!p7k^LOj zKMKi|rklB~&pW6Kx@DL5vE6MJ0x`xE8tuQvpNKhbD5Yw1T$6v13RpT$tRG$oke&f3 zcnqqSO6ZhbNz`^*?kNnaPD@(=#1)jDR%f7i4s~ffxX9q{xf${3BFz0H3v%sN))`GT zRQI%S^v$Kc04x-*V4{rcb)Dd;>c_T#R&)A1YL`yMq@nr0K7;b}6gU}Ms2iQzFN+_2 zzJdGid_Rr1e2rfmO9VZdktsu=QZRsINc;Xys+RW4>;$9AUO=BOYABC8)IQSPQJw2} z0Tr!T2wn5P{+EazZm0y=&<}1&rBz+gm%&^!u&GOfGZ=foC(VeL7o|?)u|^N*yx$`* zfXchwPhYbWi=k(9ouPMQl=~jB<23)GbLHmZZ|BQsYcX1|P{O)I<-+}>E}4z{P?N3nd)H^?Nf@lf1m{2&1o=rme^rOX{^Ty)8`?jKX5ru~*8 zrLH+>7b}bGMYEoVN|uwEeF{hAn((0H;3vgCd3PK^gz9aGsWa>S_UioQCH6VU_q6%t z@1k@=Pyz)I%sz|`4Ko|aWfU!!_13{y{qOo^LLI8i%=H6^4~K+fgUd#;P!zdI9t=R{ zjBh*CFw%4S>2&rZwv-cSCMPsLc{=JarZLJN#8glmT;Q?eAhjt1O%=h?qH!m5tBN3O z;=PNHddT@I6`Ry@ajSOxsyhn3hV?2xK7V?Qhi9JG=gL18wd)dX`%1Xqd#P?&&% zq5_AStN30%a$KEq_hk#ch&?j83A8}1eqh$vo51V4 zp-KoPt;ohNME4MCMLAjg(4Vu5?#_zR&agSjZMv1vM1VJmrCJU4+r%XapQPk6BsR{67XGTb~aIt{YKMLcK1_mRRUGFQm@X;>)iMTJkqmQ-4r z*tAz4+JXaQr$;7+1|XLL-X#9}8;ikxlbTeZ&RcrPwwf~NE-oa1)@@FPB$;aY+sZDQ zjKF4K36!qmD8zOSXj#*iL|kE<)tK=os1Y*!FpkB3r)@fSRUF1btRU;iikKA_bB2_w zG!u;e>>+wJy}B`3Hw7EHHqz8j*IJx;^-I&?9CvX`6jy#6qb?oo;T$*~t6lKM+j!*G z5$x30t;V;Px5G$1l~gZDqNw=h0y1EQ07+#FslZAy?th)7mQj9YGA4Fc5NMDfJ>-`E zEBC8;@{Kqcr+U~J6|}+6lFn-az(fS>D*_*@kmRC0Tm|Y(+YJnLn}u7%VdWD?j_w8s z=4IygsmClQ&lj0y9IbggAD7$|id_4p2rk4%{qd0r>4^@p;BiM#K1LwEUCcV|ia5Is zavv!b)qC}u(s^Dsv04=Ub;>jtu_FS7gWts{X;xw_Vvpj~O1J5_m-!XiMT~g}sa0@~ zbnDUPU{1Fa(^T3vZS7s!R6XdYiM}wq*lTN^c%YS)a53U8K`PFvP^znG9s5~JGdcB# z*}4YbY1nm~47Y){yJ(G)8#mjl4+16hE*zjnid>t7z_7p?O6mG>b@hxt-zI9DJeVXO zo;yDBkzR7B+bMWzb!5m_d774Rei@P!#g9%Ej8dYfyGbe=iY&5z_50GJ-d0U;txsxL z;lao9QJxtsT9+KtV$MfISkJe%M`}EH!6^Lm)oVke^4+BC3UvM2Y}AEhLbh1ooY~8O z13BMnyuADqX7+HrW01k?PD!?lX`)NMaZ|S6nR2s4f>`|pzObuKq3_|Sx9774O@wO) z$6^@LuXB#3xS-KRx8@2izgs3JEXtgbZu8)wM5v*{`=&_-l8`?}{#SbP;MFp6bLUB! zOlF~8^NwEh!y9fv(E`(4sN-A}#T8*R}b&gV>W#v&i<)@@|X<`FTP$V!*DX(Lvk)cL3 z*;F1;uvVnMv52!`_cDfsaC}^}h6}}BKB~>z#2@`lB%`$l34`>L)IT!9D*0!8v8rp9 zH69}hCJq&O@aEBDb92v~4btmYlZ$1S`{&Ae>nzlaMc!Y`YR=%I^)MdFBj9qy5B~8& z5O0S=e)ZSWqs~^w1ektJ{6)!;YL6#_7L_i`f%)m*sK5-uj;z2wHuIaeXq(JR*!9k` zGbqJU%fuUpd1x3GtXlC>-~49%mqBzAFp?7-Ps+aolB+Y;?M4x^ihhf~lC?25C)nLX zc$|@NLv(Q2KQ?L5jhZeJlw-ik+mc|8!U?L(8v%@D-p!7&$$e2q&jEQPNSPPPriKnU zVD_Rop)&{&F+c9swxu?7xYT`GKR4IaVxZou14q~|52gO2=M~+ne~@}>nOZLEm8Q+r^Pd=8&sx$d1SDg~ z3eGf`-~?pk7|g~H;B1BRZNIG``GkZFM3C}R>dx(|W6&ma(UY`%gTZ5_tNBz?Kdd^~ zH>sRR;ro{APrB-7j#C)eJe2okl)n16XnXf#r2mb-9EWafv~i9n>JaV2hm>lEKnbO% z=J)Y&JV}_P$4nfcn}Smh77tMlWHBA>Csc0b=kK5w`q2hnxf}6q$QL!!NBNbtoY(F0 zk-4XVmU+CH-Y(p&4X&Nf4;S0iJo^b(pF~|hak2+oCO0F?;{8l9Uy%9!+a@w3hk3O# z!^UtHf3YD$9k$>W}jx+U*!#3ajQ-A_+*x!WH(!tuF0oh8eFDf|W ztdr!w{Otp)<&+n~CHSw2dVvilT=#YaMfhN;IzQGXC6o$h!>{FQMrrz`uBHzHfD)@2 z<`5nNiAVPAeI&<3RYHi&p49Iic$%3ViR-faTv$A|g|vGeqP}H4%vDtUXSoCp3Osq}A17hTnz2mf}*Mdg&4_GgWaukO-UgV^`D#D;PGu&>8^&660 zv~jp^LBzHLAsmEK2`_In@lw$F8f^DzqFM1!HGOp7!;I8_R5}jSD{}RgGi}};0*AD@ z2sXMN0LT))T;Y!3`%Iy*qmBY3hE3A~-!{*HAd3RZgw4+#gM0QqScMd(pD!1#yC zmixt;r57=Zp>g^cs;qcW3UsT-!5AdJM6|KpAW$8Fc+`Rn{S*SUf*6OA180XaJCtki zrgZQ&uXRvB&x*_M*{{V`gHz1Nd;CH(bzqr^d}QYlpIlBb1PPb3k-zV($CTE0<>Pss z&VTjKW6U^bO>|uA>Q=fd**4W8qDGX*Z;$#LcJ|syKlh+J`qL`xF)oy9hG_~0Hd=)=;KKvfDqPalWv49D@E7nhyh;pEX(hb%CqIC8Yi@dY zZ~*-2aVs==*hAhaE#*v@$g&DL{4$h=2#QoD6^ojuq;!jF@gbA%s%45-^l^Y#@JY36 z1;oIaZ_TLvs!OWb_>Do+;=MguWMAHy2UV^kGMDuab@9}Ef4zPu=pKhY(pfE*4VUcm z@ge$1O{f`8&#Jeq=rIJR6gfhmYyv(bzrQk;RE)ZHmYBMpW4ok+55P^D7tz8>B8x^KTwgHYb5e3n2MpkI z@?okrs7ExB1<3f*s0vR=2%C-}O$oQumhzu*OGq<}=ouFr=_;|B2`I^!=@t3K;(=lW z>9787KF#BVl6MsBJ%%R>qetjsssz-(E8F7qV~>|YI(7E9@Yh>9s?TWfV>^Pxw^M4^~QVIqc`a6{hwtOKvCLX zy+;VRQE<{b2={*~c%Qib>`^&0+DIRRI(Zk^U2G{;Ra5?#|4ihoTk%3T*58&Z=U3Hb zOFo>V0-{Y3Vzl^^E_IGl(xrzN3E&V9$j5(QIE(WUj+z7{{e|9j!kNlTWZ zW6$lV$5g5+Ou4=ua}?Zj*REto9HBdRWQGO(7`1vaG*D~+-)Hg(qAWIuUq+wJJ< zpeQF5v%6%^LcDVR;=6HiiD=Cp6c*(AJc}5Adag&d8iZ2CC5(C3$X6RQWidNzFQ!fZZJ~GU5%Bud8zOzVPp$_xM`%4_3Nz%E|B``&m?X28LQiCgg z{pz=+eymHu*j9FLe0ll(sqmTi#FItlROjb$QsU!_SlySeo9{mQG-bayEl7;VgkEb4 zyQK*C*BC#I zmN%s_7chEDkhrLQ=`7s`{k3w9t3=d5@#9v zRM=|s1Kswy;qie&Q6yQsE;Gg{SvKQA$Jj5xI?mQkC_9Mh)gMcrI_rmUyylLzCzHat zSZUy&v~x*^kqsFwg3AqEZrgL2z>1tX@!;H@k&iRKWoXc$yHwBlPHJr(I-Y^_TMn|N zZ&5m!hzx@g6XzoI64IlQkA>LMV&;HB%oy>6jrSe~G&bMTU=C>>R1Z`%+Y;D0@2!2? zZRW1M2rlv{Rs79_P!sBRl6@LSN^eoHwd994D~3ud#48uSNl!?B%*d8e=&E#mIeyvb z+IFzXv=c5J(?sh=S4(@M_^gKe!c(!+y-f~ezn*plNFds{ujCcAsG)(;6Jhh5VI5$Gtd`=-{wJKj1@&ilf zHU3>RVH<%6$0e0DhSNy8p~ ziG7$IL~4{?hu}17ctxupDL?sys&y789r=eV$sVuruyVeqk5yyx4@M#)>y4|~P|W!H z2(+&XoGTSf1}tva7-nyZB$9W(zWaHdEKr$)_GAH2=-QHtL(Z{gwm=8wOON5;kZBl( z-0`)Fg$?72acRA2kwFF{7AGoq4=H}uzKO+TLE*E)ZJ472;ihgmqLC%MK|T0f+#HIZ>!iJgCh}6K$#yf69h?+M+URwBqC7G$m7qnN-6IX z&QKu&S@*E3ggY##0pjql{!11`I%9<6Z4*>63}*IHD=axcG*M3J5NB?RMj09LIZAO@ zmoTxyE1TH0UR+KZf!mKEPd`$f zR(6JF0@v&qm1rtN=@!&!XrPkxD|CXI24TdGVOm6Bixokekx$OhKJj9$>P6TvF47<5UZ6r4D1(`Lka{CdI?f{ zpfy&!O~~YwiUo}V5Rhg`Rm5;X@1)>(m?Z8#dqQr}qiDTg+b8u$6)%;`wZk9Z=jN8g zR_SUkNC~aKF5-0e=_lzYU3wMAsPyu}Zx>F7?SIXKH_k6F7ynN+_x{a5ZqAOJ_Rs^<-_dsNPR4&;A^-|-)i&nL03)OFzKq&} z%3jZ^*1VyYi8f<{qmJ2b6szlH{$KLR*03f13t`vP5h4FVQ7ymf3;)iq=is#87@>LG zi4@XtuAe2KY4~UHe11Cy+`3-9a(LJd)VnV2oPBHU<|KuvlFY)FZ6ZfL&Bn%A>h&Tz z(~DbMlkZ)ujG>Q0`}OXHxSg0bzA#b$olnoWAJ}pydBF~2g^r(IR9QZIE`Dh3)w7NN zp%LERf=G&;wJ5Ke!$83qQ#sGQ)6?86ho%V7w>GKIGmVQJAT>+MI z$m$rHLrl26n+V};1g0PSh%ZM*h7UNIH>?e68&dY*A;m-9j!Zplf{uub0m^Y3S^8jUbTjXiDHY&Rv~SnX*GrxU3nh38US0nZ(SmcAn2=4U*9YZ| z?~f;*r?8pWwZ>>!`dT z=gSO_ZWgLBw&Cr4t@xfkdG6y$5uqSCXgWi;f83rZCIf5jC6g+P-6b$&ZO7Ys_nPOa zk$f!|oj(`qHyPV{5=MfA2a+fF6>4>4T4hXHQAG{NOwj4NdQc~@EF7!k_-3hbQAX=H zu<9Jun>6>Ln-}8uxbo?Trkqt>?K7u6HEa=b=L#yd;y3v6K~1Am+|yW3by#7y4CSQj zM=fWWwMlB}M8qaY_KL&Z(%ggggWmQ;(m;2v>BPH(7VAxqQp<`9%jhHRBPH@E^7)rr zJ@2teNzvuM>mT-t?78?Cqj|jnq4*a!m6JK{oRZGYs?{?yG0|%wJn`K1Z|RdQ#sp^b zl#_9~s;tnkMU!ZNua-}IE{PotI2qf~@uX`X?~DMO_fzAdk{B`iZ$BF|c&>P~^CB-K z)L2zKLiH9GHb0-u#6M)_A%Ek15wZ5NQlH~mXup5doYzz*%M}C zZX<<-7&kz~{^Vz0irB>S`ChR-?4QSSB*nwBN4M8e57Ng28X6o`_kQcyT<|D{`yN~N z=89=h(IqMcF0KtiF*BcLKoA88lpnF{U$cN%gbQtwBqCS5>Re8zx#C#`-=eMenz~8S zzscr3(+7F>SRFyEx{pqG?YZ=8e}DA(JlM{HXCCvi&QAdg(5;ZYdls&=7MyI!U<~l6 zO5DEQXoDk}Rdwr6H=?r+wHjbLu0+I)Df3^s>Q0kml-V#&k7-N|6w<#mcq zv*8BSfMV{~6Vt`CwzoU$ZQyea9_2of;lLA(5>75HwUBb1-e{{RB>4Zu&O&D0^Yi2m zGoYqIY9VQ8VF6I8z0pz|B>LSB`u>VKm!qaH)wr%0#?*(*B@I#ex%9wLX17C+>hiVD zoUhUXdNW`nfC=i^-rA)dKDMl1y1D#@qR{oj0Ly3Vkb~kWKsm`F zNm<)g$4+5wTU!CJ)LuFd-e95u6x`GJK!J~9(0N1BihjrVDaWNy!ezD64RZW}#PhFywaGdtBBydL;C0pP}piVgPV_*R2-l6$1SAW*CH+3mBbd!GFM24O&QrCPz?AT<=t29D3>(*JL+83#B z8v3^1ADjajp!#P)Ps`8PuT_lq>(f?y_MhCDG(Nl)8!?%pgNz%)z!_bdQX+=G$zOB! zbtMB;o?y0m7c4_Jea)?w6Cx4nK7lDGO%Cf6WbXPaEF&!@2RS%vC7A;|Czn4Ps~TLN z%vurutci_}IdurrmUF%)$i)yJjKqKZp@YxGy?o=51}qC1p2D4$%o&J~`Vd}JXW8!6 zotP(I;)C*AGyF_R-*4{)WIPru6jBnC5gW=;LzhmCD;DphIEnEj`X+B3JL#^D!(BMBbW z%e2v{V5-uS=La^ssfuG`Mb+IfQ(^b literal 0 HcmV?d00001 diff --git a/frontend/public/benchmarks/tts/pocket-tts.mp3 b/frontend/public/benchmarks/tts/pocket-tts.mp3 new file mode 100644 index 0000000000000000000000000000000000000000..98369e1624a9ba8da1823c902821f5997dc8adcc GIT binary patch literal 16867 zcmd7ZWl$Vl_$c_n9R_!TyUgGgg1fuBy9Wp|xD(uhTX1)Ghu{PP!95{pAlNLt_x|f{ z?T4-XxKGtgpPuTj{`Hxb^PD~{C&>j5cnuh0OEY|_X=~+S=VxU}Y3@sD=1eIjN9k$hWo2PY>EvQ*<;YGc z=Slgxszc2Nn>kpyvja3WG^AcPi}|`)MKf<} zUNAc!I~W9dGyU&?UWfJnPZ=QRZ0+(ogZ28+1^~oF0SL%wnAmuPge2sYH1teNtQ?%& ze1d}SMJ1%=6cts}wR8=Pjm<2r?HpZP+`W801qO$OMaIM@rlhB5eaR~<`TDi8roO4Q zqoccTaAbUHYG!V6Wo=_)Yxm&z?EL)l=I-Gc`nK^`o5iaQEbun-KS2xE|G$c2qJ$jZ z=l|#D|Ht&zKMbH;%N9j6i-f?7pg@S;;Y?3YFXzCTBEk>5p_7KX`)2KZ3CcBoW|CJr z@9_<$l*NG$F$<>5q>DD!%la>ZUt@PD6b_E-H9(D)6N!{!$Wd{3AA_Kqn;QTa7=T3e zJ1oC}3;%aKjL3-mR8U4_SOCNf`cG687dLeKUIf`E=bbN<(IfN}dzS@bj_fKMHqHU@ z21LP(OKVcgV#8Gv>+Z|xhgsg0;e;j8iXj#(A&06;!ic^3IcUYw`C&hON3Q2Z98^o* zqoFxnZFE>zc3|z|;$kBwCMJGIPK=|ds3=!PPfyRFkijm^J8Sn(YK{M2oTt zDLFgh(Bn4|-&?VDpl#IlxhF*Elpk;ev}5J6Y}|4<*PM7%<+m+Na^I}>=wbC-{X6nA z_Zh2eT7-W3XXVj2Lc*i-=2s*85=g{55P^MFx<8X=!e*`X?=zZ@=Z{VCKZy+p59NQ# z_W!v>W)IfYX^~d;5|yYk(LF9DtgcT!?Z1=BsV()p=Fh#2`N5O^!(+~>RB+E;bOULCBx1MWCOUL)g za&i9sBrsT(b4E1%?MMj9TpIYXk1s&yQ7ndX!m}$XKVgcj!{9b4b40lP-0rZumof4ftLIchq5W76|khC zvb1}Xk_MUooovM1*&2LQlf%OH;j9--hBqezJSJ2LT6^>E1N6UG0@dwATfX@2 ze0zdE0fC*t2-CLnrl!=6#=6j2eN9@c3@g zKT4WwO*H6n+uOd1}n)Uq%}XN6qLUi4=w})IK?kaDZ`6U}!GI(hirP+khIH-WOo9 zN5#8d*Q$v&5pb@bM$mH*6#Dc>dzQz?D<6ua{sVg5VZy&>0zH9lw2ypxg2t_SZ9&ho z`Zn;jwj7+2L}9>ZM;4Tf4isg673;doy&a&HpBm9`eo_XFt6pVusY%o(@|=KxRSuOT zm2c~`CwFu$ajcmPHBKe<``Vc-YyxO;?twy*A$M5#VVTDkp6~Q;KH7d!@lv0r{VDBu z*qoUpb+H?_1q7zcX<>XRo6v{Bk7}f)k^gc;UCJE|i!uS-uz&H={tbm*M|-qEpC@?q zl+(RFVmc%(lg{lb-5^N7FSp8BL_sK!bsx>e5%CF?+}+Kk)gsoD_;W8p5P_&a;ECzv zi?TESw8ajfucw15+}`{kwi3ZcX@!wm-7{1?v%d>vWv)O)jvY*=3{v{+a* z^mkQhqO+4F6PP|dQREx?z*A-IIsr_>Nl^u@&Wp60G-rfk~e`uZxHEZ z;jNUL*F+KUTD~xw5K&8Jy=3Rv|IHF7&m3gQNF4xbx>AE=fyzji=1ifTy9#f9R`6+F z9{t((pY(J{G|?`(#(Owydq3xPTc&)iy7kps(b`K6R3p*9PN+RxW)V;w?;gj1R>yFu z7SaUs5R-iQbyNy!SJ~?Kgxm==2$%we!)P1Gn8#=Am%;pgU@+J#tt45FHdT3%i0Zwr za>%M(dRh@MH9>N}AnKD5ncN_@L!JT5C7t?$a*KA}d(A8s7MsQ~X_={NJJHGmd1xKf z9rZJw{1+wGIQ);*kM(;n{Q|^*#Sdc5JRca~-u#5DkL$hC*Jy-zmw2EPlu*Dn92EjC zY9cb|5E&P)pB#h=_*f5w72SR>wp{trfPljh5BD4>1Xrz}cndSQ(A0Z1+dK_5xJr<= zIHJ}AAV-gm+wAHM2;9q`ezsexTXEuh*i(O|sH@M7vJ>n7$lOU-(^palh5pt2EF4virZ-;FE>v!9=IIdt}?cD-~vEj8_iT_UA*EEu|T%^hd&d)`sz zN3j8o)+Kt3Vq-i3FD#i%aY_!2U25Q~Ij<6n9><)}p-PW?h7w2fAleEzI7gDmgby^( z=kcqA4f%1a&kV^XOADr}i~w$H=SOXoQZ#N%$cmpK`|NPj85ybikv959d&^AeA`4h{^^Z3+Z5>0Q^|O?>h&bpIhB+Y_FP1w1z(0#M+TfLW zdAo0J_BPs{xP)NKZ@Gu8Mj%!q`{xYs7-BXLQBf#quu`IhGj;P>#lZ+Wbe$+<{L5#y zgHhe${%X57zdQrZXzPpV&wF$BjZ0Wo=Q{QJN=L`Sa~lN`qgp0gp*e+8oC;1Ri3LSN zwerI6;gy92?s>v1ItKn$Eo~)*g>G0)!6PGfNgW+djg_mg9(_3=AG0zx#Q401r}l!DJy@#$$z>hmr@Z-1Y1V58wSBWl z#5G~oeDRvRF=(6^u_Wb5U zL*h|L!Py6z2^>eTh#1#PucZFs&*J}Y=M@u8SF4(BesbUpdXcJ~-$bVG}Y$O@J5 zh?eSux;$UM_`n#ywnNw(pgAs8O`ErVi}H>(kc|ERWW6as_*A@{?4)xUZDpU|^pEh}7wipmH?cvw0$^!4R; z4e!*E$(UDigPek(x5eqD7EYaQS^u>&?$#gA0#l#^MmQI#jfy<96@Lph3>UI=KYpiP zfM;@k^Jmajz4Wqwj}Iy-jathH2O#>oFl_#d==YiDKI8w8K#XZti1!y{9LZMZ@Tn7NeCcpqA_2+?9T$AS1np2@ zDQozYL_kY19xFdPpXrmYDfWm{#7Kiex|%X-*ftYhoQ~{oO}cz6614V0%UP|kD@|L6 zydVHnOoP%kx*$MuERk(arS)BvDJxG)l5Tb3XiTNB-glj*gu&(vDVGC9iKcTA}CrV~6r!}2hEDr}Y`*_#X?`MRti$IcRai#BU z5u_%}2HNcU4R$M>r=9%2#+IgAJU0J)4>u?*`C@?!-U}a}0S)2Bj0Db#&;MHQgVxL= zN9JoX2^gM5bFC;h*$4B{ww=zmmGocu5R)Bd%mc_!efT@c!~t5`sy)+q^n3-G-5`#U zaQIQ~577v^crCf~;6z2you;NY|0)9+YH7kcnmneeye_BV_cQ6iVu#V8RptTvv-ey) zKdj0&dVBQs{I#0#ITBKST*SsKS65SLRtB2E9;p#c|NhFk)F!s*nnL#3Ir|_7T1rR$ zg;o9_a+bUg?axi$qin1 zsC`I6eGzT3$z%9XH5+>XImE(FI>>fI5~H%RLnRbj0hbO1#E5=eRR|z3wFNoxH*?c) zX}$S>vk6m2mF*$%ih-zM6cLdYf{)TU%GV$O7VF^q5(RPV`0!5~2k+jwZ0iKt-DYMl z@sAr_>(c;_t`3C+;l)j;Gno1I=IpAxGU!IJ2!4U!)J{diK%By>qXp&QmLCaCvfmo_ zj2_NN+RWnvtXY{RRJS*Uol=R0JJTLC{+yl|%?--3SnK4XOQn@n3QTg5AadYyhiJ1v zD}tQ`+ZRP$)X;7W*p2SI6r}+4=G5=}Icx5RW#BR=8HS19{AZk8=C1H(BZ7g2V+rNs z9?EtsCZ{&2z+%p4AquhLgIV4WL%0?RYRo+IefNJJ<*S;wMxEnhqLZ1Wj$rVFzZ$Yw z&kYu&OKOSZUtic7aJ)C6|2vfbNPMXdeL;x}_=*Wo=Eo+Zi&X zV>pEGKczse8aqIPq^Pu(u$1%Y$U5P2{={%6B*T%6MaGh!TM&(2>)MO-vuKRS>$!O$ zh5J;+PN9aEI_m^{1(TV5jpH6o|;%u+J7L1bv%v;kCMs2-y#?w z6(x?yNK0e~|KSjRBHYwIf`ncoW=1fXPMVKti^5Jj6H$Urv`91+U{hOsAX$)0mEfoJ)O_~ulcS1|Ej?;n<5_GPzh6GO3Gm% zi<2=AttlRBQ;(2V9m-qA5ScE7AF4LwiIUa$m+9&T#{tsNZmF03>9VMcmSW7DIjKz(XfBDhL4%eg`YK<@m@0s#Fm8O7p8>9tfl5`RPJIT;(RtnSqOR{)0nu%Vv?H zaLrK=(Rub+_G#s{XIk9Clu-$Xy0s09(j9h%({T;&Yd%DhJ#y%}ID~t?0eI;KEgn#X zkjYrRIN-%RAgUr=kr(^C`K^RJsngP}RlaE&CE($Fw%T!8mK`Ch=c2LasF$b)lBkM} z6O{q%;3>cCW~Ib(GwLO}nTC26N}%sf7l~4aB#7Zy5dk6;8yO9?3+^#9jjWzFGCBv- zLNM7WEKZljc3udwQPDqAWcx*#l(bT)CG(8cjkkvb?Zc%=nJrVh_ul-> zG@8^a#_|Xv;!*D+6%V8pH#fV-7Zg9dw1;;?&oa|r>`#OGPN+ML4wID+l1E~& z(QtQxX%ZO(x$;yJL+GHUy#pqW?C+!zs*07GRJc&+ZzE-$v{;iEoxXe=`^9jwc?YsoWub)nLm!TuXeLzfk_Zyt{3#IjA{)^a0^v)Ne_VL6B5PBw17yI@`eq!bxhyZrW2iV@b*R8qOF-EOdw0T#OwBy@eF&CDf07 zyzf?fFHL@Evz-|nK0$kx6z}Z0YyBx_E|x|QWopt<6`en_F` z>>SiKU5kKb$iI7$4~Gqk0Fc*U8{nJ3AA^nrr{K|7w|xe%52)CA@T!MXE%U0{4`7a} zW}mm$@@s!+=cT1|8ZXUs4;~kL?Dp#X@wINB6>cS{6XMN1Uu%1I{}wP;6IbY8(~X;6 zqrJP5FQ6}nSSsN`LqjQxgqsgmtY1r_hy4c%BumXj(XsEG{9%5S?dRxz$MdMwpkE~} zI(HXTv$k>a=jz*&Dv4I=sk$*<)9>N_#9>hhl9;#puVD~xtz@6bwAUjA{LN9fC7MNq z6AinTr2>$c0!5c~i@}{sv*ehnM#ke70#Er^k%C5vZPo>1%s& zsBetg6CE%!g}1S==%L#{q0f~!E*%WzkWNuEw)uG04cO*sAT1Xrr7YR);X@bijO#5f=8dg($jq9$DD(s=%Yg zo;C*_!72k75&;N_AWJ-Cf^lOeb|}kbijQk;`iE^eSEJo02N-cpGt9VR`~LjqUt{pd zi#ps@C{ytHFfSmj^4xxQja^tAvND2dU79TCjKo*L@gtEI6%6RM%^shhiRy5o7T@Pj z9C3w5MlmCT6xssmQANol@Dqj#j{SB{)cmYiaft&&(Dp8wYP(Z%SSo$30)OBBmOeyx zy1l$ydvrEo9OIcfS$)t5su1$sT0N#W;-O4Mp0>LU2`NHs1G=%o$xVrw;YX1GwMi9Z z=qj)w_tG$t#Z1l3#{H49M1Xyf>fQtnv%_H3H~%?XcC91eoT%IV0s;UA4}!eLOAFu# z<#mc;uDsT0<#fL{exO2irBaQ%h8H_b-qE6trSb%|5GVp5iNUu0s05h6F*Yxp5CROS znmHX@EF2OhVKcv{Br!ZHYMn95=KIB5trTNY_T5VjT1lJe0o+nu1X)<@Jct+$1}wcU zi)vSVHEVb5@>cL_2=*Xp(Sh(9&kz>{PmD(Fs0T`sbwc>i*KA9Ok{s9^ro1Q_mF1dh z*FhZDA(p%v^yWVme>n7^Uu&!p2aXksGT~Z-q}60a47`skXRYzX~3sLA+tDw|7B)E0GAI%5hRD3E5rc$K&paQDCMC_b2I>CzIsW3__T_IWv4{43nhHX@%Lt33nQIU z8CtIN?(|Od;Ge#Vmw}=5g5VaW7&!(E5+Fy0w+!NHs0);3!M&&V^@ghhEd&%u`aqx0 z?GV1JIrr@w-X%H7c!y0 zSDB*h>>$WV1EglH`-wGEh_OssF&RNk(T20u3Zy!OuGHMC6iMmu(<`EroLH|} zA?Uxv_39%D7{OHNm|M_IXUYxctJ@ACYL}c-Jb*r7i=TBJTZ`GV5i#;FEh?1+eT1BB zQt1+d{~5>ZpWl)BKf1wrZOe+5;ynFjBh-GtgOT~eG7Je}eTrWF>*aA!tLGC}b(uTH z>E_QjzdCiY{Gha4=HHAKY2$h;Yh9;m>-2vs0kmLh-}FjhepeD8%x6~G1<~Vlx%{g` ziLEWW3OL@`5U-_6vBr4zY*0jz^Yk+unR6RR{j9aW$Hoiz?M&i!@B1Lg^&ErOk zlm9BWk~w#jKL0~xnjkgi`Vce)MM#F5MQ2-XBb>=8WvH~(iIlFlOfbJbpSlrS(EehyOm5-9qn3ZNQmciigWx=whYKJ-9quHsiI=h zkTPvW&89W`#Rnndf7jlP8SC%z^Aso2@~?I8cfZ;!NIibn)>&Rys_)|$S}vTs;iq=M5BRmyXVk)gk-Dzw>Q5vZSrW3_CTzPSKXGv z)^mrHuHZSZd1w>{c(d;*hC&yx1`g9YDK)KN|y#};%G?6eAHMz`#;5REM;WN<=n zqwg1YK!60}vXo~Zyq~mpUH4kapbCvgEXb{0MT-O@fcFDZ751*(3{Mn+m^MPpSsTmm zjQB1|4YJFokfowZbJ`Dz4JfiuV7M@X`TZ#(M-oEv9CF_W;NC7cpWf#|`=HQ0dl|9% zOE3Tr@F}7b}hb z&mH~pS8X}z7ARB-l8S66Jve3|gKiWGfRh^y9SDUDQCvwuKnN*L$wNhx=0^l~P|E(7 zzn8tUZ)`{(M18nOhJ#FV0%4-R4UI_sG^5JWl~YxQnbaos{RBLWL>P6_o9r9teUdh` z&Z_;ewK`hEYEs3sa~2tP7zUEr?7Qrx1*FW)NYaAlMQN7jw4|?WVAI!bU=U~_ccJ;3l z3nn+Wg_De_b;sm}fy1OSKPSsOpawlJI}gyFuEwgn28hd8)p@enqJq?%iS=%x!vnIR z$|o{gJ>IKyVb$NKY#innWwGQ8@qE>l0b^ zohwiU9T|Z41Y!;1!vG+LSx{!xqrk*ag2Q&0;yu*9iX{cZ@Wb1362&5;h7@AQ7j4(~ z8{=ZfUX#P_yMbKwI8Bw`{I_g^qo%Ou@Gk^F98F}zU^7Wq!k)$|&SLXi@=u)SwkrA+ z5lElP_VylZw*Cl3l$3FPRWalJr(aPw~xdsYh&P*uUnO%|CN7lI2Ixv&a-ppE%q+4`O_ZKlB2X0_|BM;nct;3seBW z_koD0Lhh>*I3{?GUmL6>&V8>$4i8MV%=F|YO}Aua3JzcW4#3O2pvfqJV4zb^OTPK9 z+0830;m_utV5T;V#ZBo^Y;28w{Jt>aXzljvgQdG&RCN(z2g7DRl)c>6Y{`+HT21~I zSxOW59^cQSB-KLV<(xS7Nk~6fFC~+!ulLh0~%PyJEjm<84KVP{vGMhA^4$Fgy_ef)I5jwSk25P*p}+C#T}F&1cS}xGqmT z#P3Vh=1%VL2ezIno3uf|zmX_;$Qe4hH(N@lA0d|Z=Km!W^V1Uc?2@F1hZ$zOa5i%) zX|^f7LPVLqeSeN(>gdb2ojr z|7ttgf6p_V7bjfB6y3!NwrAKVWEE2ZA~-^NsRb9;3E%*6 zW&3pm3g@O8QcG1UHOgY5l@)YN88d9C=Oosn4Z37{QtF|6Bq>t6~THwFbQW$Jxip!SJI;g zpK`6vL*&xJ^rS=qXl5h6F?zY8_VKK{cwBlVr?CyM7wVxj%WBN&w9`Hg$f+b%rr|Qd zM{D*y?-9Dl+2nIyE9dS}UG_=(=`yDt;~zClDIRCK4a37ORQGfhi&_cmu^N)Qw^mD? z*=@LR<{Iz$^cjsKdhg{n2?<0_b^mmwf8DC|r!fSE_1NH5KLG=jHvJk153UQdi~7Q( ztu?IkJaj7R4@pv_IHKe8_%<6~Xty@{a`Nd|I*EE<;Gx_O~MBY*4un+oA-G;G6 zp5=zfa0?3EstQbDUcGpBfW92!5jq+-^_boG?wFjbAMtz=!nd>gsBo?8yz!NEwJV`| zZOEWY!*{96fZQVOS3~W0!DUkr2-A!b2oqJHDj?R6fOb(>Mu?OqRYsD+b5o~0M2-+2 z(FY2BB7${6glwA*L@*-35(7esYhGv8 zg}|Vk?=Zx?wnNGoQvKVu5^FG}umZIo&3e;il1J_0I0%osT!k)0y9j?!1k>;Mciu98 zDZxZU!UX;Ga6y$;y&AM^VLG%ca&cfyG>DfmwM_OhtI&MDn?DOb=VSFN_efKd`dtM( z7HnI3!x`(lo)cpwmmg@-<$aLUkriiSdyv6?;DE6g@7Pvd|LuR}53fTbk{{c^O)gl_ zzcXQp+xM{e)R-rV5ip=%1EQMv@MZioh=#>)eq7d+&T_2v0=Qr?MRoiF;?U9kZLvb+ z5ZMG$cp$ozCV6TCwpc;4Djw16vvU}FpsOVeT7quVFdpsnl1f%C9~*Jjmm+&p2ld#p z@F9)EeZx?^2-hf_1MES;iJ=1t&S=YgkcjJi+=7_Y;9{0Q5d!tDhvZuPhnn|B=dZ&E z-a@G5WtLgc5^F4p5ej= zDeikSKRfwtOcPSn4<|m}59_{%%G}pug zyXP9lGR|;TJnJ|pW-P9SW?Rw~9;OS0biSXzH-7%yYMCX!U@C~rED7Tg!!|?rTt$4Q zUsX~w016eC2uHA^|DXbmuN6pe`O)j(asdAN8yg?&Wc<|qYp6hm7y(xjbk&w2XUxJ( zc-{9{Za_Y>|4|>r(t8BGU_o1Y>i#3d3_+Rc8T$#**Xu4&jYi{eKGdj0X&j5ulAKOD ziJ7VbbIf9dHXdXPJ8CVD-(+vtS-kndH2RYb$^%K#>N67K&(_&IR;H%uc6Q1_JdUcW zOVRKW5f*h1Q0*7`uI{X0GqrN+jW%e|v)juadgRZJHg<0A2e|9GEin>mZRxB2#><0+^*4s!WPx6 z-YYY`lGV^MIG#_t3+S_mUpA(HH5z)xQ`Y@QJ|$~B={<`UTO2SfMBKp+vvv)})t82Y zcG>pLuT8h};E8ysZIl|&EH9!&A({$-->s)25n}|#W2p$)?;hRsZ1R`?O=_qdnf0wP zbnUBxLN{n>S%S6N_h)3J^55v+;<)tWz1s*Q!{KTS8_HMfkgpXmf|&CP3qIXbQ;1}1-|hY zQM#t8K|Eq#wsfw5@virnm~kKT!X~;O$qgNF++ZU|X2~#7-~4X$3qK*s^LlPFkc$#L zWYhvlx;lB6_#(RQsVTflT5?>%eo7qKT`^M<-bMt1w4^#`fn|V;XW0eMX@U7C__*}$ z(HUFIvEUSP5Y;TDbz=H8P~*MEwLze%U%&u9`{IYzKk{aH&mp<|;4Tg`Q_LSJsDT_~ zMIRY%6(-xNv}X7FvgEB9&E29fG0R4ddiuJaHomSIyf8_?lravIUPm8&gZ6y>p~+HS zKS1VRNX@0Ns={4)xtcC++h6nMkD{;qndUvWSPn~`DkT{yHj3ngf&=WZQJKY*2Qa5+ zE6ra50%AAo3#*&z%k}IW9JFT5V4WMRx1fUs3-U(%2Fk>NkI|3c6kt^3T}EA%PPqz^0p!_~H^m<%+PNwdNf z_sw5G|L}7l-RI{|8cHHTR5tVZk=(S3^fb>;CoFbGA`?)PlFirro%b7cL)hGynWWsX zGHop@s~`cf1&xV|%98w^;+e5g;pn9RN69m5DCoNcBp;<;TKX6qP1zV5jYYjn7!JFm zOJf;bUdM+U7Zp11F1#FGkx!MHqd}*UKz=|ynxe|}b`|}pM-T$pKY1w;L`&3dgVJ8p zh1w@|v8;&#iW=Bakuk+izDErdSM=C?+y7)-w(#v?sNv1uM)xDRC4yz@H2kmL-BY=M zm?S%EaZc*km(0|D#Y|Y5CxFqQ!PBBnlHQK~=Ra_ovGk_3zor(?SJpTy=%h!q6aQL- z1yaxxFWD0ArREim>5kAI8PZWDS94%gYDAZ8N)CjqhqY71pEp)XF>Y1jbJce@jhgZ^ zLMJK~apGK{H!$SSmdp!E2FbDjq|Hux2ROeb6=Df+=;y+(i;hjYM~8%rs~@n2N43AC za)A*<3&#VyC1~n|buU{S-~0>o3k!|BvL^Q0cQak|RSfdNdy7#q88QmTzn3thez!2} zZ0G2pb?7cUT4;}&>&nwxWLE5o$D!T2S#YcG-MdDq1t}zi7N`K?u$E4+h$0HV8{w6j zIV%>ZhU=Ageg^baI>B(77LNOw8CfZyzF(N!yt!3^nNrlrf5xX6%-=NVvC)@@WUIo! zS1dYD)6OnYgOG<2I$>$_EYE5V-5I=Q9NIp8B;Cl#FkW?Bov)Y*@Sadng(Q2SMHamI zPgvc5mT8;i+mgtmNH)>M*OeML6ueJAgLyb4B*)GF=&&$Vf4ANe4GlgxlxzMQynsm# zvLZhFIwhGq2vUd+6Hk?O3$dKy^p!MC%)mRNI#$YBLIsmEsUBW!xG9zi*7+peQFn6T z9!<0Aro-{%OPiVLNPNQIp)7Ef{{wx|RzFRIlPpVXhm*9lP%k&ypZV^%&;)i{22;b*8A&hF>eCAIX{Jly?6u2slI!Il z4!fY6o#<{!I$W5_pB#OxG3a4xVbp;FB0ee|d3+Crc9vG$DPAKmEVmB|l|J4Ks!9=R z1yIqUL+Hd5XdGB`2B!A%gLAy9g^9pI%~aQ{t{yU%rt$1N-wi|ECKIpxXWgHE5`PS+ zsy?Y&>$|h?y#1#x^k4q+hm|EgK4Nbe>T1wBjE|Qj0FN%ZScT>!NN7uHQ`E9Fh8B(# ztKX&70FP>4KVx^3DCEq+=}%X?{F+$@?5jTBF*uQ1cG^^rz~!$9FxOD`upMf4K9%uJ zj0-&c4XTYW{-bfrU1}|$T4yE05?uh6v<~_Kh2B1_#H$w+XHK(wDI%%@=6j}DkrQZb z5>pH~%s`sohLcw&s73={Mr~i?3> zA`fzSi25=(nrF}+^DZ2sKLa}qKhBW=d*%(}!4^wL8PFRu^WH#cD7O~P5}C23bxhw; zFDe(p${SL)hXV}$L?)S^er1?p&w*!o(06eWm13K&_c!Lmq^f#G6L=r3HoEZ zsu6DJYZ?LzZ4eOMO!Fvp0Uby{0LFvUlg7bDX0{1Wx4%65+`=Pw;x&=0p{hA+qH$3b z{=OnAO}A!(%iG~%niOn?uGxVfCu%xU#)=HeihT3`X2fZXI^3K4NrVV*7B1L0EKX0^ zoFfa25as#&2w5`LKu%4n!tMtk(k%4r{LY$Gj zM$oCqA*ikh)>Zga&d}BNi)yDGA59cOM;DMeY`Jc|DAxYGam4(Cx$w8?&c6r3)8x%Z z3}rS!jx!gtyv7QiVTalAgL{L8+qw}11Q;?BVRVCPwwEizvNC(6sbSIrW+LsSh2K*} z-K~~NLkByQj*UR_y1rJ--Z?E=ejZGn^jy0kMI}sTqpkbRlD6zs8UzASa&P_+5lQA& zWoz8ia~?UM93Hq-xVThynUBN~bwPbTP`a>rSrK3B^-?B^d+_l?jWU1Lsu+&~cN)T+ z+HoU56T8h$&1;b%kkx_*bLU}nR?GEuQ0;G$IB}F1qRf)|JTx%$C5i~U*jPXv3Z27l zgmh>@Ti{-YZA%aU!N|N(-meFGO*DXkJet&MchQPu`C~PMcmz9x8b3ch|48+kw@ay< zSkz!6P&Los(8*46xGcqAe!?38Tdupro%hWz%s|FIC@qgRwZFS>S6~{OJ1l+{%}%nQ zLbP-Bekgs2JWpA#aM$YH__d!UD*s?YoLCrRs6^bwQ%GnE(o&|8R2BPepA~28Yc6t2 z-Q^qf@~OCOpRyfN7#A5*Rvlt7B4Qz6gvu@^hTkf{Y73#zt6xT`Imnd(&hm_tvE=kS zKj%I56gz@JoHLhnH3u%vfQO?-*e?vO}3v$FwO#O=)zf^uFM5ib8wo zIEZHn$H_B&^Q+U7)OJdnr3_vgp(rC4mPUtu`RmuxpY~;SM($Jw|*? z!SbLd3zq*da$Yp;PutOW@^{9ODz%3GpQ=CWFjrc$9b?6d;o9U{n*Y~-;z&Q3Ttq*G zzCmK&=HJjn8>>YDVU!z=#*tkzj4mi&$+jENmpjd$%8td6yPQEVL`_5usoH#wpUJK_ zgziO@TK@daM61mIw0)#6w^%1re6It~3RRKU<`?ji*U~2puGKhQVNHFTY=Bbej{UWwvQs6UXb{<`Pyw z-vxoow~6F=F39I$F%jfG;h{$ZlY!e|2=UmeMH+e2{wMx=moXxiu<0#b{peX%&E5*c zseSb`Ic!U@Lru-+XOF2|DQg@77I<8^8Xa1Le;5~JtFTYBG(T#eawA5aS+pSI6G{1h z0G-hM`xx@(k7P=2^hZ2Q%*${KF<>IDJa;7n9QBho1n$ z?yg9nAm9tqg*WwNiklOIr9p@rt2^jGjhbO!ao@Bx2)t{i{ zxMvhF*#_CJy~gEGrY1py7j0eSL?0Ko)^N;fCNeXp>dTJU*@Gv5oa#2}iS z;zEPbn1R;U0`!WhopjX~bPxE|1nj;-%7;-GNH=3pYvB9$;;7E$2_`9)4WoLjKC3PV zOhPT^!daYG$QNb(ibR#p9+E;nsgoKM4rw~+l-^X#JYRZj;g7`~tknH^;V2(%ULvgZ zi-!PUrxDQNE<)njFye?1l-NzVA6cm#8BgA{ar;yie3FJ5Jdis?j3ndTT>9H++o_B&w5$B58|Gv@HT-M| zMOlU!df=zepJyzCyXw8p7|u;rJFgbmoVF4pogo#hFa&jd^F#PU%L_HGf|}C8^^N%E z-nF~-?~CnHM=F;gUDE&{p9@e7~9%Ua_ylz|Me;a3@cqR zP;UeV?_3widgP8g1g%6-T*na&G(az*^`80X7?Eohm!jc;y@*(s)Rt>*fg;!`p=qe> z{)JPT6^Rd;iDUSQh@!gv0Cq(tUA)Nw{}Hg^BOdY~Xk$#ya&lm4sK0%;)PG_dG^E=< z2*eu&@}ERNrlm`XXm#xh4{J5e(WK42JrT_;H0b=^b7;Kzw`d1(bYJT~5m%HmEQJCc z2@ftdqyQ0<9TP*}nw5S^v>#4vaukTO{G}8%FR4~{>X!KPi2VdZ<<5rda3){>D+@O!?EP@!6zAhx>< z$N)9lFIxTUj#!W7Ib;Z=UlATXEy;8^6$PCL)^L~`2L`}v>Q6S<|76G0pQOPA&((BI zgadZlE|(}V+XwVt0A(fK{7`T<4nO>zH@YmhqFjB!u%7OZ(0(~-&QuRmxG8#SCR`F! zT`yKl7SM&)7P`>M7GX@kGYfTq5f>zx6nwQ;6J5Tm#PwRLYj;UwSOE7Rd>U{o96apXR@$ww{{O2nDL#3jak`&kLOTD&}eQ~x$ z0+^}OFrh>#Ay|W;wP-}f4N-7{E6ELsVhd%d0uyK`B$ay~Cu2^o)Ew0B0|*cQumA9j zdF_-R{_lzr9&%Tu_uLc=;c-8YSULgBgmVk?$3ti4ZzhF6G{n)I_i(GcqA{u~ z`A)TG1->ubXs%R=!HnE~5BG$Tm~g|el*g%sc$ELbk!~X5X^jT>1rzz+8FlZfd9ttoieCR@*^Iq8YZ9=bWziYsVx?!g5zCwik90MMXZOZ z)Wd5#w7iZBkxAb(j|}1M{6Z0Bb>o@UoLjWf|E;NskC=x+YI&oMOS>f2F(>Q8Y zWi?BAmWitdSfsORd#&VYPUwNNPJD9Q3-7szMSoo<_@2xD^5xOt;5+jE!j*F#^xycm zGlqdH5W#{Pt=uESUYgY&i=5;A!~Oc&=~JJRdGfFU=NCUOr!DWxu2%5?^1x?Hej=F6O|ld^`T{nK!^s=!@t^E9*o08etRyDJlGq zKNGjZ3b_>)x1)FF%7P3|yu(@zq~B95^R9o~7_1}V(@???!4{&zzry_HoB1uuQBEzF zVr){N@uoYu@AH%t(b1yD7gy<%XvwU|cZAMniV681PPynOT4gs>u;B82h7?HI&-ywK7Jw~lcc=Nt3a_(V+{?gT9tXtrcY{`8(*N3$MT zrToFQZ1IeALL>kEoBxivS>PUhK{01miYD1NYf#T{KVCBCeKQ_io|5d6Z~0Ei$A3=L zUj^vDO78awFi5L0G7`CUXeq~~CRNha@l?xS<&xB-M5{}O>$Bb~e8MqoO=8SAqP6GV zH#Rn~ilu8p!f9mAapxS4cGA*iR8X7zpFiOLeggc@sQ?Uh7~m$Q9v7b)#9*#e&98tL zHrPAgY ztMoA%_6gakN3*1>3EA;qt;8ga6RHzw!J+W`_EU5c^xCZgZ@-MM7@)`Nj~`nONWdXt zk1vOWz>}kA{jYK=kn-4avVS1NoyQ6uwJ?RnfGl2@k@^3P|3AC&|8@H2-=(t@u$=ad OaQ*+f(*KW^`@aB+V?9Iw literal 0 HcmV?d00001 diff --git a/frontend/scripts/import-tts-export.mjs b/frontend/scripts/import-tts-export.mjs index 041e58b0..fbe0342c 100644 --- a/frontend/scripts/import-tts-export.mjs +++ b/frontend/scripts/import-tts-export.mjs @@ -149,6 +149,13 @@ const LICENCE_PROOF_URLS = { // Both quantizations are the same weights off the same card, so they cite it. "orpheus-q4": "https://huggingface.co/canopylabs/orpheus-3b-0.1-ft", "orpheus-q8": "https://huggingface.co/canopylabs/orpheus-3b-0.1-ft", + "indextts": "https://huggingface.co/IndexTeam/Index-TTS", + // Kyutai splits these: the code is MIT on GitHub, the weights CC-BY-4.0 on the + // card. This column states the weights licence, so the card is the proof. + "pocket-tts": "https://huggingface.co/kyutai/pocket-tts", + // The bare openbmb/VoxCPM path 401s; the 0.5B card is the one that exists, and + // it matches the 0.5B this row measures. + "voxcpm2-tts": "https://huggingface.co/openbmb/VoxCPM-0.5B", "speecht5": "https://huggingface.co/microsoft/speecht5_tts/blob/main/README.md", "styletts2": "https://github.com/yl4579/StyleTTS2/blob/main/LICENSE", "dots-tts": "https://github.com/studio-dots-ai/dots.tts/blob/main/LICENSE", @@ -198,6 +205,9 @@ const MODEL_URLS = { "qwen3-tts-vd": "https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign", "orpheus-q4": "https://github.com/canopyai/Orpheus-TTS", "orpheus-q8": "https://github.com/canopyai/Orpheus-TTS", + "indextts": "https://github.com/index-tts/index-tts", + "pocket-tts": "https://github.com/kyutai-labs/pocket-tts", + "voxcpm2-tts": "https://github.com/OpenBMB/VoxCPM", "omnivoice": "https://github.com/k2-fsa/OmniVoice", "bananamind-tts": "https://huggingface.co/Banaxi-Tech/BananaMind-TTS-V2", "dots-tts": "https://github.com/studio-dots-ai/dots.tts", @@ -301,6 +311,9 @@ const DISPLAY = { "qwen3-tts-vd": "Qwen3-TTS VoiceDesign", "orpheus-q4": "Orpheus 3B Q4", "orpheus-q8": "Orpheus 3B Q8", + "indextts": "IndexTTS", + "pocket-tts": "Pocket TTS", + "voxcpm2-tts": "VoxCPM 0.5B", "speecht5": "SpeechT5", "styletts2": "StyleTTS2", "vibevoice": "VibeVoice", diff --git a/frontend/src/data/benchmarks-posts.json b/frontend/src/data/benchmarks-posts.json index 9da7ee34..5ea125da 100644 --- a/frontend/src/data/benchmarks-posts.json +++ b/frontend/src/data/benchmarks-posts.json @@ -31,7 +31,7 @@ "slug": "limitations" } ], - "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nEvery row can be played, so you can hear what a word error rate actually sounds like.\n\n**You can run this yourself.** The harness, the corpus and the scoring are open, and\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this\npage from nothing. A single engine takes minutes\n(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it\nspent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template\ndoes the whole thing in one terminal and shows the report in the built-in browser.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for twenty-two of the thirty-seven\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones, so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the thirteen\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2x for readers who only care about what can be spoken\nlive, and sorting by RTF or compute per phrase draws a line at the cutoff so the two groups\nare visible at once.\n\nThe line falls at 2x exactly. Nothing sits awkwardly against it: the slowest engine that\nkeeps up is OmniVoice at 1.51x, and the fastest that does not is Chatterbox Q4 at 2.03x.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes, feed-forward over one.\nDifferences under roughly two points are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" + "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nAlmost every row can be played, so you can hear what a word error rate actually sounds\nlike. VoxCPM is the exception: it was measured but has no published clip yet.\n\n**You can run this yourself.** The harness, the corpus and the scoring are open, and\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this\npage from nothing. A single engine takes minutes\n(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it\nspent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template\ndoes the whole thing in one terminal and shows the report in the built-in browser.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for twenty-five of the forty\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones, so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the fifteen\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2x for readers who only care about what can be spoken\nlive, and sorting by RTF or compute per phrase draws a line at the cutoff so the two groups\nare visible at once.\n\nThe line falls at 2x exactly. Nothing sits awkwardly against it: the slowest engine that\nkeeps up is OmniVoice at 1.51x, and the fastest that does not is Chatterbox Q4 at 2.03x.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes, feed-forward over one.\nDifferences under roughly two points are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" }, { "slug": "agent-skills", diff --git a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json index 07333e56..07c9d574 100644 --- a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json +++ b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json @@ -1,7 +1,7 @@ { "benchmark": "open-weight-tts", "run": "2026-08", - "label": "2026-08-21", + "label": "2026-08-23", "machine": "Apple M2, macOS", "corpus": "84 phrases", "rtfCutoff": 2, @@ -224,6 +224,107 @@ } ] }, + { + "config": "indextts", + "display": "IndexTTS", + "group": "", + "sample": "indextts.mp3?v=3f702911", + "cells": [ + { + "v": "IndexTTS", + "sort": "indextts", + "tone": "", + "align": "left", + "restricted": false, + "href": "https://github.com/index-tts/index-tts" + }, + { + "v": "ar-lm", + "sort": "ar-lm", + "tone": "", + "align": "left", + "restricted": false + }, + { + "v": "Apache-2.0", + "sort": "Apache-2.0", + "tone": "", + "align": "left", + "restricted": false, + "href": "https://huggingface.co/IndexTeam/Index-TTS" + }, + { + "v": "-", + "sort": "", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "2025-04", + "sort": "202504", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "868 MB", + "sort": "868.0", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "868 MB", + "sort": "868.0", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "3.4 GB", + "sort": "3481.6", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "4%", + "sort": "0.0427", + "tone": "good", + "align": "", + "restricted": false + }, + { + "v": "4.04", + "sort": "4.04", + "tone": "good", + "align": "", + "restricted": false + }, + { + "v": "4.09x", + "sort": "4.0888", + "tone": "bad", + "align": "", + "restricted": false + }, + { + "v": "16.46s", + "sort": "16.4642", + "tone": "bad", + "align": "", + "restricted": false + }, + { + "v": "Bilibili IndexTTS: q8 GPT backbone plus a BigVGAN vocoder. Clones from the piper sample; with no reference it scores 0/12.", + "sort": "Bilibili IndexTTS: q8 GPT backbone plus ", + "tone": "", + "align": "", + "restricted": false + } + ] + }, { "config": "chatterbox-turbo", "display": "Chatterbox Turbo", @@ -822,8 +923,8 @@ "restricted": false }, { - "v": "4.4 GB f16 and 14.2x real time, the slowest here, for 6% WER - mid-pack accuracy.", - "sort": "4.4 GB f16 and 14.2x real time, the slow", + "v": "2B continuous-latent autoregressive model, SOAR checkpoint at 4.4 GB f16: 14.2x real time for 6% WER. Their MeanFlow-distilled variants target latency and have no GGUF conversion, so this is not the configuration they position for speed.", + "sort": "2B continuous-latent autoregressive mode", "tone": "", "align": "", "restricted": false @@ -1032,6 +1133,107 @@ } ] }, + { + "config": "pocket-tts", + "display": "Pocket TTS", + "group": "", + "sample": "pocket-tts.mp3?v=5dea55d6", + "cells": [ + { + "v": "Pocket TTS", + "sort": "pocket-tts", + "tone": "", + "align": "left", + "restricted": false, + "href": "https://github.com/kyutai-labs/pocket-tts" + }, + { + "v": "ar-lm", + "sort": "ar-lm", + "tone": "", + "align": "left", + "restricted": false + }, + { + "v": "CC-BY-4.0", + "sort": "CC-BY-4.0", + "tone": "", + "align": "left", + "restricted": false, + "href": "https://huggingface.co/kyutai/pocket-tts" + }, + { + "v": "100M", + "sort": "100.0", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "2026-01", + "sort": "202601", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "209 MB", + "sort": "209.0", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "209 MB", + "sort": "209.0", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "457 MB", + "sort": "457.0", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "6%", + "sort": "0.0601", + "tone": "good", + "align": "", + "restricted": false + }, + { + "v": "3.61", + "sort": "3.61", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "0.27x", + "sort": "0.2745", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "0.87s", + "sort": "0.8665", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "Kyutai Pocket TTS, 100M parameters. Faster than real time in 457 MB of peak RSS. Clones from the piper sample; with no reference it scores 0/70 with 42 runaways.", + "sort": "Kyutai Pocket TTS, 100M parameters. Fast", + "tone": "", + "align": "", + "restricted": false + } + ] + }, { "config": "chatterbox", "display": "Chatterbox", @@ -1436,6 +1638,107 @@ } ] }, + { + "config": "voxcpm2-tts", + "display": "VoxCPM 0.5B", + "group": "", + "sample": null, + "cells": [ + { + "v": "VoxCPM 0.5B", + "sort": "voxcpm2-tts", + "tone": "", + "align": "left", + "restricted": false, + "href": "https://github.com/OpenBMB/VoxCPM" + }, + { + "v": "ar-lm", + "sort": "ar-lm", + "tone": "", + "align": "left", + "restricted": false + }, + { + "v": "Apache-2.0", + "sort": "Apache-2.0", + "tone": "", + "align": "left", + "restricted": false, + "href": "https://huggingface.co/openbmb/VoxCPM-0.5B" + }, + { + "v": "0.5B", + "sort": "500.0", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "-", + "sort": "", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "1.6 GB", + "sort": "1638.4", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "1.6 GB", + "sort": "1638.4", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "4.7 GB", + "sort": "4812.8", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "8%", + "sort": "0.0798", + "tone": "", + "align": "", + "restricted": false + }, + { + "v": "3.83", + "sort": "3.83", + "tone": "good", + "align": "", + "restricted": false + }, + { + "v": "2.60x", + "sort": "2.6019", + "tone": "bad", + "align": "", + "restricted": false + }, + { + "v": "9.27s", + "sort": "9.2725", + "tone": "bad", + "align": "", + "restricted": false + }, + { + "v": "VoxCPM 0.5B at q4_k. Clones from the piper sample; registered but never measured until it was given one. PESQ is provisional over 42 of 84 clips.", + "sort": "VoxCPM 0.5B at q4_k. Clones from the pip", + "tone": "", + "align": "", + "restricted": false + } + ] + }, { "config": "xtts", "display": "XTTS", @@ -3145,8 +3448,8 @@ "restricted": false }, { - "v": "Second-best PESQ here after piper, but 11x real time.", - "sort": "Second-best PESQ here after piper, but 1", + "v": "PESQ 4.02, but 11x real time.", + "sort": "PESQ 4.02, but 11x real time.", "tone": "", "align": "", "restricted": false @@ -3549,7 +3852,7 @@ "restricted": false }, { - "v": "Meta MMS: a VITS per language, 1100+ of them. The only engine here not served by the C++ runtime.", + "v": "Meta MMS: a VITS per language, 1100+ of them, and one of eighteen rows run outside the C++ runtime.", "sort": "Meta MMS: a VITS per language, 1100+ of ", "tone": "", "align": "", From e5e208c7f534e379a3c168260f6aace98ca23f25 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Sun, 23 Aug 2026 12:19:21 +0100 Subject: [PATCH 18/34] Take the corrected VoxCPM row, and give it back its clip Three corrections, all to voxcpm2-tts, and nothing else in the table moves. It has a sample now, so every one of the forty rows is playable again and the second line of the article says so without the hedge it needed last import. The ballot pool goes back to forty: a configuration that can be heard can be ranked. Released fills in as 2024-12, matching the arXiv id its own card carries, and PESQ settles at 3.78 from 3.83. The note drops the "provisional over 42 of 84 clips" caveat, which reads as the estimate now covering the corpus rather than half of it -- worth knowing that the row is still partial in the other sense the README describes, 78 of 84 phrases inside its time budget. The note is better than the one it replaces in the way that matters here: it says what the model is rather than what happened while measuring it. Tokenizer- free, continuous speech space rather than discrete acoustic tokens, on MiniCPM-4 -- which is the "latent VAE, light AR" family a reader asked after by name. And it states the gap between the upstream 0.17x on a GPU and 2.60x for q4_k on an M2, so the row cannot be read as a claim about the model at its best. Nothing drifted again. Eleven cells differ and nine are the normalisation layer: REWRITE on three NeuTTS licences, and CELL_OVERRIDES holding omnivoice's licence, size and date, bananamind-tts's size, and both Tada licences against an export that reverted every one of them. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- controlplane/src/tts/handler.ts | 10 +++++----- frontend/content/benchmarks/open-weight-tts.md | 3 +-- frontend/public/benchmarks/tts/voxcpm2-tts.mp3 | Bin 0 -> 18595 bytes frontend/src/data/benchmarks-posts.json | 2 +- .../benchmarks/open-weight-tts/2026-08.json | 16 ++++++++-------- 5 files changed, 15 insertions(+), 16 deletions(-) create mode 100644 frontend/public/benchmarks/tts/voxcpm2-tts.mp3 diff --git a/controlplane/src/tts/handler.ts b/controlplane/src/tts/handler.ts index 4f8ac160..8d8dbe30 100644 --- a/controlplane/src/tts/handler.ts +++ b/controlplane/src/tts/handler.ts @@ -24,11 +24,11 @@ const CONFIGS = [ 'supertonic', 'piper', 'indextts', 'chatterbox-turbo', 'kittentts-micro', 'kittentts-nano-int8', 'qwen3-tts-vd', 'kittentts-mini', 'dots-tts', 'qwen3-tts', 'kittentts-nano', 'pocket-tts', 'chatterbox', 'f5-tts', - 'kokoro', 'orpheus-q8', 'xtts', 'orpheus-q4', 'nt-2e-q8-metal', - 'nt-2e-fp32-mps', 'chatterbox-q8', 'vibevoice', 'nt-2e-q4-metal', - 'cosyvoice3-rl', 'chatterbox-q4', 'styletts2', 'melotts', 'omnivoice', - 'fastpitch', 'cosyvoice3', 'bananamind-tts', 'tada-3b', 'zonos', 'csm', - 'tada-1b', 'speecht5', 'mms-tts', 'vibevoice-1.5b', 'bark', + 'kokoro', 'orpheus-q8', 'voxcpm2-tts', 'xtts', 'orpheus-q4', + 'nt-2e-q8-metal', 'nt-2e-fp32-mps', 'chatterbox-q8', 'vibevoice', + 'nt-2e-q4-metal', 'cosyvoice3-rl', 'chatterbox-q4', 'styletts2', 'melotts', + 'omnivoice', 'fastpitch', 'cosyvoice3', 'bananamind-tts', 'tada-3b', + 'zonos', 'csm', 'tada-1b', 'speecht5', 'mms-tts', 'vibevoice-1.5b', 'bark', ] as const; /** diff --git a/frontend/content/benchmarks/open-weight-tts.md b/frontend/content/benchmarks/open-weight-tts.md index f0f1250b..5b4ca09a 100644 --- a/frontend/content/benchmarks/open-weight-tts.md +++ b/frontend/content/benchmarks/open-weight-tts.md @@ -9,8 +9,7 @@ toc: false We tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text. -Almost every row can be played, so you can hear what a word error rate actually sounds -like. VoxCPM is the exception: it was measured but has no published clip yet. +Every row can be played, so you can hear what a word error rate actually sounds like. **You can run this yourself.** The harness, the corpus and the scoring are open, and [talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this diff --git a/frontend/public/benchmarks/tts/voxcpm2-tts.mp3 b/frontend/public/benchmarks/tts/voxcpm2-tts.mp3 new file mode 100644 index 0000000000000000000000000000000000000000..ed8946d60bbbf0e8ecd933466436a617d24daa08 GIT binary patch literal 18595 zcmX`y1yCDbv@q}lP0-*D!JXn>2<|RLi@RH)6iRS+r$BLcDPCNQYmt^xoKh&-qWR$e z-ureYlif@*`FbbKW|QNH%>2i8yhQosEeDmjWZWi$s79g zQCBx_s0Gy1#>&mp+QHQh>Sp_NXz%0l^vS2UTR7QxashO8bmX2+i}Q3^Rf{*af;?P8 zTs+*|&!+#~5O=zL|F{4DpOsu~-JT9`pB`oafFb~Z3c|pK;1iLOQ$gt%S=czac?E=p zMPI_CmOm@kuG^Svh%m1x2M5RkaOGt?ivX zeFMW|6H~LZ^GjdXzWw;QdvNso&-vBA+Xuw+$)Ai?PevY*=Y#*-7*JvVv#SXBGV>zi z|6cyz=O_OV002VaL@Up-l#+@qZ7?qZ0fB(VdfF&+DHK|&Z<|4C030Md;eZZlIGi>U z6A7IVBo-(|khkj5e@}xXf{*DZ@kfOo%MRf{GD6hV?@96Viw@Q#cZjyND;8Pe#!}IIVFS3nD=Zjra^= z*Q*a_Y5R*#K?DBUe1K#Ekd2q&x zGN&Zq%E}cvrosW?e@CfYXOwJnXCSJOsZ;)eN=RC>^U3=6|&o-{Kbu;koN=KV(7!V*N3k;T4 zJ_AUj?QgMozrs_HS8;?`qIamOshmGDfVRJNou`tT&a=k#{YkW;?!PVF`>V1mPSO0X z6k%liAy3FJ{m!K1ZAQS>i$-c?0rPiAOPNAh63r*^dgx$aSzfE=8Bp!*+Uz z=Z0|gk9+jVus9^D+9x@vHe9ThnBsKUo^+ujatI!9Z#669B z{wH(q#Zo0kEO8ALG-)Anv19i4gBjy3MbsDN8gl?Dn>RS@ym^f>*l>&AC95E~08`cz z^Gn1iY9O0h-s#sI<$p-jM0#YM^NC&*%as!`>eMt%brcqo$V8%CZGd3@kiJlDyE|vA zp!dy=bR}#`8lP&xck=0Kz_0)7Pp0yUV^C|<&;NaSLDiHRV)Xi7H9^hYC*KgVABH?M z@-t3d$wX+85Xa-e#VzoOmBEU#);FhG%!g#+T@^D|eX6_lR^y&p{kIG@rA?5NxK7G~ zIHD%0S*+`?eY6I#SZ%Z;!wqgd|DS(dwa4@j_d3Kc0*8+>$luO@Jp=D;y< zgt(wA+}@#NNV58?z`f#G!Jl>e*m3<0TesSUYhRkq1N9rL-PID5v$h6_s!nTc4ds9R z6^9<>Sq;ew2^>8Uq*<4Ab(VAXO`3^D0>AiB{uI+uR?fV-E(ou)&6!R(HjS}h!h=tL zd`S3-3{zmIdV$M3HXyyr*Aj#{InoyH^hprMqAHE|df{nk1OwU>euYD>yyBGP%TX|N z)#5Cz9dX+w%JE~zY8Vb*yv$u^HNsa+Qr+Hj3cTBx4=z~~GH|MJpOtm5n}d(hBKOMH zBt<(~zCJ+(((I;J{sOZB7$Yk^fdrl3dGS}a$)5e_0C1Xv-F}H3FgT69;ew}9!b?&V zmLg?LSBMA3L{_x^D?p@{yGCBa{fgQG(25Y3S8uPOe=A>su~ z=jWFy9zWoyZp{Fs%wq@=v3L9;qg}=8?u2r;ur}2WyN8hV>d#qH4_z+4-QQiXX{bJC zH+=DR`A22mXAq$Bc8{3F)!_HaDK;&*xkADXv9i^NK}Izw^mr1vSqZZB-{6~Ece~+J zG#DSZhQhU2l_~#=KNHI7jY<}7le-`E ziteC4=TQa~Q@qtnxtzYudIxgLOxfAlS#Jy>;5MCzYNgQOC`-IxY?;am`{(#$0l={M zpgj~1Ewa=<-G5?u3zw2OJPP83&6syQ%}Nl>o{O_GQ@gXxuC3ZzNv$#@&N{lCbE(ZP z6sA?~G>~>(dCIDatgJUqq}Uw<(P0AO)?nmiP>rzZsI6E2sRMy5VHV{X)rmIw$qYiq zK4f=o1PqY!YcA8dDnjzt%q-13a^-UYtJJvYJt3oPjb_0@=(FN41w}HHc`HhS$O<`r ztnSmI1kWa4zYP=b-Ms&Jr4!ChY3Y!PWb*7M1AyJ1{8;MA>}c-y*U+5g5K9^EwKFL` z0FVL493prM1mJFsm+BX@pFLFQO-5l6r*?|8?&LrJOR>*T2S zczV*2`OW~SOli?AvZ8442{hnwo(IF=piao<4nb>P-I> zy=$zj^W#?iqmh!nLZmDY2_B##b_YSK80s0>D;R-;tQa7yT&FlPMG~2}61W%MrR*oG zpJwJ07i67B3bMCY&LByDm0vMA7ZOM?qFLj6wa=Mcl&NTu5w9I~xxlHa4TxsuD|1JGH6J=$W`ToX)=`b zD2Oa2%CBxU=tk3sed_&*DhQ2$D9$PHHM*e_*`cilw;$c63C9iwxHi0hVz%k#_kZ4n zdOlWJ)=r2UNt-vz23^hUsVqgeb_G)BAUfKd_XNnXn9VoC?1__{@z<5HAl&m2%0Gw| z8AUEHzjq=?Q{H{SCE6wTi!Untno#?0bubzV((1kA{nhngF!Sa7(1zmbRb`a1$Os*0 zD_O*8{j*;Y0J^+;TV9}%%}(v+eHoqMfQ_+;0)P~Iq0egCfV!#1y?{R|urd!+6 zQnKJ)k&6{O5oFRb&ligvjE$<~Toh_Fv_xn) z56+0$NKZ%j&AVN})_9+_lTeGL&yn7d^l&bmQ}7T?Uond;P}#uXecMTefP%9QpIbf~ zmnm69fbJD9+tDD0IbuJe!XuGVp*P?ngW4)-oov1kVj5KG4gi3eLK94gXieBkdAK3B zXq47!nJP#JaW(P=zU8k^HVcn^mz3&q{k|qh<@$=Tv+z-dv-WP3VbZrWA`t$UKe5bp zJ&M#-7W@yGkV%m{Ob&I)|~_JMx!Oep0t?DxdJnO_<&udev(P%N2YfV_Es*O{$RP$QE?< z>Siq#CPr#s#TxdC6j6;)wTGDPy}l(sY>0w~AH)=AIUubgFUktdnl%D`9Y!l_mZ3wi zk(E}w#sfsps*$-7-`S0pMv+gd4EA&OmbYm?7j?gPNZ*w8A%P zr@ZiCYPYOYRn^QiF1hX=`Een6r^NjjcH9Zv76+@NkdPm@2T%=W;op3fN<-~Mq2@Al zh+E0(3#kZ=s>_F99IHErwxwDC<{xB_V|Ia*1W^`JX}$y6U5a6~N?>?1zc}4bf@@cu zn!CS7%KmGf& z=pn|`Q<#BPrg=tUi0;5=zX}|K71ySQv@Y4V$jPTPy|X=`m=Y#-%YCoD!XJKAOh`({ zF&ZZyG@=IpAh|-xm4l(oE9%!D=xE~gQ1B_H|?xfkhH zvx6Cr#-|k+*{)FfLM%stBaaUa+0?Le481KpU19)}UPhCaRR2h7GPx(!Fq=y|yr!v8 z1*65WR#aTSK%!E2y|?@Da}R@ZRXjak)Ox*Dn<2ozz+ICKJGfY#ckIzME;G39*{=iu z!-I`$UNR;3Rvcem(Ei=O)h+)!jF{=X_Z1!@f4`J;EV#2ZP088750w>OgnGzW>|CG? z$eDax1h;Qw9Hor1WekrG%Udhh$o4pDjbpYMHOqY*$L>E)KJDH%{;*Bz3TlkmyG*;|+iG z6(c2YTMqO*I((~pdwU&bAvwt%{(X~2j#eH5>kEDM!{B(dk`9sG3paZ#HTR<*$RJvh z&i$yNYx2IfkqXXzcp7%2$3r(gV{^HEz27z`=2_n#J6(4l_jZ_0P9wH&@B9Og zx?)avqhBkc!V7RNF!a1X!_dV|Pj;I4~C|OlQ#(%D8Id=O+af)>(HhgQlj~?S9 zIL$P4Uid<)Zkb?Br}xF^kFiX(ERKtjZB&%6j}s|am<+6#h^-h3KRknYFk}}gi=}2{ zs*Fja>EMy6{AgDAZ~jaO2g73n_pmhQUf{pD?~4hU@q4N9Z5+{wqkyuFSz5&Z!M!MV z&VMxT&X`Z)RHhQIwalp^D%;Q2m<+5W7aCRa1ueAEUYwB}J;3&`{cm)#IKtHAQJ^lx z*VZ?24okZ2uHAf}+fbfX)m~HiCDnT|!DK$tQ1lqawx;{kM?_I|XGPQ-7jw=DTrveq z(2HsBtswc27j*_d88j-UJpw?VHnqrj8v!WsY6pYSLBD>K99j}&$&wKB>nZ)W{-p+h z{bD2c!xr$^@ah7RV?Z0j^t$SQrZx1>#s_!s!>jM!c@|M^z>uNHAmjv9c~sOf#R;x( z`rp3eZF6d`*LtNBj+)z+)a&o!G`Wjp!&tZjm}t0N7!*RVtCkob8meBKQ>6n zd-(V?_aLt>{zHxD2yh&xi!f0XPqel>J+)ozc77f|EC4WnrQ1$)d@xH2S9fxmIULvr zi^hX6u2a(E5^`E<`67a8H=ncj%2VU$EG6 z)`3+Ow?Dn=+QRMl?B|1nN`%6n)?aw6M5o=Kute1!7uO00YO1w}|32o+i9%Ji?aOo# zNqc-Xp}(&iNhs@<5qJ0VT(v~e*%7`r&vW2|mLwD~L-HTw3?`1Xzi%mnR~ zW^R{@RAxjvRm_hhZa~stck^09=6zv}DiI_3GHS~M=B4mXd7qAs9?oRUCw9XO+{KGM z3r{Zjkiyo+Wa&YLYBD;h>6l-7Ry?ySEsxWi^P`v^_8z5G_Dym!R_{6g^GgAuv67$1 ze=-xETl;zZd0K}4hk1OqbgOQYAA3|Zl+jfccGd=CC}mWd$<)Cus$Z3Yr==5G9DFkk z3c_(DM=&oM#$3K&m58rrA@;7%rNT1ygtgw}k#1LfGivF+{>ke@pT4hIa~UrF+GdG- z4W^;_)1d@E(%3?~&ci@oOL4j$uCp8)u)IfawEp7-rUI(|RUDOmD)HUeKNHH)A!@F3W zoYGf3SwBdn_~_mJtyVC-o9!mi53BM95Kz^?afqqJ8i^qR5FQODirGj#iyfCPC%S=} zs!J?RsT)804FO;j3yWz7x?ryC@X18(JjjH&n=dlhd>MwHDcNTgQHO*BXFL%AW}a z^bRU7e*NnvcDdxl$-P*-=t7g~R}c%HucwEjLy^)`8V@J7guIB(XTLQZoLg&G-cXs$ znD6R+3I5La;&Ah&$>OGpYqEfQV{dJQWQ_Y@raUs80*nR*iEC{6XI*yxCL24%#p;Ho zj=^~1RR8rus}7%6gkg+*Qej_UzhUSx+i}v|7G+x6ynkr<{>ohBxDiD z=5RefqQP;DPM%|(xa`@nsMjmUK@^3Kj!tLzn`$KP#Ik{22w}E_XXH>Oiv82(NwdTDZcbH`YjDIY7* zip%?Fe*he8aN|^7qhZLNUE>o*Oun=7GP|gp|I?`DRnyD!#B7kGM0TwFceXSie`K6hOv@nBCC%20*0ER<< zh`siq^TB%>+2*6H25pW_D~GmU|K#uAFOWA0zpBY)YGx<3r24eCk;m^AYexx=7+oY3 zX+bMju_Lvfh?T3no+#lOsF-!!-|X*ZGt=PkYbc6fPaHFINqP21KIO0CJ<8)2@Zcm* z`9oey&aj1#=O17?Tr;=7tE8Z4Wn1judd9NbP}%Y8GQBZj{snEKe&`Jc&rNJb9E2JRhSnnrqCyaLMjz0>5*V1 zNfa+5Oh2qjHJBu+D7us4O;fOy@O#kpX&T1yF6_T<W{ntO0B*%zqgh&#k+YeZ0cW5B<@-Afa6vP^G1>Vg*{J+I4#CQj`Gw^_2s<+T^7e);L|^QO zmmQTAew~)n-B@KCXZejXJ}zrMT!^==QyeztKt3Bf>C_qb?2m$jONxW`2;Jvi64eKs zMbG1+ig-8ojO?Z3OsY9n>WhHZd*#?)PM#tjCpv>Pq?rQ77N5W|YKr}Y>zfh5vh5hbmIgzpT&n`wg0?9h{zJjS zdj+YBhU7yxjvI@sVR4zR${f%qm}8wlk)wWK3AW)v%n(ztwz!*uV^-)QcO||1q}Geu zeP9+rt#Qikp6NM~e?Fr!=CeN$uFT>RxF7GXC02IocTHG@5-S;>(e1TV_Mz(C=7a0| zD*J%u?25D$Mvjbilmhf~ZY%~$sTnGyh-yt9aG!ZN$&attxpE9u+OY6^{voI^b}-AI zmbhS(2GXH?NR;rrjioq22S1XvG_;{;3zY;faF&cWc%M!xJ^~NuQ%g7ak)L1G)z;NH zt)H2z>7(_Wj%a-9@?`Y<&v_{~!{N^5qdZd*HM3iyAkUgV#Z5YhEX-pzemfzmXwv`k zpJ`9mPqVlCS$4^cNhkgd+-PWX_Np9-+3c$|Ydcd)A@D;13G-bl7hw~X%q8DX+HZHT zN*)LEf)IBULPJIf1mYm*%VinZ_tPn+cbt>})RrF&hpfXr2QgfIs#*0e^3U-vg5wp^I&@S5$h2urnDZOZ zODOPL5+3P86ivL)DqnZ!DT2VDj7hsZf5tWJxaf%duD5XgCS!Sxot5)IR zMJ2?OF?so)HqGo(?=1L^2gvx@xcPDWiE(Oj1LPM&NnhXDYeI*V#@rMW*; z=qhVY?m_yZ6YWWKBE8rA~^4MTt}}?l*qM;8aB)hIimGm{`c~;H7ZC zxQby>X!YGU-BC6qN_;-aEYSp8`^k?hgF;j5bU^~~`e?;+MmryTN;Lt1BqWy!TBMMl zjSN3IXF~lx_9#f%3R@BAT-kHaI@#DZX&I`sFsEk6QE9S#U_HGbO22Xnx)oZ<|F8a- z2mtH4nLE>R_X{LAoBYiZy?8JzjmhnU{|X$38F;N1F@(T8Ap~;$nu9`^6P*1>W3R|7 z+NY9ow0Ug9fOF)|`{MzbBZf`sf~Rl3ZEJq>)%YtD_eRdv4AHj!NiS$qIbAxO>@JSs zGGplg5CU+e)v+T&^AbtX5(>xf4zDvw{dv%mzcHHAc4^E}E6|F_gbNtH-^5yUAk54D za2w9EXKhD=nERYNAC~jg6vIz1(4>e^9S|bj1aY!?j(^tE{rB`;d5O|n4yt4O3!;gB z=vavs5ffR)a0svTTYG8u%nkTVIGiaucGXHbwwelr6xAX@%pZRr$v;9W@X@&WFUj>4 zF&&YZ3OO2964hlFF8~E)*II-bnKoodVCrJS3GE9H)=Oxtv`-;zXRua?bi^@3bH~Qg zP~$p&e|8Pxk(ZB22$B9duaRbPO?{caAUKX31Yt}E08lca1dc%XXE*v~^|L-t#)bVR zi1|;pGP&v9To1vII?w)M0NDJ@sXR-=gPyF${Q`XA+}1P?&t377fwZ?Gw$f!pvq=!c zfy!S6JS6+$DwyZJd|>TR2*WZYnI|rSB`=e%QW0C5hc;}(55UJlZYX;UtMGq2msk7p z>-r@)-Eb90@TINU;~=A~J>qGmq&p}%>V`OJmLtOn%|+Met3wIFUe`8u1O^xQr-gU- zF&tUXEpSIo@>?lD*9c@d*zt@`Drb4ny9z#`g{Pe&`q>Dlb%P%34^H*kBcA1aWzPny<-=fW$B(j`ZB2^7d>10OtGvrV7wt zGMt>rz8FOOVf+{Qt7x{>4^)*DzZ~CVBn`z&F?W^7iSfNdtT7iFT-k$V#6BR>R`ek8 zw|AU)cZ;dzVv!%Pq*57b(AN9h6|b-y9B-ce-2hO_u0?r?>KR9oGyf&g8v$k58}q+Z zbK1yT-MHAZA7wXmP0*}8fJH$YUUpaTJbt|O@DsR zs=<7HH~Q=!2Y^;_ZOTg&Zy0E61R}sMzWXHp{rjOWTJ>~SeeXRf8xJP49hs!cU}R+4 z!b7jW1riDnGQ!6xhUn?Zf-rj@KWP|YJfhP% zs9DH=t56*Z<|9kb&qXKR+fANj5n=npe8a2P8)@s2u`{*7eSdDB^z2^_0c)n~)sSl8 zk!M$3We{oapPsg}32E~F+cA7q?+}ma{S3Ju zE}aTW+qv=QE@N^aGiTqH{nWHOLi*8u?A@nD+A$<1T=1*%T8ptdWEyJ=k6#nJEy|e1jC9#O7hnuFXt^f-yg`UOb1GLNykJ{PM~iPcV@T>q!~GV}A(5xUxDKp${Cl({~Fh+Jh|+P{#Es*vb`mDEH?h@S8x4SGcbW`v)}83qA>itQ7D=+ zcqs~%Oj43YF<0yGH%1bBL2Wiu5YGFLt>!<71~E9M?Xb|e!HSjV67LWt(V6357}UA` z7q*T1;thk|hi@)*gjf!xjUq5;5#(K$O&6=4j@BoABD}uZU?5YIr+xLASe+^Jr%Wy@G@nF=#|H1jsn0zvd;><~vP1+fI|xC`LhK6A5i^aN)WQ=2-eS6Zr5t;y` zA3%D!AN+OCr9w2!34IeCzt_06|uEBJNkgG<)`AB+?X)@YIMo|hl z$&AiP-KN;#;Dc=QVJ3?Rb4V(qgfLVMGr{)LO9^dZDeUp&56llMo4tFoTRQBGoen|= z&;Es{^~-~Cd7M@<6)FAK3v54xK4LWw+5gB|Ag2xS7lDwL#)Bje6?+KBxiXRhB!Bny z#B60Qf?&;%9q#i*j87h!Ebi8yxswK!XtD?*wm=0h6|W0xn&+IBiRxuGn%`li%^6dX zHhr(HM!Ygh3}{ZU7r|a?i(hoF(y{AZK|G*nw;sJ6HxsC7I(?eWo-&*EC5a~X*nNDr zXLRrJ7=%D{k<=|`Xq%qA>ps^5zyajKagKvEoem5I=+FKiAt07$%8@J!Dc6{5!>jEe5i9Ie2#knr657E5+Z z-4S40hTOoxtr4c$sfOwIekcVUb7&3!mdScfvvz_>-lYBg)%7PiY8 zVS8o#R7mCEijvI1e`>3aMX)A>p^V8CP!Ptb8zlQwldQ)(eq-Pvl;_t1aK zqCq$7i}jz~{?3Rdg(-%hk}#$1~eiRXL_q~TxB;XrEIt&(;owS&EDdK)_fZxOcuFeS`=wf|D?IN zx|;3Y``+ZEN_2)8=_HFf*I~m_LM<9DbRs&{{WE_Q29%^7PmDZ!eEAY)=5b0@{9$I-#d@FI!<$x(*aRV7gP@VexKgH?Sj6Fi(9Z2j8(X8?U>|fg zAT%KMyNKjG00N+W5F$~I&a6P`VOj_jM^Zu>qLrPo(j$xvVA2zg5s0AZi9B+ z`~FpH9I?oTh)WsLVYl9Y)u57Ed}5p$Rz$ZlCNgZ+m*ZlffQYkp?7LzCeG%8PuIHzk zIc^j+9DqWE3T^^%3a>UtLbIN5WQOc;4VTBYq?Bsmv@W0O$v}Wc{`Th6nARPNhl3T* zY!tGzrI@j6d?E-B%J*llkr2E;Eu&@?_k&Vg0nD5kBS*4;39J-+@_z09smtKK>5#Yg z`8&ss?)^z76y&ja({xo1KJ=7aW&Cyt6xM8a)D@qdxA(V8XXFuIWADqht*~$zZ&O4t zSpY@NHDc9bV&#AB*gb;*7;00xFU(2#7_C}0c{fBIfZPT_9|SMu^HP8IiCFF;XUi15 zQIflWp!HFdply*vawOA^Y@cZ^N_%#wdQ*rz`?CRH=hi2GZ8B$W)m2y;LbF?T0S4o4 z=e0LMe&3pj72~~Vt0}}pYJ$c*j(GREB@aJvYSI2c?}1alrYM?#dVEMt67k@WjM@?K z4cRLlCSTgb>E)`#kRrkW2h{opnb_oo&C-4jTS{}qrO$y1BrMHK2>WCO=zbMI0c6DF z{r&i+^n^h^!%=G$l`F5@2q|du$k@Mv z!=sg$w|~~azADx+-Tm%1ShgmaTO{O|j7F#-3bk6_JJ{DVltUR~9BOl&XnvkMByez@ zD=(bh>!YHGuO8)#Thu?Eh}%s(b%aOKwGUIa=C@QasJJMzG>13BFJ+d-h6^p+n_T;; zKwnEL2DXt+>t2$v@}fn$_soBeD@Kt|C$5PXSv-)d7Y&SGc)V4W{-6ALEF3Qy-LCx} zfX5!s^*8?yTAqd|CMWHw_=wU&9fp$UmS; zQ8=8XQPY-DMa^xgS1}Aqea!f~>BPDA9Dg4GH~`^WMyP(nQ{eK{KO@mM?#}O_d*QUm z`Ev0@rn01WNYw0d>vJO>s|42yRmeDjxdOf;l4d>8V8Pgzj#>h}f&LK76x?CyMgU<+ zZlTVvlh4QgWv?_rGkpgsLez7*SrN&vt2e%G{l(Rl7J7(yRjad<=vH|V$?Yu9E=|C0 zH^9SsEzUG&hdS)Pd8%q(_^~hUopunz7kWp_DqAWeMI=vT=mJSaI_5k^VW$nKD$kVq zU5BJ^DDN=tcl~p;F+0V5~QRF^KQp!->B6ECD z@YoRHflTATg=UfP^>889`N7DD9qr$&caIa=6JOYc)EVZ0V$^NV{+OrvPpZ!z_(d>7 zPD8+13|;&8rOsHh8e@g86{Yr4Vv%fBRoPj-PGqPex*<~I2h02)@d&yl^Rg11FvJOq z`k}_fLeqY{!PD(*Tbj_)RNKB2QLY9?S1xa}^>^wbp=J0-qzW2$&U6Xi2|m*~7Jf^& zhT(_tDd1qNQ_<;?5P=ia#pUZLjDBHDhEdjj^%5&F!;C+f#)u}Dk!!*k#=VXRLI91-!>$Ip@dxj_%yz0{dX^EV z{*+yvo!MFP%SapT6`068$L^ewMGM;1sCtCld~6&XV-MM!p(C%X0%9!x z0=(g8f6-I^M=W%YyI+eo`NdQHT{ubHGNjTkj=Itono?wFubDnh_o5*{8_2tmG6eg zf&|asM@@(T$8`aVHDQ9`hBWt@SPk+jF%}8Sc5MD7Tdke0HJWn*kj=h?3EbJQA}58v=Rs)*|`j`iFDioO=W z#5GNN%-^^Qr^BC?x&&|LJHK3P&b`E_n|jIQI1e{d#3mMRjEb?&!Kl4rXJZJ*Zos;g zae{J=2tdhol)FQKD^q~Jv)9l5E&ymu#dp6!O6x_^i@=t+@0f4o``iqfLHOU-i0@Yk zUj~*{rqw$eor@RGaQ^%o_eFCRd9L*7UoWJbz<&IioIucfKJ2R%S;Sgc(CU1srlzQU zJ9zWPT3nZHi>_dlD=MN`7Ve$V2;9{cw4U_su>Y&v2gRsxuwMP)I!x(*$Ox>FJoT7NKJ{IGTd-ZwX8GU-B=Q@kEby2w2i)-I$YIb`r7)D zVwdnClN*_-_GrwQsMPzQ$+&tt7L2|2iTrEEecGjpUijOedamt8wWr8=brp-hR2+oKpnLi%S1v2cp-l5^zM#DYb3M zuabgA1&=>OBU3`=8_L40Tl*6%HU_a^SYN~__Qx}+82b55*Eds72Vcfxl8_oz9g`g@ zO-#sI%f*<&SG46pEgJqf*~CPp2`dK2(9`N)4+r?fU_-_un>}7*a#nv;j`WyvQVUmrP+P zOb)R&a>;fxB@GoL`OPvtZLMq&=!ez~p8yvJ|r_L!b7deKu&wQl$j zFcHU(Az(|E-rbIW-oc#Vu5PJ2)JduF%onllbR?ZqT^VUbr^Kd;#ae`7sQG%h$J?vU z;5BuOavfbdENzCfI~YwejeE9u5Ty^StRx3@RR_DM^A8sMNG~u8Y`utj?8dB`KPZ{ZaclxF=1w*Mzs?Kh`&?vLnKCgYjprG;D z{{s&8HVfS8meG>Wa0|G=o<~usv|hZ+r?i0zGzwX@L!`}b>i+pz>X+(wpE#MVX|;b& z8L9dm)E7o~`9-<9>S<+v_Sni^y1-|pYwj=td=)gY1fZjrCDe1}-ANE!fT5r;zz)aC zP-O+R=$KcX2{N@ezJ#kdN#YLKxTb{DEZ=EQ4gGUn)h>p)iXHRRt zQ^el=wmcB>|2rImG}&dp#0AgdBgdx(I1&zu4q*)RsjQQ@N`m-uDQOh9 zpQtB_&O%CjpSSrPn1#a(Pe46zL19Bxk+j5%>mdtklBkLwsv$P|Gt4?Xccc}8ZQ(#t ze3bfN?@r9-37T~G!68E26Q9`8T)j~*kMn<7(5@*_QpG?>*eY$M3`D{9hKk6?Ixy=p ztBw7#F)k0L)mDbO*yL2KaI#vOh%e%D3=}jj zyv(dsn31pxs@@pk4%J(-2(Nu`tpw!r{;RoUBj6YEz61=W501{ymEo)?G zcV1K)5s(h`18XQDr}9VpH$>b_xbkR<6-~}=7c6*U3BA8Pw~&%I62z5y?}ON0AU<4e zTP$3;j&{;#h*DhQZgd}+qX3Eff(1I zxu`oSF(d(kPg~Gdomtu`JnKrP;yX=aS>DF-4rSjM@oJ*G7g7KFOTpobqb?qXbuY}u zcCr!Qr<7e z56|{e#N{ugIR(isl+)*;TL1fqM= zYpX5dJkABa3_D{mW;tx*cCAA{Vxli>M!jqe^^LWLAp@?+Dw{sc57{lLabWg}pe$~ej#sWqGKtFZumI$Q2exrE7) zJl$jiYQxw{!mma7$f$(Ixt@#hH#@;}>=G4ZeVU76wzU8?p1}gi$KcdCGyEv}iW~2E zBJxv87N8BkKCttntV}yToQGvhIv!eWg{%;fWgY^hn+EV|yTnR1$f+%iKl{Hv^?yDX z#&M{5aMDz_Mo^x%JMxl(ReBSMO28fmNFWxv7!#ES`pC84Cz9EtL67kVYg9#M(?pu) z7h-COu~p`yuYuQM)e~uG5z%^CYqVS#NSe`azcoP6u18of&cw@DX~%PX&(h^#Qn`dW zPsIW|LN4ZKnr77V;MufPBFXqH&M@Dg`JpkK1vRAlaKt&l$)69WJk1bA2|BFJy~ zU8t>VXz7r0>3QC~;^mXpuK*JABGUl6-FHk!%{(RoqQgH`8MA6cMrxP#yUB)QoOxot z$IgF6hqy7@cli+1LNtm4Rn9m)BC5B$Ac~QB|eW`SG2;5k(Q6$n<2` zT=;)1RzTeJGQSa7zoP28EzyyZ2AWq;Hdsf_MGi(Vp=E^;n(tO(CvSkQ4y0D#M#JXa^Ec zOwhBC#2|+UYZ3Yd5MfGCVVB}EX*upb`wv1e0(K=;<8){hDVeTn()|1&Q=oPy5{I(tz}1{qTw&12b}(I;`*nmB6nZp4I$05S$*#9Fo@;o( zlTLxq5gI+<3g&kEi4(>0__b2u7qAZm0szzYZ3>jBDWRWtKQVHz2VobPd#~KOlnh`` z!REqxpZ&j(Fj=O!)a}cqw8_d$yQ2;$3;ISfjW*tUB3`0l3|q+8_G`QA*~q{Bo9@W+ zuV0Spal`5DkZ>iTaOWk_FkDKfuj%bm`Q(GN-mQyrU7L`f7H)7R1Q&z31xC)$pnz%kZ)bo9^ z+%Zt*>C-b)Ds=KDNlj)mNR~@D)^?oHm#Oh7EAxumol9AKFksG;r+Y+p z%*)F+);|@xWM(tF=AN0NCt1st>zp-Tt^2yUMPSpFx#v9p@_T0Qzn5anvPwj&Y0{P2 zXA37SRz6kKF`Ms@(A}JiCzlIPos-F*y76M=)&Cb)Pn`b5C#Bn2^2=&gMlIieHI_dm zc+zgCiJX#}oaU3bipgwosN$=sLX}M_B6oYmUno4Vn8wBA8?M&A{dc0vX2WB8ra?dI zd9>b~U72foB1Y@#Ewj2i3v`p^Zf)Nj3QqrwTs`|$&LyNCbmo10a(Yk2L#^p6O;)d( zzQ*F-u`@M!W_Pai`i5xCt&^G@s@5AKAZ)s%u(vECn{TC?X{^ghA4gwTW6-e}p#uOw z?tj(HWnnJBv+z-4?8bzagFKF*p#$nBQi1{G{@2W>9`>^^vg96&-WYqppns^je--lq o2F4`}3=9ek3?>c07T6FA|6%BUQj*|+ss*KgEa5+(ZX(5U0M-iL3;+NC literal 0 HcmV?d00001 diff --git a/frontend/src/data/benchmarks-posts.json b/frontend/src/data/benchmarks-posts.json index 5ea125da..7da27c3c 100644 --- a/frontend/src/data/benchmarks-posts.json +++ b/frontend/src/data/benchmarks-posts.json @@ -31,7 +31,7 @@ "slug": "limitations" } ], - "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nAlmost every row can be played, so you can hear what a word error rate actually sounds\nlike. VoxCPM is the exception: it was measured but has no published clip yet.\n\n**You can run this yourself.** The harness, the corpus and the scoring are open, and\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this\npage from nothing. A single engine takes minutes\n(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it\nspent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template\ndoes the whole thing in one terminal and shows the report in the built-in browser.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for twenty-five of the forty\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones, so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the fifteen\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2x for readers who only care about what can be spoken\nlive, and sorting by RTF or compute per phrase draws a line at the cutoff so the two groups\nare visible at once.\n\nThe line falls at 2x exactly. Nothing sits awkwardly against it: the slowest engine that\nkeeps up is OmniVoice at 1.51x, and the fastest that does not is Chatterbox Q4 at 2.03x.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes, feed-forward over one.\nDifferences under roughly two points are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" + "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nEvery row can be played, so you can hear what a word error rate actually sounds like.\n\n**You can run this yourself.** The harness, the corpus and the scoring are open, and\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this\npage from nothing. A single engine takes minutes\n(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it\nspent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template\ndoes the whole thing in one terminal and shows the report in the built-in browser.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for twenty-five of the forty\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones, so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the fifteen\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2x for readers who only care about what can be spoken\nlive, and sorting by RTF or compute per phrase draws a line at the cutoff so the two groups\nare visible at once.\n\nThe line falls at 2x exactly. Nothing sits awkwardly against it: the slowest engine that\nkeeps up is OmniVoice at 1.51x, and the fastest that does not is Chatterbox Q4 at 2.03x.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes, feed-forward over one.\nDifferences under roughly two points are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" }, { "slug": "agent-skills", diff --git a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json index 07c9d574..b2b410ef 100644 --- a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json +++ b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json @@ -1642,7 +1642,7 @@ "config": "voxcpm2-tts", "display": "VoxCPM 0.5B", "group": "", - "sample": null, + "sample": "voxcpm2-tts.mp3?v=76ce9f86", "cells": [ { "v": "VoxCPM 0.5B", @@ -1675,8 +1675,8 @@ "restricted": false }, { - "v": "-", - "sort": "", + "v": "2024-12", + "sort": "202412", "tone": "", "align": "", "restricted": false @@ -1710,9 +1710,9 @@ "restricted": false }, { - "v": "3.83", - "sort": "3.83", - "tone": "good", + "v": "3.78", + "sort": "3.78", + "tone": "", "align": "", "restricted": false }, @@ -1731,8 +1731,8 @@ "restricted": false }, { - "v": "VoxCPM 0.5B at q4_k. Clones from the piper sample; registered but never measured until it was given one. PESQ is provisional over 42 of 84 clips.", - "sort": "VoxCPM 0.5B at q4_k. Clones from the pip", + "v": "OpenBMB VoxCPM on MiniCPM-4: tokenizer-free, modelling continuous speech space rather than discrete acoustic tokens. Upstream claim is 0.17x on GPU; this is q4_k on an M2. Clones from the piper sample.", + "sort": "OpenBMB VoxCPM on MiniCPM-4: tokenizer-f", "tone": "", "align": "", "restricted": false From 0e8f2436d0228f545d1d97d51f60102ca7f91166 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Sun, 23 Aug 2026 12:26:13 +0100 Subject: [PATCH 19/34] Add the four vendors whose models are in the table but had no chip Kyutai, OpenBMB and IndexTeam arrived with Pocket TTS, VoxCPM and IndexTTS last import and were never added here, because the chip list is maintained by hand and nothing checks it against the table. Checking it turned up a fourth: yl4579, whose StyleTTS2 has been in the table all along without a chip. That one is not from this import; it has simply never been there. Cross-referencing every row's MODEL_URL against the listed organisations is what found it, and is worth doing whenever rows are added. Two apparent gaps are not gaps. cosyvoice3 points at QwenAudio and qwen3-tts-vd at Qwen while the chip links QwenLM: three handles, one organisation, already represented. Logos are each organisation's own GitHub avatar, which is what the rest of the list uses and what the comment at the top of this file argues for: one consistent source, published by the organisation to identify itself, implying no endorsement. This pushes the list to twenty-five chips and therefore to three rows, which is not what it was trimmed to twice before. Left at a readable size rather than shrunk again -- the measurements are in the message that follows, and the fix if two rows still matter is a smaller one than it looks. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- frontend/public/icons/tts/OpenBMB.png | Bin 0 -> 8454 bytes frontend/public/icons/tts/index-tts.png | Bin 0 -> 1562 bytes frontend/public/icons/tts/kyutai-labs.png | Bin 0 -> 6584 bytes frontend/public/icons/tts/yl4579.png | Bin 0 -> 1528 bytes frontend/src/components/charts/TtsVendorList.tsx | 11 +++++++++-- 5 files changed, 9 insertions(+), 2 deletions(-) create mode 100644 frontend/public/icons/tts/OpenBMB.png create mode 100644 frontend/public/icons/tts/index-tts.png create mode 100644 frontend/public/icons/tts/kyutai-labs.png create mode 100644 frontend/public/icons/tts/yl4579.png diff --git a/frontend/public/icons/tts/OpenBMB.png b/frontend/public/icons/tts/OpenBMB.png new file mode 100644 index 0000000000000000000000000000000000000000..b7441ee304edc22334185143521bb92ce6b47f64 GIT binary patch literal 8454 zcmchdRZtwjx8?_TcXtR*a0oKEyAwPR+!@>n?(Xgb1PBZeTtk51?rwv-1-SX&y}Pgb zvb7K0=c}&jm##W>zH@#Zt*$DEfl7)B001x)8M~Ys7dIFTb~LDR7eUyhlM|(&mjM7&qnxS%}!3& z2)T#S01K|SGU2&GsgWB3^^sc`V)6I+hSER8>~-R+cRT%>d9|a%mAobwjemu{pnOkhKf-_rB1caA0Q^)EX z;$)1w`Ej<^XOdec4F=}nM}u)BfXiNybfti2X2YC`HGO`Y7g%Hj1~jyV{K3LxHE@)+V0Q@bkIswe>Ca2yo!lCso-ysH7!vl;lkb za=Wk;1tUvJ3n=5v%lI^sfN67|qlc5Kk0dI7G7O3?q37Mu0qkG*x|s(`xv63Y903)y zF$Mzg;0-EjvCb>nhsySr zGA02wWzqFR5uQgik{IWhg@p;{6>8g*X$d$p(gDHhlnM6ErztkZOw z49p*)whdn_r<{LUiKOl2D#EK%jkpMWJge_=!^Z%vn1p;eS{=}g08Zb|wSzALxk)IE zB*aNEBSB_2Uk$aCRCtfd!8Hd9t$5fvaL=rh8&9{pyWnSUEiYGgD zv3zI>zs!rX$Qa$b-6yHdanX!EU;3&ftTPl@AFh3k=)(b&vgq_7aDKB5@jjb6LY32< zb8Lr8Q8$DKp%1Fy&xZc~Jy$Du$2*g@M^IR07Tb(Ifue{(IKHIs)c4~kTR7uSTBtzF zuPs0nHhZ5PPeU8%i9O` zzL~;HVkY$Rnnk!Du)j+YWHAY`llta>i$hU^Cz(`a zF*mcpn}IQM?IyTyD>qFyOqUi+8>|Hr0Z~?x3owux+ejp_TtDaEedn@aQ_|g(tR`4u* z$O$*Ab;~PD&<97ZBapl8d!z3I{h3QReQJb=I5vrL0a_9blm6vnpeNe(YWH4#Ub_~5p5F3#0*FGu#TvVjSWhb9e%*k%hR z(9J>{q~Au>c{PKz`Fy#zQAa9*n`hFdYM#Nre=^qynme%mtxT9-L0w5tJUDx^hCZ;? z1%2j873jGcZJ^!~{E;T-D*f2CXtLn7(et#mF_`%wJof&hdqWuoT8GGWjr3|>ZnM8V zgTRlhtj5QTp+Acr9xR}`;2ob^NuOJ$`hcq>Wa7+m1iPRrjezrWn_ba+rY_(5s4U&8 z=%-@3h!LB(=A8>$I*M5P3yfly?T(4A)#&W7A8nfI{&TCDvQ7YHfEXT0>^Ei=L&ga$ z*PM~4n|ag`OG#ED#R|S=I46nNxok}Q+!7q>Dbb61YD!tLV>VaruKRv>AiDYs|6a1@ z{>Oa;nnS21;`A3efq>!CwN@4m*$Ee;=Ya`Q>x-+_-%GlD+z$2-DM)-tB z=MDY3eeCYQVDy|^$Mr4vlvly5R+~YQjQKR-&XT7tSJ=3?1k_0e1*VF$gj!O+z|1_Z z@W`}yHyv;k6@1kb(9g@Gt`7T^_`n|-4Uan_Ns1p9-{gQ^^AuNZn=I51J6>t{#zRX6h^svWeO1PPO=9 z4ojPCL%A@;o*AN^X*fZLBt=xH0E)bYPTf2!h8S7c0^UVK4dp2^4DxgU)_{bMzOQ0p zTPQLe3oAwH3o(csHnbuqFFV5V&(1=A-uv1D)JYW5#5Hf^-qXguViV`|F(_!02njP0 z;Fq(TUz|WPI8y8dgQ&RA`F2tHzF)T);Qxcl%vSFn|KQ}MV$s7STSRh%KU~`UG^@u2 zKIpkC^KD-s5&KtpO%kHSmeyflp!4nSgP4C}IjP_v3#qUZ43MDLM`Fq54(R1^Lc77TD z936{HSnw8G;FTq`%aJryr*^I!c)u$CJp6t%g+(&>1xH;B=jp^t?C)lAXOIUwKD;K3 z?$#(m?k+deZ0K}z;?7(O06TQ?S%B%`YBA`+`F&(et&SAc0JB4iqC8k$mqjNOi!aE{ zKONH)8~(w&@Y%Icl*&WsL2R;HrbJ>J*eK$CpNjP0`o|Bc{6pAI5_!ZW2R;@i)(~At z5x!zsyvjxf`Yr<@b&Wv?KKmY{;l6&-L;d^t-t0%Y_rE9?li5oY&_4yfe6iSYZADOI z&LO!uo{0&$w4j|{1c^QiT@eR@5=<9Od&vGgM!A31rHt5pj z)gXAa`27MRIlOG|vB3 z%%XXmBq1zqzy|n0H*x>aU!l6=y*HvrUXK9gjP+vG&vlhmVDoym_tTdk zGn9&56~-(vw4mo-dYD~O0Y}Q;y~8VwpPs1#t9>s99vE9$mn=h%d;IRw!ve26J*%=9 z3F2(<-?f_(SV4+O3YLU7=-|Qv0mpeFJk&N{k{XILuj`t4fE^M14j=OALl$( zT>)0SiqkYS-{z1dqNGE@EP$|OxSA-N%n-1kwG?EaEY_h_2;&o;X)J@khARyCCaK-- zKFmlk4=wT&Z7XY5puUq{1L;EJ!(pugs*_bJzPyU&5qC}*H9*`T6<=7ni(Ahr1-Arm zduGUEY{YZ(Te z!*%Mv>J{BSl`~RT&X;QcnP+?Ys(kUN)WDdDc>%=5b-IXpx1)vsb_3Z^ue_5X6A zzDiCY$M=3%7;xEoL1jWABA_ejR^%H)epoE^`#{-yvgiFVjH>6gueJLM3r)ar{`-T> z=JWjcJB?twr_MUI-J!kJuwfP7lCb@7E6=aC7z%+i-y*Qa&_~y&K2eq+(97oES)zmr zBjc~vWTxxvMn^#b6+z(yqSVsl2~4RSR5$5eCq;%C*pjyh3+oY@uFe1t5JNqwZW=rSGhf+VLlGCyA z<|b2w*9$=ro7bza|U+8C}gaWwnb!{z>2Y+KXH)j(3%qowrIh!2#o>QO1ONf3?b z0NNZG8@0^(-}5pbVcMz`MiMuOLlYh0FU!?U-!-TVX4@=r;Z&dDDT2;O^vpVEc8 z*t-Hyq*s5TgBVfptyQV?0RAXa+;nTqXa@8+b<;f#dn1Mz|G5g354VGS)z>h0JW9rm z zic@}jiZkPXTR`Xa%jMLbpeQTiE}Xtjs0mKNkt0C+d-Y^cP-nLAJS=f%-6>(OyrRs` z`C+~=4-w0m1c5Wg?FGaLeelU@hTo%>nILNB$5%~~6YNAuRR?RF==aSKM;SwhL6o(4 z-b4&)XR_RB9G}&txJkHw-al_~_Y++P&%^*}b?IU2`{&Zu`7db)(GAi!>+DkRXgFE@ zAEr%RXFOJv#bgWx3#@wTlHymdY&I(`Y5InmXFZzj7;V3otqZ;sC?KFQ)d+)V5K>pF z9x^_QzG}*8!zy>CZDB{{6LcSVL#$Cn0MeJc^X{S7Uw}&Xdu3=Y4O65%TCQ z`2Y<^{$`bf#+zxuC*CubYOlD%h_~RW2m=a{N5sSR*g%2Rrgm!oIna_)WeU}&t#EAF zPTT#qhP4m6qf=8e2g19}T7mM%dOMUAnz%>_R&jg21L6Dl2?e+XgIP5R%yFZVG#RlL zE%G6a>&pn{t`hcd1E1o+BR(FmI|QiJP*NkAdFyg9d%smo%Gc34$;mm1+pYPGA2C1L z-fp*EHw%n}+PEp zdlPui_(WFs{x-hO=#{9xn1IR(5G3qjrl%j5-Sk6m(j~8iOa~NO;fN{2!b06S8~z$%-!jJ zQkoixN81}pkHQSU%FFBs<507f=?YKtsr9b{c2$TE*FoiIU(h2o^B#qvt-FP`UF6mT z$NXiapQCq91i{m0XbY?SG zfUdLMLy{6JDRM`$+p;~Yx)Etu^;MMeaGsBo-~Tf#@ys20)g0y~*49vN{Q{0h?9h;3 zc@$LI%#t&0SB}gdf|B7PW3r@~gQwyUvF4_!rE2R^M(KQ4my8q)<_lkrOh-BHtLNaF zM~IvMa{q|8$8I$|sHOY<#n@(QMje>|&72_m*&=w$CB3fpKv+{*DUj^t!S%v!vUGS) zLIbv>>CR3i+{xnUq`Fgp`?sJY3atw4;4vwri96FXxifwBG34g`j>f|G$ytmL*?svUbh^p| zl_nVOaEvk)pQHT_9RkOR3$TxZV#%YS+g7@p6UkN~2?kdsl>v-ZF~kf?f84KwM65Xb zM%n$<>=9oV%|)LZLI^O}O|5zTCNvz-)U^UIjjxW0d6mgU8D9_4i<4CeR|(iWxCJ?9 ztUYicjLgNt#A$kn;Z?%QWS+8obYQHiP7J7&S(6nap}5gG$=MH03zS9X7{W6}U3LIw zr?{lZsVyC3{n!I0zCHS3m~;XAb@Kh@IOzFgGtl3^SGXala6U<_h0%o|kH(MorTkt~ zJb+&OF_H5R|?<64Jj*fQg3 zhM^UEP@rm?^L7l2poR3?*1_3xr9k&pe7)&id3ot5@ng&H&pqu=J&b&QGG(Nj2)z%_ZCTYZaL53tTaB<sB?+V;n! z@<37^=8u~?mFsSSPd8dB$|>>PkM;NY$RL-k$m&u3Quo0q4c7RWl7w>*1Gg+8HkPjG zJnxmz`X}q~nEcu6(qm5vY;x2TT)AB1c_L_;C!WcHl(4yV)$A>oW-hg!TnE=E6$Lyf zZoy4q(r&2PJJ7SEVwgfL$WVk!hXB6O*VQ?_uDASAz6xMSWRorx!!n34imhLZ>d-=; z%zLSbMxHgQ6JK?+SQo)+$3SQ|bH7`MJJs|F(KzOMg#@bHVP4)-7Q^5;!?cEw5Y3)k z?-KB{u}JG842>HjQBG>!AgB_y$c$ZHvaA)3f-57qG;R79zh(jJ?9S^ws@vGn zSGY%fCvzDX{A-cP6s2hMi3X`T1uPr8-$E{~l-~x!v?S~@!4s+%DNQ-O`Cb7dkdOQ| zXar%_XafefdifxTdD0J0TagySLq_33P25RiItBCYWv$4Yd%Y%m!MXETM{$>p$_~p| zYwIehh2vjlHOeur-T01ouiEb^;s2(A0o8P1i%}14M%(fbJ8f*{&1n>dLb)M&pgpI` z?u5(o;Z#()LDv}+rbMS<|AhX~qAA?zEG>_8aHs4{F~L>ulm*1ipP-x{m+*UTlTCb%R@l_?2#1G|kK}1wBwuE~!}$lhTSK-SZg=%b*9j`yNWba3{RBr=At2K% zuSMSdwVW9)jCFOXB6*jDljLmXR#KtQh9}RALAho9={gCt*&L|E`GEzd3AfN!MlthW?$1#JRnVewbwFy{gx>tmqMgglyW*rTEXTrm`JfHGP z10bR(syn+w-|ndyaFg9IQ9{`{Cir(7gf%7E^TfW=^cOa5xoq5(6VXDq!nh!4#!-?X zy_!AEp89mfdRDU8f{iYmZ9l^iyndqLR3q@vFCSv(f#s{HBjIx&b^UY54fHr5Y^vm$ zQK~a%B3@57$$QbUlWG*R>xD0RGStbMeKj#95p!c0EU>b~-oYihAM!8+qiX2tqf;W3 z-GUEwsb_^&7ezd2ES|d$gj=j?Xj}d=$3@Gf{D|TBAoA}e(Lq&b9aajQm}#iUaM}k3 zt4=we_p+C#Z|rr(@;J9nOX>v;G7O9a5s4!$AByOd&qO@xzoS*M!ASy4bMs5bD_Y%u z@hYcsWVAf1sri)l!HclMRaSMx`Xe5M*6_?4id&Br&XTI)YqV=VS7M62bbO{N)2?3- z%P;q>OWU|>)8-Tmkb%(4ZjW4@Rhp7uA&CV)3x)`%G)`SEYi^PSu(*t;?Ds{Z7Gy&~ z%&AqnN*557yHG{cD{r^ILtUlZ7q3F=nAF!&mqji?Z?Bx6Lyx^vQ7TN6sQe$&rPOHg z21#Kew+fh^H|#lQoyFQj-(IWO?3au}k83!|4GQ!)8KOFVgp~5zp6wLD-fO@)i2Zjj zre)Zn$5(?SeHVDlR+oP`vyw@F!%zXys<%D+nRT#Jd}xv5KJ9l`NrIcIKidm(2kx$QBSR(E2>RleDqjcEE`d(}uV@WeBeQ6URaGwiRd z{x;e9tliI~g0cxdhIubczufQNW(dSPwb^!Y(o0$;@E8>4=Tkg)h_di+pit4g|Hm_a zjvq!0aoOMZq<0!1fj&5{hDr5a9|pN!XF9J71|s+7 z+K+qizQr!~Tn}%GqyhGFi-twN^$4Rl_qdz`cn;wg??F&1BfH(iC^Je;t>izatYxf&O>aav6Chmwb zrt|eAw+DIf1xZpSZ+-3zA0=nKgTo1V@vdqDi7Ey#K&sG8im~x}xoYaSk4(}M7@Hhe zT?!`eLQ1Rx>1@7Qa*MQu2J(_)p7d5n6se$G)MM=P>V}mZ)r6l8%=(a&#r{TE7e~n18SO$gq`1jsNImJ5yI&xQaf3htt?N8s0)`s{lDl#8gBYgt#Of`h+*DH zG@+^%7%fx~G>!>n{NMD!<~h9bltZ7*FF2jFt9K`M7MyFB#&43ugn+7KrWx!O$mjVV z2*tCEtDG~+?34mYV_u17mPBeoHeCc)mH$HzWd@OXc>mDD=cN;OQC41LyYQXgHpMcA zLh|j*wLX5y?iSwd-R2G_gAP4uLh*dgoZOkzsg1i!*JfV7?rF8GbCU#I{KtZd;F;}zOIStibwmS=Nu>^s3AV)`CjQ2m8x zpap-tN>Rt9Pm#nW-(=oF&=J1KZkEC!m2QmF3w#H>`cGJs%WifmEmDN479DCx`uyG}tRgVONAzCXf`4VU(e? zexY}P2BE6F{9zg+G{W+&Q1xwa$dFGe1CA-uto@;UQrYE4L#de9TuHzUmp5X`mz)No zZcx52XXvZ_U2peJDPZ!n5gpV)Awer<9CPmQneFM9RJVk<@mQ~)?V$HH103NR*iZ)z z#IOu3tY`X? z&*i_G)gd_%dce~g0J+$NQe7z>@5Uik0#*!m!V`?|FF5`kJTbfwt zyMYn85p!;&n6vDDi?rpu(Glc-hFxOjQ8ijUXgqP1Il23yI4OZ|S28jBvCXseQu+7+ zMO5zy@w3xKZSnK$Xq|Rs_)8xd=j&x1mfHC`xR0*yIQ*oApjUI9Nmtwuq;NOqKL5~j zb3K(B^+R@Zr}dNcG03aES7oyR?J^y`F|-MgJ1xF_jg(f4J_J9zx2wW+lefQ9K2!Pn&YK37jm$zv_&fxx6&!~jrzQ6*!@bkB zajVr2%>N;kyyo7Al_ia;amV*n9diF9%}t-6r&koC+mo}J{X-nsulc*qXMbh6*E*XX zDGl8jbV=SoOLH#!0lj%E@ST|Nq79lpg#z@@zM-%xCa)^>bP0l+XkK$yftc literal 0 HcmV?d00001 diff --git a/frontend/public/icons/tts/kyutai-labs.png b/frontend/public/icons/tts/kyutai-labs.png new file mode 100644 index 0000000000000000000000000000000000000000..e5b6156146cf28afe8eb7e5e2c4401144e6d7632 GIT binary patch literal 6584 zcmeHMi8qw*+aHyP6taYnN!bQ9%1+9Nn2>#^42mIp_NAhct@#ScmI`4QMvV+v!&oQr z#n_jSeP_t+gfILy;)Y2;#^Q)A+GIA9w$>IRY)KwoI7H1|3?=KXJEfn$33$~X-w=-%>7l&uwDsB__< zGE;v9d)ngi{^4*Xx);bi7CDJHgN=Uy;?_DSJKQ}EWyKu<*KaQMZ*oO|dbwM_f`eSZ zpiO$0hb7@%!qxA{iW&%Zs?JB(k9aCEciqdmMJ(35Mb!Bif)N}4?9OwiQ%(Bz)iUr? z?1BjS{hHN=9zVKqa-y7y`rO8~D;mt~)Yagi@hWbQMCr0OKX&fy8RW4u_lO5LvR zu)amw&!69~>vb?=c%TB_PWDW4l;Y9Hh-QX&(d1}LTrSoL=D>qxEAPCSdWI0k*meu0 zSyXsqhVX>C)>_3|avMAdg$h_1__l_`XVwH{-eJ_BV$h1F*Lq_YlHzU5=2C|a_YP?CjWiXYp8LR zgIbDJo}5r|dF{xm_%Z8CA16B*A=@gnI^*L_(74P{c|g4c)oNzYKsCMx$5wb_{ORJO@LiF13yn7Zo-+|~ zh0%Mn`0dz}TQ?k^uVq^AERV?1?Cr~=4{bG0N>N|FB%N}06?dsF)p*TLy$p^b9OqX> zHnd?n3Ko5@^xlp?HlI^zk`Z$OT+QhekrfrF0<1b02$5nmp{1*`=OI5Z=;+>Hvjrt9Hc`V^;CDY z;9Fwg0mM)f{04U(2IG1SAu}R`NYO8Ge&=SFn-;)+ynaqT{?9nv-&K6)d6iZ{H>3kh zcX7w%5w&JUEo7u1pE#)F1wn*_od=O`&lBn|&|dPUazh3&=+)k@&fG|OyjoNg_p+QCYa<4I8LbT4FoCuYtlrM=X+{0m zgz;QfR(_+Ki-BR;B4n#qKat`7{vrQ3ptVTUthEIf<;bbBm=8vUr&$Y#BDj1vVT6m>h-&v#3fG_M>Dfq z3f#_Ie^cWbzf@MAte-CpL#t=!mloYbxr_R>xy_?LtgM~<z8q3b({SL(?%^p{VW;U2gc{;=izEm zAT|PrWh@SWh2Q)NB6_dF@zIO8qu15wi{(`Gw!j!IdM-v)G@FudB*9`Wzujtu(p#x7 zJRa$^*lXBD-^F~38y8lQy^Lb8Rsv@^Uhqx)eW^&ZZXZuIP0-Tcj$>$aG;F4Wucw^( zs2RcR{|XuA__WE-mKBc?6cWlSHNxPyT_1aJ2TFSVO{s_hOAGyTpwi7=4hjUP&YphK zIM_g}s7Opmc$1|Sz;3zmAt691VPh@%3fAjw=iqY`jY}$l{6-8QlL#K75$IPzXut*(x=V}7)EQxOhz)M2b9^!on9={uCk$h5SygM))JV){rV@^zLTGlym4 z`~gu5l>wkO@l>r0oXdpjuLV-Iz~n_@X%NlEVPNl4TDY9I(&O_3{#vl8Ee7nhg2J2{ z=OG?!PF^$yw}h^$^K^EOZrF<;zxgJC|5aLCnlOce^RX;Ey_P4&ZfnZ6MX_Jv<(q%n zobXa>d-Q8C;99pn=EIeSh6Wy%sF@jO$_COvbl;+tzY#7EU4)?}li!sdoF_P&TLj_+hY(|QWewoiFpi+#=Utg=J} z4!=*~E`vF+-M_e?V7Gtj>yIrfE}n~szDHYla`lzQ*fD>?(vij=X)-Ks?AR^SPNMW@ z>G;1q7&to|%qjOobcn@T*-PVICzMb#==;SKl>sVic9};j`Uw~7pT+BEsXh$ixvP!( z-5C5ygeS43oQgAz5)>4iMy9f1aN*#Tx$$vj3Vo8fE~hwSo`F;ZAk>y7qsMNU%`D>> zcu*&Q^ypD8o>&orx>sU_9@|#-q3>N+ag$~3>wDc8Yo+qHj+BnA_Z`*-`5$ryXOp0M)pqiM&5#*4j(0?lt`Q9iYTg z<-U`r=SzDZY78Cqd$ohQ5{pB*w1{1QF#qy3c4@KJKpv|T7Z*o^XHF-}bH2`E32P z{>`R^jCFWPadBtX(lIqPH7NBRa6V}V(mebXx z-(&2l>t?|spolPw1&CE)kPW!T$;Vb94f!T3sxfRJ0!onlAha%N$CvJG9C}x}8%<5bsQtVMKh-uxDlx5fdeJiV{FESEjTOg_<>qmFXfyUn3ngCD z%x^+^Tek_Dp>J2zx`VC029pfd&cPD(q*FwSi>1J%&`|t!7PHwi@(W9WB5K%%aPF#Egu<;| z#h;nZhUU3MPIFjZxpoI`zm_L! z1kuRvrtEoS*5~C9sz<*0+5VYIHv3SRW4C&)n5>Aw5sC8tT~-*f8DhTaQ9{5CguQ+C zMcL_bV+1h-=qxbOR#z9%SuuoQ$4U#gTUcN+%ZVZcr$Qj7-LaJ%GuslQ@1sOZ63bYy z(mgXJsaisS>3rV@Br~m_$Ca1MrFVJ^f9c*tXrS7SDz2N>9?EJ(ljK5fqs}>ig5I_v z`4*5!gPVwV3FH(_um(H5XOjJXJ4QdN~2MNW-3-?u`czvV0fW&Vd~CGe$^+ zC>wip`W=JWGd9AG>{#1Os~oY1Kso4&8D#A*%DxHPF3;x6?9%+A-pmE9ViB;a=+{xb zZCwqQ@7r>W;j8D_4N*_l?JOFYZchdO1>1dKNK((k;}chot4}#SGYvSfswK}9FI0us zY1-+r_O&yEVjme9iT!P#)QXaE=~hd5|NebvXJ?)ZhX@X3Ug#07J9BO+%8g4NxxTy4 z872KY(y=9BSMj}*XL%$Av#9$=YbdAl@-e0SnwzuSrU021GWzbY_VK6-K<67Ma|DP^ z(@vzPr>DFj2=K7=*7tkLnKOH5!`Z3aSgU65Z&JvV11=SV9K#2X9+_u#O9J6_lNL=5md*FGwzs#Fd5J(v4Edbw@XAO?IMri}IR>q&{$I3{zdATHM9EMxZa(er z??1TV)#or7L&jr~#&*Ld?EFr=+daY+F0<&zkHPLgxIoYYRT=2~iw@Ns zQ9Y2?u0G=K?0hMy^{Rq`xzWIn7AnxA!z4dF4_EHMh2n}Ib5|b+FHBr+Fskw1Q-Q8@ z^Wfubsk?>W;@&k78=2+gl23EtSbNk7DXP|q8d~QD z(~bVIFZ;yZVaOPI)Mvr-&MmO%p7G4~W!yUz0Q zX3k69+)aH~KYxEKzmpw)BY#EmJxvtvCE#_w%>ipxZ!j1o9eQ=3>M?$l<}|){)(heP zhy3ez9)Qe4sB;lfzP`PMDkR;71z!7*?oC+3Q?kr>tSrFPCJFrc_2SZt#F1{`{Axn) zAO3R`Yg?CUx5tY0`fDE1?AH_&==%km>0|qOrbEsu%F3q%l~F-FA~#U8MXnzcbBQr0 zO@HyT^D(J!gflGwlqqv;S*%pgBauk8wb|)xz)$aH{o*8Lq~~*%>6+OJH+CL3QZ%E# z{A&Z-RHycDI5J_S)2WW1-TKbb$D$TU3Le9yvzHhs8~cX=oNydUxB<@N%*_(@mm~wu z3nk8aD+zrmK86h)KPZm21{eXl3|KGr8$`jQKDMcu<;mv>r~i2REg|W4i&xUwly zSK>a(y|lDsmRULJv0@q%gvw0_NG7)EvN49F%mxzGE~XRxMTU}w+noL z>5biOH3+*O#BzFK?{zZ97p1T`Dg*C&bTE;RmX0_LGTv)<-KnQM0JA`}yr5De^#MPUxF@$Xg(N%?Sej;dfByvUHzO&?+9* z@XMF$9zSK&QXpI#I;K9bF{!X`F3yDx)~endiBRz1`0T_t|rk6 z06Fy-m)`ZgV7(iE8MvXxN)ZrE*l_pOWUQ=^q9+grxHBVUGmpJVKRb;GiL6;mnwNZ1 zs|8I@sr;P)wT({Jw7cBiphB8lQ|xc)0CV+EO-t^sB`Vvw@Uc$8(~#EILmeG)K!_R! zskro)*a8z9`|FnfWN+p5uKk+o>`eZ)JDJ5FyCYuJ*WVb@e*)YoWj3kx!(5Do2nm}B zx^5$~*I_FzyM~aW>fM?2y!_G#w0wK#aclx`u%i_1Hd?_@e%D$K#H=!8mk^a-OjEb! zoFnUJb#@l7&*PNv0kKZ7MaQA0(Tg#NcX&Ky6LvDz7j?LEy&6zLX<^r>x}V2q+v6dI zV6&?A%0_JBP^BTt(2tl(@3IBJ7#~Dv;Lqx|KID3y;|@R=J3#0Wqu@7Wsn9g OAh)&j5hXVrp8OB;J1y-1 literal 0 HcmV?d00001 diff --git a/frontend/public/icons/tts/yl4579.png b/frontend/public/icons/tts/yl4579.png new file mode 100644 index 0000000000000000000000000000000000000000..bd34086a8f8eb3025ec780b7cad00b79d388e084 GIT binary patch literal 1528 zcmeAS@N?(olHy`uVBq!ia0y~yU|a&i985rwk9xc3-}c;it|=Ran?FBzA?K~%=70y&zaPl$0Y0-?!PPl%FrNrgwI33TEX#61ItEc zA=+|ogz@j#@Yf>TzheLM$NQ=dS^w;*__&GIw$jUYqxR4*CDA{uw_Roa5&!LRj4h+x zq0__Jqrf6yN3Y?tm|OkD=M*1A&FlZ&qJ6tXI_AWj^>@w2 o@WVo%PNgEylMnS782FVdQ&MBb@0Kd=bga7~l literal 0 HcmV?d00001 diff --git a/frontend/src/components/charts/TtsVendorList.tsx b/frontend/src/components/charts/TtsVendorList.tsx index 867d1008..d1b3c3fc 100644 --- a/frontend/src/components/charts/TtsVendorList.tsx +++ b/frontend/src/components/charts/TtsVendorList.tsx @@ -3,7 +3,7 @@ "use client"; -// REVISION: tts-vendor-list-v6-two-rows +// REVISION: tts-vendor-list-v7-new-vendors // The organisations whose models are in the table, as linked chips. // // This was a scrolling marquee, matching the splash page. It is now static @@ -22,7 +22,7 @@ import * as React from "react"; -const MODULE_REVISION = "tts-vendor-list-v6-two-rows"; +const MODULE_REVISION = "tts-vendor-list-v7-new-vendors"; if (typeof window !== "undefined") { console.log(`[tts-vendor-list] REVISION: ${MODULE_REVISION} loaded at ${new Date().toISOString()}`); } @@ -54,6 +54,13 @@ const VENDORS: Vendor[] = [ { name: "SWivid", icon: "SWivid.jpg", url: "https://github.com/SWivid/F5-TTS" }, { name: "Supertone", icon: "supertone-inc.png", url: "https://github.com/supertone-inc/supertonic" }, { name: "Canopy Labs", icon: "canopyai.png", url: "https://github.com/canopyai/Orpheus-TTS" }, + { name: "Kyutai", icon: "kyutai-labs.png", url: "https://github.com/kyutai-labs/pocket-tts" }, + { name: "OpenBMB", icon: "OpenBMB.png", url: "https://github.com/OpenBMB/VoxCPM" }, + { name: "IndexTeam", icon: "index-tts.png", url: "https://github.com/index-tts/index-tts" }, + // StyleTTS2 is one author rather than an organisation, and was missing from + // this list long before the three above it - the chips were never checked + // against the table's own rows until now. + { name: "yl4579", icon: "yl4579.png", url: "https://github.com/yl4579/StyleTTS2" }, ].sort((a, b) => a.name.localeCompare(b.name, "en")); const ICON_BASE = "/icons/tts/"; From 4584259806d669c468b7a4231b7164021ae71cc4 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Sun, 23 Aug 2026 12:38:19 +0100 Subject: [PATCH 20/34] Size the vendor chips for three rows instead of squeezing them into two Twenty-five chips do not fit two rows at any readable size. The only sizing that managed it was 0.70rem text with 5px padding, which shrinks all twenty-five to tidy up one row's wrapping - the wall this was always going to hit, and the previous pass said so: "the next vendor after that needs another pass rather than another trim." So the constraint goes and the space gets used. The chips are bigger than they were before either trim - 0.85rem on 9px by 14px padding, 38px tall with 18px logos, against 0.78rem and 32px - and the twenty-five wrap 8 + 9 + 8 with 167px spare on the last row, where the trimmed size left 11 + 11 + 3 and 840px of that row empty. A third row with three chips in it reads as a mistake; a third row with eight reads as a list. Measured in-page across candidate sizings rather than guessed, which is also how the ceiling was found: 0.88rem tips to four rows. Worth recording that shrinking the logos buys nothing, because the width is in the labels, not the images - tested, not assumed. Narrower viewports wrap further as they should: four rows at 1024, five at 768. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- .../src/components/charts/TtsVendorList.tsx | 30 +++++++++++-------- 1 file changed, 17 insertions(+), 13 deletions(-) diff --git a/frontend/src/components/charts/TtsVendorList.tsx b/frontend/src/components/charts/TtsVendorList.tsx index d1b3c3fc..5ba33bb3 100644 --- a/frontend/src/components/charts/TtsVendorList.tsx +++ b/frontend/src/components/charts/TtsVendorList.tsx @@ -3,7 +3,7 @@ "use client"; -// REVISION: tts-vendor-list-v7-new-vendors +// REVISION: tts-vendor-list-v8-three-rows // The organisations whose models are in the table, as linked chips. // // This was a scrolling marquee, matching the splash page. It is now static @@ -22,7 +22,7 @@ import * as React from "react"; -const MODULE_REVISION = "tts-vendor-list-v7-new-vendors"; +const MODULE_REVISION = "tts-vendor-list-v8-three-rows"; if (typeof window !== "undefined") { console.log(`[tts-vendor-list] REVISION: ${MODULE_REVISION} loaded at ${new Date().toISOString()}`); } @@ -71,27 +71,31 @@ const ICON_BASE = "/icons/tts/"; // inflated these chips from 36px tall to 92px. Both need beating on specificity. const CSS = ` .tts-vendors-figure ul.tts-vendors { - display: flex; flex-wrap: wrap; gap: 5px; + display: flex; flex-wrap: wrap; gap: 8px; margin: 0; padding: 0; list-style: none; } .tts-vendors-figure ul.tts-vendors > li { margin: 0; padding: 0; list-style: none; } .tts-vendors a.tts-vendor-chip { - display: inline-flex; align-items: center; gap: 7px; - /* Horizontal padding only, twice now: each new vendor costs a chip, and the - prose column this sits in is a fixed 1104px, so the row count is set purely - by chip width. Trimming the sides keeps the height at 32px and the logos at - 16px - shrinking either would be visible, where two pixels of padding is - not. Twenty-one chips currently fit 11 + 10 with about 30px to spare, so - the next vendor after that needs another pass rather than another trim. */ - padding: 7px 9px; border: 1px solid var(--border, #2a4570); border-radius: 40px; - font-size: 0.78rem; font-weight: 500; line-height: 1; + display: inline-flex; align-items: center; gap: 9px; + /* Sized for three rows, not squeezed into two. The prose column is a fixed + 1104px, so row count is set purely by chip width, and this was trimmed twice + to hold twenty-one chips at 11 + 10. Twenty-five will not fit two rows at any + readable size - the only sizing that managed it was 0.70rem with 5px padding, + which shrinks every chip to tidy one row's wrapping. + So the constraint is dropped and the space used instead: at this size the + twenty-five wrap 8 + 9 + 8 with 167px spare on the last row, where the + trimmed size left 11 + 11 + 3 and 840px of it empty. Going further tips to + four rows at 0.88rem. Measured in-page rather than guessed; note the width + lives in the labels, so shrinking the logos buys no row at all. */ + padding: 9px 14px; border: 1px solid var(--border, #2a4570); border-radius: 40px; + font-size: 0.85rem; font-weight: 500; line-height: 1; color: var(--foreground, #e8edf5); text-decoration: none; white-space: nowrap; background: var(--background-elevated, #0d1526); } .tts-vendors a.tts-vendor-chip:hover, .tts-vendors a.tts-vendor-chip:focus-visible { border-color: #d95926; opacity: 1; } .tts-vendors a.tts-vendor-chip img { - width: 16px; height: 16px; margin: 0; display: inline-block; + width: 18px; height: 18px; margin: 0; display: inline-block; object-fit: contain; border-radius: 4px; flex-shrink: 0; } `; From af838f97852ecaf58c1f924b94311d13929cc7ac Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Mon, 24 Aug 2026 12:19:30 +0100 Subject: [PATCH 21/34] Import the six Magpie quantisations, and fix what they falsify Forty-six configurations now. Magpie is NVIDIA's 357M multilingual checkpoint at six precisions, and one card evidences all six, so the licence proof and model link point at nvidia/magpie_tts_multilingual_357m for each. NVIDIA already had a vendor chip. Two sentences in the article stopped being true, and one of them stopped being true in substance rather than in detail. The cutoff paragraph said "nothing sits awkwardly against it", and named OmniVoice at 1.51x and Chatterbox Q4 at 2.03x either side of the line. The boundary is now Magpie q5_k at 1.98x and Magpie q4_k at 2.01x: two builds of one checkpoint, three hundredths apart, which the table's own notes describe as indistinguishable. That is the definition of sitting awkwardly against it, so the paragraph now says what the boundary is, says the Real time only control is a rough divide rather than a verdict near it, and says outright that it used to be clean and is not any more. The passes rule said feed-forward engines get one. These six get three, on purpose: six builds differing only in precision need more samples than one run to say whether they differ at all, and the answer at three runs is that they do not -- 6.5-7.4% mean word error across the ladder while individual variants swing by up to 2.4% between runs. Worth stating rather than leaving as an unexplained 252 in a column the page does not show. Counts follow: thirty-one of forty-six lack a medium.en score, and nineteen rather than fifteen sit above two times real time. One upstream inconsistency, neither ours nor blocking: the card's prose says 364M where its own repo id and this table say 357M. Verified rendered -- 49 rows for 46 configurations plus the cutoff divider, 46 play buttons, none disabled, six Magpie rows, and all four prose changes live. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- controlplane/src/tts/handler.ts | 12 +- .../content/benchmarks/open-weight-tts.md | 20 +- .../public/benchmarks/og-open-weight-tts.png | Bin 82561 -> 83663 bytes frontend/public/benchmarks/tts/magpie-f16.mp3 | Bin 0 -> 17806 bytes frontend/public/benchmarks/tts/magpie-f32.mp3 | Bin 0 -> 17388 bytes .../public/benchmarks/tts/magpie-q4_k.mp3 | Bin 0 -> 17388 bytes .../public/benchmarks/tts/magpie-q5_k.mp3 | Bin 0 -> 18224 bytes .../public/benchmarks/tts/magpie-q6_k.mp3 | Bin 0 -> 15925 bytes .../public/benchmarks/tts/magpie-q8_0.mp3 | Bin 0 -> 17806 bytes frontend/scripts/import-tts-export.mjs | 20 + frontend/src/data/benchmarks-posts.json | 2 +- .../benchmarks/open-weight-tts/2026-08.json | 608 +++++++++++++++++- 12 files changed, 649 insertions(+), 13 deletions(-) create mode 100644 frontend/public/benchmarks/tts/magpie-f16.mp3 create mode 100644 frontend/public/benchmarks/tts/magpie-f32.mp3 create mode 100644 frontend/public/benchmarks/tts/magpie-q4_k.mp3 create mode 100644 frontend/public/benchmarks/tts/magpie-q5_k.mp3 create mode 100644 frontend/public/benchmarks/tts/magpie-q6_k.mp3 create mode 100644 frontend/public/benchmarks/tts/magpie-q8_0.mp3 diff --git a/controlplane/src/tts/handler.ts b/controlplane/src/tts/handler.ts index 8d8dbe30..60b71207 100644 --- a/controlplane/src/tts/handler.ts +++ b/controlplane/src/tts/handler.ts @@ -24,11 +24,13 @@ const CONFIGS = [ 'supertonic', 'piper', 'indextts', 'chatterbox-turbo', 'kittentts-micro', 'kittentts-nano-int8', 'qwen3-tts-vd', 'kittentts-mini', 'dots-tts', 'qwen3-tts', 'kittentts-nano', 'pocket-tts', 'chatterbox', 'f5-tts', - 'kokoro', 'orpheus-q8', 'voxcpm2-tts', 'xtts', 'orpheus-q4', - 'nt-2e-q8-metal', 'nt-2e-fp32-mps', 'chatterbox-q8', 'vibevoice', - 'nt-2e-q4-metal', 'cosyvoice3-rl', 'chatterbox-q4', 'styletts2', 'melotts', - 'omnivoice', 'fastpitch', 'cosyvoice3', 'bananamind-tts', 'tada-3b', - 'zonos', 'csm', 'tada-1b', 'speecht5', 'mms-tts', 'vibevoice-1.5b', 'bark', + 'magpie-q6_k', 'magpie-q5_k', 'magpie-f16', 'magpie-f32', 'kokoro', + 'magpie-q8_0', 'magpie-q4_k', 'orpheus-q8', 'voxcpm2-tts', 'xtts', + 'orpheus-q4', 'nt-2e-q8-metal', 'nt-2e-fp32-mps', 'chatterbox-q8', + 'vibevoice', 'nt-2e-q4-metal', 'cosyvoice3-rl', 'chatterbox-q4', + 'styletts2', 'melotts', 'omnivoice', 'fastpitch', 'cosyvoice3', + 'bananamind-tts', 'tada-3b', 'zonos', 'csm', 'tada-1b', 'speecht5', + 'mms-tts', 'vibevoice-1.5b', 'bark', ] as const; /** diff --git a/frontend/content/benchmarks/open-weight-tts.md b/frontend/content/benchmarks/open-weight-tts.md index 5b4ca09a..af03da21 100644 --- a/frontend/content/benchmarks/open-weight-tts.md +++ b/frontend/content/benchmarks/open-weight-tts.md @@ -48,7 +48,7 @@ face value. **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it compares directly and cannot be gamed by padding. - **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger - `medium.en` was run too, but it did not finish for twenty-five of the forty + `medium.en` was run too, but it did not finish for thirty-one of the forty-six configurations, so `base.en` is the one recogniser that covers the whole table. Being the weaker model it hallucinates words onto trailing silence, which understates good engines more than bad ones, so the true spread between engines is **wider** than this column @@ -72,15 +72,19 @@ face value. **Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family. -**Nothing is filtered out.** Every configuration compared is listed, including the fifteen +**Nothing is filtered out.** Every configuration compared is listed, including the nineteen that cannot keep up with their own speech, because "how far off is it" is a real question and a table that quietly omits the answer cannot be checked. The **Real time only** control above the table hides anything above 2x for readers who only care about what can be spoken live, and sorting by RTF or compute per phrase draws a line at the cutoff so the two groups are visible at once. -The line falls at 2x exactly. Nothing sits awkwardly against it: the slowest engine that -keeps up is OmniVoice at 1.51x, and the fastest that does not is Chatterbox Q4 at 2.03x. +The line falls at 2x exactly, and two Magpie quantisations now sit either side of it: q5_k +keeps up at 1.98x, q4_k does not at 2.01x. Three hundredths of a real-time factor separate +two builds of one checkpoint that this table reports as otherwise indistinguishable, so near +the boundary the **Real time only** control is a rough divide rather than a verdict. It used +to be a clean one, with OmniVoice at 1.51x on one side and Chatterbox Q4 at 2.03x on the +other, and it is worth saying that it no longer is. **Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and long-form categories. @@ -93,8 +97,12 @@ scored on **trimmed** audio: scoring untrimmed rewards models that pad with sile to 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has room to rise. -**Passes.** Autoregressive engines are averaged over two passes, feed-forward over one. -Differences under roughly two points are not resolvable at this sample size. +**Passes.** Autoregressive engines are averaged over two passes and feed-forward over one, +except the six Magpie quantisations, which are averaged over three. Six builds of a single +checkpoint differing only in precision need more samples than one run to say whether they +differ at all -- and at three runs they still do not, spanning 6.5-7.4% mean word error +while individual variants swing by up to 2.4% between runs. Differences under roughly two +points are not resolvable at this sample size. ### Limitations diff --git a/frontend/public/benchmarks/og-open-weight-tts.png b/frontend/public/benchmarks/og-open-weight-tts.png index cf3b38b352be92cba2288281296e69089388ffce..4bf27ee9b80b391346b58cef9fb3e697d69fc63b 100644 GIT binary patch literal 83663 zcmeEtRa9GT_b2Y|F2&ugIJCHz7K*z&#hoCfcyV_t?h-6OaV=ilJ-7y$y#MceXRW!L z%ek1fa&=B}a!#Hnd;e@lX{gC#qLHD&z`$TCDtyp_fkCK)fr0ZxfrB3DeRqlu1A_yj z_(59RJNq;T(FA{yym#YcW>W_>_9J1m@Gy?2I6AzwlVc!Zko+@78?#?;P|#?lT&jQ6UqRE#cL**!Or z3Kh=0b!V=I<&lh}tQGr+-ZlIMVT-#M#QclW zJsmcMNVnBX>geQ;uiK4J;Uq&5HR7j@`yX`D~vS$vK&*L*oS*-PRHI$T|MoY0&0 zVHAU?QAk%Ii-!rsFc#4E_yfg+%ezO6LD7Gzm60EU_aG@H)COIy(KJ)pPq%QojyQ)4 z?eG;S17?W;fH(^m>}#NZ;0TVZMhO)|NwUumKv!4SS&>on9`6HN1DmxS%X*UQDO||r z*`l$p)2KImzx+rd!F=IY{0EghG4L6-grwIcT7@}En2ZzSwaOkIoc|5jeo8e1`d_Pu zlaCr$TT59RpW+(kVx}Mdbe)_d9IDM)jxv?z%nouG%w8YU0{&Jp#0 z>xl=knbjFr&hPK;x3_nYcx#5qfSAs^G<9{gRSAB+|80NnFJ}KVa96D_y=A5R9tF_vrGCjA^m3L>RbWSdVIHC%&6iW^82M4)wjB=O+BROjiGLS+E7K17Up;UqD1?=cmMfNQSP6*#|d34j8x9AyC*xp*RJUJmM)$)KhdrapI>oU{ylKzn9aOL%ueNMCbrL}nm4m$kKxPkXlk2iN$ z#kTMH*vMb;R$N``yxd(CRj2A+OFyI7d{{R!(yHL+>IoJTNZ6uss$sD`+8cXMr17Iu zO#J%lD!N!p%k{(1An9p+Bfgsu?Q-HwJ>!(Bnxeou3mYYz5D&}p>QOL!Lg zYqT@IRmZiys;>NdGP;lsF*0&w%=oxlSQZ>IyY_X=c-0jmPeh9{YqQAS^!LZdCjug( z%&Z=ZJ$w+`u9L*~C*Q9(>2{A@%Y%PSolu-NKaLxZ~|FKIm`qvIx0tF2mEBv7;y zG+xE-gvpvzrgA)M?(BSic$kopp_)s^BrmU{bC8`aX6ssiYb`B)dv)WTxsD08cjgk* zA^E-@laiH%T}@h#`;ty$bc@>feDgC{ag;Vm%#zx%{Z+4&qQa8Ji^<3;bSwBd6fP^% z$;Pj%OYF}dK1bgaLN-d;DJtmBgPGRCpw_ZGUF~^uJg@^8w4DE?CfR%&?Q1{H?e6Mm z2e$9+9{a{YD@5Sa6W?7KdRoovImEc&YAR1W;B>e^8WXMn9pUKs6)>D)FGwZqziQwP z6MFS9(?1kP_LsXHmd_1)@<+7#Qj7b>-5SLG`U*477_z87ImvMOf`*2cm(@e!^xk@Z z_XvMT#?G`%rXa>Iz0UPy@;0j_$?bAhC~qDc!=&53Ia86(c=Jb8EE!TzpVSeLG9l3g)-5x0cgq``aFepHH_(uUBw}rolRcxrkoNj~jOT5MZ9a ztK_>M+I*qpb30B(oo*Aw>_F?ct#Ehi_rPR@=k@j6Vgr)2DNjeez$Z|aX{@XX_FQ7m zmzOarhfk)Pyr$#du^jco!v*c;xKR+kd7o>}xZ)qh}f8!sU@fO20;~?|Hw&nS6yDcs<~eYz}p;N%p+uFC1ns z4XNirj#`;KCQ1b-EM(CHo^ExBygN>;j^lEko~uRgm*C&G*o%AoNvARtZJfukD4jjv zlyU6(c8kOk>l$$?`K`*<`%oF(>i%pwM^WF`cRyCtD=p|YN!qGGDmR`&5GR+pNmF(q z8tV+c>G;iYjr5*7F&$J|Oa_Bg6r}TNW^FwL_PLz*iuq(Sy;SqH$95^|fL)$%Fap}Xgii{)bjA72}|&e!7&9>;`o zbE#Ip^TNyxzy|n%(pJ3AgP?_fHP0ahA*O;kW1!S(dErTat71>fipeMrLAI8s&bhX( zPEI4fmo0zBW*ZNWj_SikJNe~#spCF|ti~zuo5NzUt3v-KP5D?tNTO!_J@tnipv&HL zebNBN9vCO_7zx@hgS*9nBCav3CN^*214(v@x5Zx`NkefrKBt-lCFi*c_6^ewIjjxu$sF)EmM%HBzXW9!C$d(^QntW*1O6Z4Fx z%`_xE8x?KnYHp_6&Go?m!2E&~?df7AvL6fzCTf)8srt2*Dk63@h(HxbP8ub6>gBsF z7rRf^2@d|nHbZm+Ma9F#-WM%=8s3T?ii&n=5YSYPT@wW=|lPcN)!}<5GN57xp68miGw~mJiKhn&%HZZzP)U>I=A$H_Z@}0>-xP{F4OMC zICRU#QV0fJW~J2_Slh1k33b=U*X@0XtKO6 z9g_^qQsv09n^x{IxSRY=3y=8|7L%)-;$pe^d>Tj{G?D|c%cjKyoB#^o)*rzL_9MQl$Mre`O~KJIbIYhl7j<9~y(>0cHX z2YRiyS(brJ%_Fcw?lRUoeGVk95#7j7eN8yGs7|-`7rNZ#AI7rynmOBH4>fyZq z76H@3#1gN;jy?(Xz0o!k7Kl` zB%f$`qpBwwZf{$8wZ?vDs;!nH$TD``{dq=ow|3E2Tf?iu1Z=@GQ2*GDuI(~6Sqv_F zIG)As^&ekDB}Ms(>Q9qME_kkcx96pvcdg(NA)jEUE~)m0w%QKqz4E4=F6A`lvK*zH zZY^sENcx;(I_VEu4hKuTL2d$eaw^{**7LD9;c}>%#o%v|KDv+RN<>v+x*aW7UtjZ; z;jM@m>vh}CuxA5Nb4kIB!xWnWmO=$b4pdZw&_3p$LwR)xT2WpPe6j3)41P`U6BVj9 zd(O6mD_xAW58Q#`2O51j3>MLHMtG7dN$;^Hpw-O<$fSu;Aj4kh+@(4_AMxd7#Va=# zykKlPqCC~}a;!X&6)};=_p-tz$=GtycRfXlK3m)w4qN;vuD?KL13-!)7k8M=+>qko z{8tN*#QZfMeQix@17Oz(>xOL8B~Dj$!Bn**;K8p&p^N<%b?(HVHp_)}m4?6hs$r1= za21)Am<_?TR`bpqJT*E=ZNTxir<8y^q}!7&r{C&~=-lv~p(#B9$VNG6>a2AK zy|`ej`x;72tYE$Id8G4?A9w@Tz=rRgR_gvpk@G8T5$APYzv?S1d<_=0JlLNf{7Zx0 zSy7@)6iPF>lY$qrZqRUe9ltI(jnp1&{x)WL9%`j94u3E36{V|7-1zH*zl+98{6;|k zE{7Dw+tW4nxZwf_^a|mU+=u?~`G#=fNn*0BAMm)Ns(D9d?`QUpoh^$~mD@#K-BBal zZqGM*q~0vIzUKRbSDVOg6g}hPTNiVy?J=9A;d^&5by~P6x4X?s=!!?B!|RYtxOZ9J zv387qqWX}U$%MaoV0~2wh07uB-yfaamV^{wa{jhL!NAtTX;EwYEj0y~NrNHKRawPA z42X~Q03~o;T`KV#?jD{<*RWSpBs3<;oxq`TwNM$Yt4@G!Bk?=l z`pExkUCWG8GO#NKx2@65E#?}hiDp`MGrCCu?TW~s|H;&ieO&Efm~ftloynWlX^lg` z44L59wY-<}^MO}gKq&k`#mxXWyp$CspBV}e^`)l0L^Rg*r*H$9g>#P+#=Ogz*jk*w?DUZ0f@&&69pTz^ZTT~v`Yi1Vp;t>^Js)=3e*>>gCa^!n_# zcAMqZhM(H|@TbV0$t)s`c#?K+8(Le;vsYZWD&|}kh!p0j(}_>4r_u)uC2|WW)Qd}O zOT&l>Q^@quRf!rU+4R%i2>+CZR2BzheHnxp2e`uOaLjZ)jV%Tm*Y6J;UH^t|zx`&X zWrvZ%gD6rWaml2wi*vgeTEqA*`t@1e5=LW^K0{I;{F0O1P6w8YcbSyqWo9pLt*Kp% z1H$PEg*~s>?;oHGO3oi6-jS?@dr!fxNmKA?v&PqVd#+tb@&O~89(Yf^-z{C283GAU zm+vu{7^0%^qn=qL8-@kcUvA!e#}@q!;lRsO1;c6#Ij^+`-3JzfIC{0rO+A+5*XJ{2 z@>K+b;OV^Tcxu*B30TG2_UG z`1q)#C{8iEvVB&l=7DZx#qgVdHgb5`0t}m|>j^%F(YM3XB~?bD7Vwfi_Xmn`G-w;M z2p;KcpK@L6)2w&hL<)jJzAZ;L$|#qiqffa@muv>_Wqwgm@an-|T(T0XA=w#bl#wvU zt{atB|5kZ}CG@`Sc*|dbC;fJq5a`U|lW1?0DQS%)>VeRq?vw^Eb4Gs7X)a`J{M=XQ z!%!Hdrjl`=J))4MGJCP|L*PWnQCWG4Ao}-oPXiNP7fX|6v_r-p+>_iMaqR3+?8>R%1V&CQX!jtOWA>g149R=1bIgslJHZ z&D%6$7r(>QiA)(Gm(5|+hu-~?BuLu{&K)&Rl(uvC(`#!_5G0k23N$P2JoL?b`UT3K zLkKr_kI=0rjd|05!IH^r7ZJE%^JdY=$tpI2Ti$;qm5`6idv#u^S!q50B$lgDJ~n?v zh*XZ!)!}6|7255&I(I-CMal|C;8)XXI#3s4{Bl8l9k72Bb)nbpWzcAPJvaf}tdLep z=k}!+#>TXeZ}}Ta@Tc^%J6M%T&A9g%FCmw($G@M5@WLva)GsO>t_fV8zOm6#H+6@&I zrOW$Ujt%?~3*OR%pzB9ub2*6$FPcM_bA4JzGV|OA`}v+B^_=MMF`lkaegOx3xNa=1 z8wxfPvt2;P#a7a@s>3~9s{0X}jnKCtc_Ne%g9}%}bE$cyPwnsdvS7%(#E}^e=<81* zC^*^_n}h@JSRCw68a|Nmc~kuuVvEiVdJJb$dM$ix%m?G-Ow^nGM3ZWv7WIg@p)B>y;FjxCf<)lGhtrVP?O?S|YG7{@c>;XTF;adifr&7N3Kc+3 zl_vv6!#?Qgu3h$|R*_)zlObH^xqFp^)C!><-R)EzFhfOO>8_ZfHd#F>aA- z%v;rpMIF|bJ)x$(idm(K7HvS&$dfWFZ8MVYa+H4FCJ!A_2YwCWO?E3?;oW&^wnfU7){N#k49e+v4wHY>iFw$$zxFUjCG+jy z%hP$)eQ>@WlgP_Qb!F_Gav1|3_aS*aa!~TiUw+n8%P!CeXD2|H{^Nru0O=<1XAJVp zd7uI7m(fGu1Xi+rr06T$fC>>0Hv ze1WVb6S5noEk~y03>EF?`b*Rc#9}l?dYJwv)s^;H9VuQvXl-drolETJh;4cIK3Y%5 zNwPqLc?&HvJ`e5|rW=4$=Q1L?Aa91C2|0^Tm6eyd0fd1*H@J>*f?+4J1ZV|Faneht z0x~jO?J#!fnWy;5)C^vR?uYK2sKov2K(IkIdP5I}(m2n46LMMa7@)%P{_n+gai1r# zRdG=!q~W9w;(%NL+uqn6V(oFj!E>Yb!9YlTwI&bQBDg`m*}_i3pL zmg?-bRT2u~aoVJ{S#Ou4b@Cm^+bG8|<-+^K&VQG@y|$i7YqRfuR$1*w^M41=TBi!x z2Ijrpztrl~Y_R`S$~YyHXp>yze4QLJu45&FDZNkI@aE0N;1rF)w+=@)u)FKiDJZa_ zDDVP*?7Up(1;b(w{w zZ?PETt5j2M?rEgjwKmLJ_{^6x;9c!otu{roXSF^a-*dg&**V)Hw+PbuWUIt8h7#gE z3_z~vXL+xlIo^roPqfpb*b}ab!#T2R{+u?~P}(GFMlmx;Cbs0u4|1mGKPe~EC2qAi zH^L{YtGO|diP6z5H9WPc;jqBSYTwp#Ic~`RbS-$O$_m4*WhdMw;940I&m&Y!L5~Cg zg>AfD>BtQx;S5fvHak_E@Pe==D9gOA#(j2hxyq&tEPDJpfLYOq4Ym+YfOuqF?yeIO z!b8J`P}S?%VXo@F#L3%hxcJp+I8#nSpy}YxBe^OtvhNj$j=_rcVO~=U&o}NwP3&H` zl47Y5CcIyjJ z(ggRx=uM#4cK#0BjB2wE=_Vk(wfH7v(;mY?0xRt%NM2;^%>lR@m8r>T^<`}pHV+1R$#@Ch}zhJ``{qbU7lrN=lS}RR8(BXiC{^{7c)nFl}U)VH->u zJ~p7yX42R8t%6ugir_&P>ofcp92#&Z7SUl#Z1wB(gxL4EExC*Pg&`*H^wBn#^>MIV zV?4%Rt_yFH_%ql>W*rnO1&%cc+JkzbWp5+-khu94QcauZwIF|Tl>pY9m{mER=-=1A z6#D}%Lql$z)@>2pft6-kSS0hh%5e8D%GPT!1|S zRABiJWYq5W-Oa90ddElMl$M@8zV^-M_>q1d*2Si?=QN0#>#aq;Rc&+VaJt4LIInhu zZYY+Dh>*C|?KVlnkH)Wl>$Dnq^sNgy;>1N;!4$?>Wz0%duf+jvkCdL3n2CwWdT4<| zS@pZw0DRt`MklE^9!?#NxgzuJn1fGH-QVTxa`VX~jl>i;Dy&4EhjMSYl=;oTX(Nhz z-KVzcv42Hn4u6MLBShUo>SMDlcA*(Yn2c7(xYW}dWaxv9O^wN!J%3dWXt!6-hP!B035_qT;Li1nl)ToxP)Iy0+*&OuQz!A^R)M%JKu zo^>7m6A9Sr+o|kMwEXYn(J7dqSxij_g9>-)?Wu{eBn73g^8tAlq#^xlgAc|5F=ZLQ zC1Q{xMMMj+o1~|i##$qCZ?(0dC|#9T@#F0bAV~+meyQjt3=Vx*b8FyzIm# zn<6KNqTPH9b;uEh7u(CIf1sXjD=8rz;TC+C1ECQhMb9SW{HNjnV@DiiWEh0kDeBTS zl*N4N>r;wN=UtQ)s*EcTa6NbdK20-5(!C|%-vgHCtdkqv>a-rS0Lt(Fq541Nd zUFDQ0mqa&ONL=db=&HFPNB1EW0?P$-v&zivF;ldbXlmZp2iti;93-6iIrm}_UKe1KmGCJP==WB zJKjNE+8Z9dnEF%SW83$~fLnYnOzy8S?PFpvU~s85eQ7B<7D^SgOm(^;Hro!`BF(JQ z)F?jIsC@Ip#4lRZE=}dVzvR3X=Bd9NKx5^uVE4PHHYJo{zBYW{V&0_TG(9t>fm=Z| zuM;A-BFNE3o2k;LX7aTe+_OK?hw%XFi^qbXmi_w`5>#e*Kn z@CQ<0S0NuH@2kX_TCSm%OM98c!LCZy^;MiWvI zm|IkX%l8mo@uy z2efTwI#EU8hUh&#`>@WGRCPpspJPVoviO}W+=6xzOu&qdO5RpqSb#gml?!?Ay+U;9 zX=na26A42slykY{I*jz?u-Qx9hEXVPP;NWY1hoQr^vs}@Qd(|HPpG}j z+A774;-IT@ZnsXKlMz0&o`k&xE5lg~uYc5p1Z5^(1p;*m{5;P4K(IETL#OT zBix81K3-dO3kMMa@vC5joYuqxz-GR3U!q~b=s%`>-ThIEXq6{53$B&*>e(yjZ#9kve2ow>BJ-?>yT_H^nJ-bu64UnoI`(zWFvrP8hAUbcxzX zxq*B(EYhr$E?H!<9R)@v7E+S*+s8ct_|L!SO^Q^sKmkqFOUZvnk%}{odyFU7uDK;H zmns^e_QWG%`1BfkI$syMHLoCXc^wqwEynj$pRSke_Tn6smpqF39ei$UG zzoaMWQedGr42+c}(-#&v$BPu=gV4x<)k}9Pm936*5Npup1y7Q^ub-c98_@cIWH6Rn zS1{9d@{f|b2Qxcsr}raSg+CUBgwqQs|8{SGkFRo1hHFVoEan!cp`y!s?`b>SJ<7=4 zX|0gd3l-u{T-&gLIXJd=Kum%FUzajBr*%#5Ascj0Muksh)B4kE#Q(0Gwzt{hAx z@9R_fPIq)vGSDnzJJVQEX3*EKQZ?s5BleYx{_{HiXS*yUV(Vsh<}RI@wK?s9UT^fe zu>l?uW^N3Vw{6bA#;d&85tfRS#5t(dTZcyb{iihMu`3gfjIupvF!wjl6|b{M>=}9L zvz0PPQBoRv76!-O)B&=i{Q{1Q7z z9EBn5BZJX(k!B$i@+ARU@6SbH?@+%$HUxJ7kyyvAyT7&g_SN&cY@8;T3yoP{Ca@)T zz#-(W@Gbz0-VXv$Hba)~u}ou?BiaOU{7nX^e0=o9rd*0&TlKrJd1^cpD2N5RJpP>y z3CK{jd0h(TA+AaOrU(7knKEsAo%~5`?(Ca2?7sS-E9v+8=VnAb)%0w&pxQNQvGntG zOSK;Nh5*@_@RPwMZx=KT~rqAK}-UwQlA9kuw z4a}I`R$bqJ$g|xgh2ui`l_GsOyl8ARycR|j!IKR|Vh3R7`F9jXY>TB|dW@vV`3ksV zc}RkEmwyJ|LRHiInZ4epxi5m4fLs!Gsb&Dt(k2X``zI+g72aGh5R0~nR)w1tM^rth z9VQQImq}PiQQrj0>K0B;xTutx!KOXrxs1OUEY^bgb^Bq84(JC2X$17b%p?r9tZ}o=(SoN`6a<7h~ zlB5xQx%w-7G22A#QKN&oc66HLI{ToW=RN}*o+;PCOf|M`Rt$%d6BOf5h^u*2)j2IptDp6lh6hZFO2Ni93Z z&HX>kDWh2{1$ z@-0+yoV`1hj*O^waE6VqV_&os}*8*`YW_oi*U$>d@wP^G^?<0zAiuS2zx zq@o^Fl}i|hvq_bha=!<8OsQB|Dj--CS(4Ud*Z(AUPRS8I%_W+bwaa(3w+BMQd8bxI z3HjKPkj%9(UTo(d)dnWFV}g?F-czsrt6yQLJG*wDCPb?o;Rp!9Niw)*82DwAMkbM| zIBOUCPt*O~kFRB?iFuuGNM2jbRi_kAw?TsJkj;shafddpU6@XTx?mULQl_Ue8itWE zL~1N@_p#a8o&Hv1i1(w=I5WIWRp7%0uhW<_^PWzT|KQwIT3(i}ylP~Z!m`${S=qvFW~Fpv!S%Rxzq zd#$HSpBN;V!FFK|H=WSTW;-V$H-HC?R;Y9RQCX*bF{Jx?k?>7wpF4}D`|Wb)EiM~I zazmg^`}(2|>LIx>ttT%ZfU=%07gNrAn#0!cDYS#me&k8n>iqf~(;B%l{DmDARdJP6 z;KBe(!s!YNomV@NOQN;;URPSKAD2HZCr(Xwg~)o0hRuwy*ka@3{WjxKaM)z#7PjA$ zBws3U%E#oz{UeondwEI5>r3W?HmXC?tKzqHH4g=rb=K8SI}m`p+G^K5O_OUs)X#2g zq>c^oMV`bI{EO`W2sIle{STV;y}$y_WMDg2)kIZ&FX{WKAAmLBx4+${4Btjf%Av^^ zsOxOKIH|7k8!BOGN5GP!tt75xnQd=_meEVi;Jb+LjO#4Zzy1eJUzDn4OKg-`>saAy zCqjc(RN{e8f11hhqGX{e-B1igpB4HNR?E4VRBmS8+cWuq*TOL3yUpJ_erh>D5<?HS{d(Vk5=M_HP=A%o(yx!7Jl~+=4<^y8g#RghIce39-|+yCQVN>h6>{QC zx5PS>2<=Vg!y`^v%)N_HUGbC@a)blIA%Ba8v{ZMwoiPd@1ax*mL#Vxmzg!o)`&Crb zrbU7vC)C^{^^&pV0!@AgSn{{Zym#ko{bk>zc8p&Q442`9V`0_TGJ66M3~!{rVLTXN z`1;!v37&i71wDgbkXLEOlLli?R~lDdhHaSS6=xX>_EVLtxpkFb=VG9N4-_g*o4^bK z88&^qIxWVE3yBVYT`uX*tWaOZir5KO!rNLAD9FKO*uHy#BOo5?>H1NnK}`zqs$T-|5$bMn^>9b7%>w1q)=P<4 zA|xOXX&5X;^T&4|^j-`AWTi*MeO2sk8wdJBHiEjl$Z zrjGoL)%UwU?aNVyuY*JDY|zXmBu&7qebi_=`<-cM=pno%rjjKdNqTQ2na~ zpusE5SIH>c*!_Wq#u~FsWx{K5peAF z1w#DAFk~TLx$JImi=Pwfdpzo>mu}bU-5b|Z<+R80F^8B$@^&N6Iqh6}r(g=}GRBdj zPGFT&tck09Tc})oNqaaP3HI>@2KagGkeNXfy}sg9AguY_r&v)FMIN8pwucY!v%BfI zvYYPEC_Cx(xA#(e^JMx^0sFkjv2S7S3+rYxz;|FY(&RLk$C5%6hb%jAS~%Eo@?ik# zF5Mr`gjaLXFu8n0dN_?b**3-Jf=2B|GLs)f1LesfgKB}Z~K(kNI>N5N3=nO z!PUZ$$t}+#8_F~^c{OHofZbCQ{JPY@jVw2S7`^eUkHu~f;DfP>yD6KHilwe+zz%eU z)PTOym`?fm97)h!hNCMe(0mQJoR}IwquJQ!^o4g=8dS!xwzBHEyJJcb;zAst++p}u z$p?)x<=yt-O`na4;kkL74RX}>0AuFA=U)TeUPc_mr-5clwX|Qbpk>hWRK&)~pYnSP zx{=Qq%{q)%?wkBrkBF_60fX2Tg!+wZ?K#w3196|-f8we`9qPwk8|#e{zGM3#4vBYI zHubI79v)Or4e~lZ{b)w_57JyLwX{4?OP8QtH6V%$i8y?*e82O)f+o`XlQX_Wv|kzNz#}pk<@2Ex2%ToMC_V z?FGh=)eBqyWy+|N*Yd6~x#=4^F|xSJ7MnA`Zb-Zf&7G@1hB1b!W8#47m84lBhAY=)%K9y-64EoKyKx4v^JPPc;n`s*_TD0nh z-FA~|I*6(Ce9OtW)AJbmpRc`s! zU;b&eA=SLMM;%Er1TWf-iY)Kh$8rVcr^`CMjeYaJ#+|XL_LN@5OrTM=pnwO4SzB5Wjy@~H1 zlDvof5EzV>4W7M^Q{tsPTLD_tuIhHjawrI`>n##&w%0I z%E2bkoH7bUJ!O+VxM+5QOe>&gOcoweKmuaN->n@hDa0_fBsLUKC4wlg0wdj^q_u(7 zFTsOEw8)>(!7PP|f#s2rjj5Hz#i7}LaUm`)0S-C=fwcxi@g~jT5K8sc2;{hI@1tg~ zjPp1=MVp-nBPyzqiVD}uD9ajF*_%Z86XYkzlP@|VT%9o^BLk^deKV?;&SEf(u_QJ1 z;TsjBn39@eb43ZXWEuouUyUxUu0Yvs^;?aJ5H_`Fv&+azv5c#$CmP1Z{rxTOUsUuf z^qP%s`w3u;TYgid0G6w}hpVgW7iIzxk@lmEWlN<<7qkZ?Hd$m&dDG47M^zbeTYkH8 zW5IcX0*z_za%q$NyeeHPQcCNSC7Cwo7hl*wiS>(PP~;u=E3S~YR=5dHl+$qfQI!^K ztJ5m8F6SpH#uSmfWhK`0DAs0GfYQPF!O7kc7Pu`E6_qGdQis5-zAexMW_1>{E?Q z$miKVd{P3uE=*qp`ATCBBA8%-B%*5bv$I`_ML44MhciXVynM)d@fUR(iBp?x)>r4KMkG zFg6z@pI?VI>`_Cto!Ob8^LWYd=xC;ng;pc<%+j>YBp9p!nyK%@hq_?0ud4mBUXFSJ zPpiM7ai!h;{rOsIVasP5rvU=1bfSNU#gyQolXpIz>;dPqm~YRcays|dx4F<@r+-X5 zwDQz;zwk7^=lk!Ro;DtdHj;EMHE?p^ai2X0*�aFs(;)R5ytfA^psr|p_4eU}?Lh%T+qCxq*C|bYU+B<|rMfY}h zm&q_O{`)Aoc>l*pqOb+Qe^30sIBQ5H{flREvL-H0jR?!d&8@{>s^m|(=3#G0Kl3lS z)WQf*?9(0j2b7PwWI{sNZ3NF8Lx!7+2?>K2A881%sKg4THDt9o$pkH&EXXyy;oy!h z5u0bZwL@y#RQ|S{EY$=sX}ttf1UG>E&;OLj->MW_!oi>1AMv>yPHPg>B6;b;g}~(Y z{r(-R%?JkP#^F`^B3Kg>8?GH(J9cQtQw2Qxzl?mkV7D((TwPtIg$0ADiUqa0<2G3; zzF?xSlq{%HP|zD0t?y>gBnmH7Zb-4W2zkNq`v$(ke%4*a&62`46+k3wt}*lt?nd4~ zN4rwb74`)`vw&e=o-o0G+JQ-p;Na)yPWJ=#y!SrIV*0*!r{qKnH&3^0U%!$Tm%;|f zy}`=Tm^bTT`n(qt6GOtJNKL=s77|*y;Ea`l+*3QeauQ^FKR;|OE!jv)zDWciwRN;J z4GiI~*v>Im8TUvt#Zed7I%XhWdp}D9BElmCZm(~Pn3aka=p}w^bao08?8JUC0ue$6>68jD}$04Rn@bFRaheh9Yya>XI3^OH7e(! zZ}Gc1g!MjRjK{#lq!w~Q>oj?&(WGH*?nqYQ`Pk zCRafwzMSmCTwa91Rz5fBsOv@&&_@KQPyB5nprj1vQ9uBOtU>XpTN z@XFV{E>ts8ZIntnI=Q%$za@~7q#>#g>j&4~jw_xNB6bwj z@Or7Yc66#^1QzKLv+HWBGm}q*!*a*W${3K*4lth}ABD=M^PKrJQ-w9p)tsr-e&ceX z`lL*qa$<54PF{MW8{XLWsM+J>wZ`eiMoMb)`H`zuzx5crmWAMhme2S45>G8)*DQeS z<*(XIiW6EY-2Z@QtXKQ_H>qH9v4NQg6+-v*X!`x)$R;DvL~NkFYY(UUF`4Z z4GLb1Ij$am*#tc$J0-jd+cuz|g;nXIjU$fDMop1Ybs&yUyf$3^q-r*QS3ZsN+SBJ4 zgWR82hqsFx7Ajq0qv(iV_WTP%*0!4J4s)rO; z)ILggcXz9aOzedca-o{O+=c4D0?zApkUQ0wXkGz^<#kX&&;|(i9$Z{nf>YcPie}mj zFBW2=uBxWA3r;2?6cz>YZan(K#t)5+Ol(mlWMwhn@b6te+I>PRsEs!+n_TPf%lxJN zqr4p1dNP-Jc#*{o4|$~1hbzRX7yknI?HIhbla!bnxY^4&&Asd{`t9&fgXKwYFAdJv zCX=!8n;IqX^BsZ@#^`8XcJo6%4-(cMCZI(A_}AIM1^M z=uV~B7vRMixf84Frt*ou_qL-GJrOou3?JDQ(=%`jj-JilPH?@SVX8LrdawR=(M?c!H zD+4_#L8Nlh3;${Xaxmu%baW}rl)AE!cR5K()Rbbn#{?BkhZkF2;vM%P{=<`h#N%pf zjSnWp$-={~5TDc2hsSg{>qL4jPM>Zo<>PZKw1x>mkWC4uj2A}>ezgG4%w1}r$z+hYkLTN^MIf^HfLAC8H(V*we zfz{;L$Yl4oqtgph^J5_!lv=#vGrlr)-oB1@xQ%9e?6%CjAQKt?)iZq|r}d6Oqv?r> zF28t*g!yTL)o%Bl!$kPim5|D^wzl+J#G}k`wKcLxL0q&mD)AqZp3mHLOI-~Qs!Z0z z4-K2)znkQM$9a1 zzEaAMda1PKOp*5@9*0XGD`%b`c64@+^dyBVRadZLWyoZ^8R*3j$k9%Cn@hRt!h{@O zxg{b3AS#?6<<`#^gK)feyPvns%Y%zMy?wp=_`pl`8C??EUG|Hd&5nyR51UO*bI6If ztCwLkH3Ig3W!@zc5EAOSSCp3zw>Z9+QXC6dMOLJbyBHZ(;A=~3e^S0_c6+$si6xJ$ zn78w0@w-1WIkyQI%w$WViN6D{1^xbEtczSxYuvjVHS2Z0KIoDD?R2Tx?|vOme(3OU zc#L=|d@;78V|9v69$8H$4cv$^u^ch)#-F0vGc$mMk(fM3jSL@8Obe zW(Jp$8i6blx|trHkjm$O6&FJn#tYhEGsGGcy&D&*SA^`A+VZ(285y5Wmy=A+O#*#0 zi^;63MnXY1bZ^hs#1GS%0*j}TnwppgWUN32NwxWup$Y^kkxG_<4e{H1{*r0CH0 z<_heF)mU$nm-%1RePvf%OS>)Z?ykYz9Regka1X)V-Q5}q2^t{KxJz&e?(XjH?(W>R z-*fi4=l+B{Mt@*5gT?Av)m8JE@>JFNQa=2oMtmJANFqzg^atvU5Jq=azsL2(_W%N) z`!N>#@h&J4Ib~&b{tpe%pQx$M22FeImfehTBrMOauCG3t8=~`j-x%(Xt!{FtP?nBe zoTm9AsHO*^VR`f5_7_3tJO&1J17HFJzmRyWaI&b=2;mm5hoNB7$=wO^l8Kl1=&tOz z41a6xg(kOSIBpFV zom#7f%E`YrNyGA3wUwxnMC3y5-wUkp@$2#RwNPT(-HxRO@P>;+VVJcl2|^Ne>MKA` zOO0wC_+P&|IAk8rvk}J(z=Rg)AO@%12QSsjk_`y8SoQ>5f>ge`9z0_Np7|S9ESvzQ?UR2+^Ef> zz_&8(I#hHA|F;*byD|X*fp9)M$SsNlAQ`k=ckFmP#}lf1&)gjrH8?(OrjI0ff7Yt2 zx5}y~$X4gEp_b4dK;hblzP*h_PJ@Gs#|>bfyOT%8e55Obj(y2>xLq--t@6w6--(Z= zV`-tvhB>+9!rxoaJz)G24@j?0xVmsD_{5!m#KbWD!uv*#8w-u|S+CW*srT6B7|D-%6)gU3ByvZs*85$%N4hRPn&3sA?vp_9d> z{6?w|lZMWf*EE+((MN;Gwp(jS68C}R5M<>QWR_3z=a(|*a8;1b6ot-;X7}SwcnoE0 z**e{ZhC=#&MUmohI7NMC#N2``EloGEEPmDOeT#K~etB6%!OiCX$WnxF_!cMB^z0JT_jtN3aFIt2&nDnM7DyzgGBP zG~FtMyCqFoXBU0+GPe`_Y4B~SeciA~Q&e}snoOD>Dd zP*S)7RaV}}iHM~1jFzsLDbQ(cqx>VQFpx7gF%c4$IDtM+3*{006i%kdC|d!*FTU?@ zdc)QN11nq>V{+p%t&VRfHxZ}H9w^8*?w{Fiwdu`!jP*bU;V6UxjuVxor6a&P6{wAN z+dDe@`t~ZcaAgTujVk4P{3?JqE12T9&}dJGnE+m4ysJb2qk{dS*{% zI1$G$81&|@A?6UfTzN25S7Y;2vn)RT*RNmL*{Kk{&>2KW4q(E+mXjso;*|JoX|Oga zNl8lW{-CT80^$rJI6m#l4YN=Qr|UkAiBcjywD5U&{K@t+Iz7fo8xU8fPsW_J!(k|m z=e9z3s<`?G!Lxt_e-Vx+tdH&ytE*Mz85GlA69^=pCriy`^RMpt1WSme< z>g60i4?wTNkv};mIycIK>=CcfZ%(oX^7B3fm67YOp9mmlr_g;64e@z=hIK!mBxWuw zmU7Voymdi592^-om!>~BzN^X?qVGbQf{wj8BE^fcirlb%~HpA&#fLKOAPVPLDuj0>jCVoP_8@UV(&SPFd%hI zr*||0(oG=(kb&3k=-sK{=}IH)SSTD6R-^N7MI`RsPO%AF>%&ve7p|MN&wu=$73VAT z%zR-dBF#+;nwy(bn~`1pnKa*qq9a2;b+u|K4NOqxe*6vf4#)L-R^rzJa`1ytz?)pw zaGCxqKJ|vSd>N})C*$Mt`Fap43U6n|{m@qZ_s=Ax=>cz#ItA2?gm|JBcgTjnR@wuS zc56mj@h@y6Nrf;&R>j_4A_n1OEpLC>EIvFv2Kx~1&<%$xqzFTQC^HCnOX2vQ!SA(` z*eh44l=?hcpUDCLv2U%#E9&t27d1Y$niBfNsJNEFeR{q?9g5?2BXQV%9?9ly2M?o$h-6 zmOcPq(7z2H_VqlE%^1~rgY8Cofw`MW()T9K*gBYmKH&-;y`3rkrKRBaj9RVCsI?*G z*r{2zbm{54SCADQEQ8AH7C#7$>AU`xUZ`~V$riuCW(mqMJVjg2tryVFSMC5;IaR> zyzWqIe|z$wP%hVhh1dNptwPa|Dd-mIWoEbR>grhee%b%dK?V(E;zEIPfUj;zeFHma zkX#kE8i&AOAxYKT2JRawO%zW@Qu8hJXQ!o^#V=oC-XPk`^?hwIW-xl5+(`v9Mcqc6 zv5g>#SSMr)3zM*hJ4WvNGsRLPDQN2O(GSEzk|k*4IS7Y)ht=@a+4F)iQM7ysu|={^p(t%pr?(T5h3epY!d z8_HaO$r5KVhpuVD5p+@BtCc-t+MaR#QUuix&<)G4>2a zvdEr)kBB=7jwe>urLlxFJls+6FY?)_T_!8P4o6&4JFARk4aGE$N+jgdkg=yKi`aqi z%c>K#K+4CfS!75^$c^a<1=&I?%cN6M*(r3YhiKB&(bcW3{F)W7U%*DP7ya|+j8L1e z-wOc5ptVf*M}06`2un0y5m`uUrX{R9udy*OJTJEdG)IEj@NYCZkV!;alI#R?si;3j z!Dojjgf`x9kt>fYT6*jZ`16uyN%PI~5pR?m5L1hZ&3AE7G|!-;G`U`q2noZjgcT}f zUR>Ro&+`ZMH+t3dkfPFR#KLy^jcQ+lDfKrDO)Ccf;5BV?O*^bKOi?rlBinx51K}WC zVGHe@PUX88`l0M4SU``MiOzgSC4MvVHio-^3Er2z@We#pKi=xCIjO)Kc4X+VM~#5t zgRNO?@nRYvce(EGxE#Q}L?>Ix<+~IfB(V=dx5D|OSwi?2xH+;Iubh`H)8SDxv*BXW zE5v_%vFS(afK<^S-b&#q=srVkNtDG)&ds$WFL}(#{&0DPYbX*_%0=h(bpxA_cKOHL z$=GS?Nrlq%QxFjw-cNM$rG85g1iiz0z&p+MKy@2C#D)P)lgeVgukPUBm%S}S$@r@= zAvt;f-z+};51w>%dLUXB*4XPkUoDR z425`GX8Em@@b;KSVxWZcBHFOOM26WOSmbg$yMoBn*@enhNA{ zKNSX^$G$(U%-Js2O;R)nc0XQQ)YXQ{Xilc9&9 za#1-+L~HS_X|(Pm(Zg&nFE7&t-Dp4H{@^3v>4=6ir>^AbW>AjE+U0hy>+aUJkB(&o z+QrQ#0h|R!D}1Td^qyf!bWDt1;}u`dG~8)sfIp^|vH$yPcux-uMjcSxXK?N|HjcAB z-R?VN;cn>n^!Jk!B?Uo;&sEF^t|HPJy}rxgg%6H-bAFJ_j>X5tg7!Xq?_r{|b-Pc&|2$pE#F*M7 zhuUGSX{@fR4`+v%sHJlNb(OJJ!4T7SgRk z)Wu86I~HQgFwOTl=IjCH!c_H_iJCPeCe`|-qYT@iCd!a{95@y zeI){J&jyLR9d~zkDw*U)wGj(ZRK2e?1*6A22+l5Upf0s)Z(NOE_P}KiC_uZMBdxHO zW*54nsr81m9o#l+NF(`sIT!g!{xQl|*-W?B$cFch$iA=M;>z9@qW8kh{fc?mM&iF= zlc2tT%CbuGcb}%gj=a41ff|Mto~70e$vk<#}W_{a0xFJI@SS2>%q&xsv0*#c{d%nC3@At$er`ga5!1aiurzB$}#iyf| zscLEQ-+sc+X2~2Ye25K;z%Hz5s@$$q#QvqB(JG^Vt#L)A@!#Dmgjy_`1Tz~em^tE7pGy}TYh zpHG&o=C;dt8HKGAwJF%sxhVPx@*oFB2ggQa<{Nn)e7k*l)fhXD#a(~>_#3w+&iJ^y zUm;#&w^ph#JZIPJehf}~IXF*j2d8mo6&sBSRq^aeu~YYjR0Fuw%)Ww#PPMF?WtBF6>W1vD3{m=C>{+uI4f3P*)c0=8ecgsq*phfzr{?eo++@HkW~qfmuC}GnzW5*XY}( zka`Eu!T?Mrpk&eh`=sPvBp{c>Dk2q40?+C((Ux@#e+|Q59l2~5lwLl{F+!e*L=<2d zElU!)tH5|~=mHDE6yw(Z+gU5=DGlYfzJG9_u1T&Q2GJdx7!T{5l`2>4vGa#~COtjz z+umV9+co=vP{(QUtYxUXJFSv{`^iTB&;1TQ`nviUrNACK5GEa0P2S|#(2%nmSK96N z(H>mPmso2Q#*o=)?^|MS6no#h6GZnSEJvU+pGYDI7viE99-a7-zkI_#ps zIE>#BB5C`9;WIv0VAQ0<4JofjLulg1W+)bOeg8Fp3?C(hp6Ua1q0L<4UhdsSfax}9 z_wP03W?_AMyZuNk?CUAUtu#W$P6v{Z(V35gYJYo6Z#RJO)a{STY%=OT;#5P{sY}8EjJ1HOs2fSQT3k@K7_v*JkftiKp5{y{+?DdbQJq(pW8yjI!0J#a^ zVCD}o&4^OQ_+RI1J*>St! ziGs$`o1=($+5m>0HkU{ffKrAwHh1b}nWDx>rg4`9!a&zBF7Kf=eThk4=^kbP8X0RV&Kqi*BXS9A0Q9zX;)U$HMd$edYT4qtxj{y8>0%;Msu zp_7Uh4-5Jxjim16;el@d`TFwC?4^YFRvZtaokSSt*v%9HWx~$f{C63~lr~EPx1X;p zY;p)U0QDIPWRVRh0;y0PGSh1j-CH{Vaejrhn13QNkPkOOf3EX*^AI*l>NoBZgKFpo zN!QW9L^id~U)$kT=-&hfFUgwQ+4>DI79#h7zgd=I`#Xyp26n4UN8~3`9@32907P}X z*2>55b0_c91ebIq!Lo88jA}GiAs`1gH#cD~L-w0@Z(x`^qe*T{8G+zqASD0R{FHlt zlU%4nN%^(xVSn`?99Lv(fn!o9ui!;x!|O2j>%jY~4K7QlE~MUYw&b(UU+d{soeE%< z&)uAyoB)bD0{QdMhFtSqI`ewI+&tNdhl7KGGL~xmaPc=xJ+UMdKMg<}Toh>&_3uQL zQ;?kr_hnB0mDBxybJ`Jk%jXK!Gv}K;|N!pfIRoHoCht}nkHS) z)ok?LP}_bsM?u>giAx!E26K95mvr{m?-VYf{Sw9Oq6AO#2|FJy7Wf?2t_ti2q88D}eswAc&nV<% z8k%2i>_|sUl=0fk{aD>TI}-pzW%DPbv{{NGO1}Gs@iExa?Zd-ELlY>ws6}Xp{ocqfE*q$s!G3 zEE_rxPnSGQGcij|PM6$6EG&$fQ*)IvgKX=OrS%)~Qa!EM;LF19lN?G^a!Y_ppy9UXNk$l;D{6% zH)x=Uhz6_?S@TikVz|OVPEM1)Z4gV%06vun?;lx$ufeXRHR)FjaTlQFMM5zL^tT{R zXpNL_NnL?<%Y35mFF$K6S{+Z9dbhTW*>8g!+b!t;TV2WF(TqgzicI0-DQLMkbd`yc5)(5?`C_;93LFq zGAX*-`F&YP4n zA&q^WLqZw=hs6ZDWTHNB6{-_yUJY zgjSIVR|C*San%kg8#NLFbN&o#Sot<3U|+gm&K;&=m-r&Cjmc$i&>g z6WUA{D1>9vMSl>fGw!G{Xm^6uc81I($0bFC2i9u$_cpx|?YQpnnA6GJd%(TXIO!UD z1o%v6;}V$yNI8s2v@D-)i9g+_v;*W8a$@YGM!xFX-7|WwR1|c-dWHV0p^}lxX1?71 z-O1t@so4-z>u_I`jnCZymr)mYB{41@-q~6!X_JnPDikh2JEDU?#FEm9iHZLAXX$TG z9~O};Jxad5^SB<+SKbD@ealzZYjuyXO6pX&+0PVwuplbGX# zXYsW7K0*~~II1Dz8#LSRHaMwS00yIe5TyU{9O`eA&ZNOk$>$Jv_YJADiEOTeBY5-^ zCwF(FT?lXsLF6uM=;NLuzy!m`Y`IykgCo)*S{WW{-o9Pq3#HJDQrdSv1$^Wq7Bgb} ztU_UEw^rS`qaC1GZ8DE{a0ilAX9VPv#Tu%q+7bon$l2EqH8yxI?w4(xPF!9Gt=pZA)-d5>ItY8mZnLPXg7z^&m z7f@AArp-s)IBpclf{aa^4=_QyszxKDsq^A^kl%)`2v{y?f~R=AZ+gT*x|^5hD~(p! z7t^)w-C+S6h8O|jp27I|goIT-_nPKUHsbzo?Z7Ds-o?<{eY+B?Eg>>Av5=Ikt%n@i-v=U7YKqxE)}Ta5wFW-r zO$7>v7dZ5MuNoI?|9-HYAOY$l!00T8$NL=|{A&(k20e$|0$~Z~9_iv!q=e15}q6p{E4b;~ET{tNJ+Ppp5Y~>w7!=r_c9vsM6UYaNB$82_Fix~2!@;edD z;3&sz#*HpAYh@iOGxVb#hRN6|t3yI6Kfmy;_2?cb z$v{rA2l*)`{ph6AumXHADK`K;7)c9>0TrouJp%K@fDa zLI(0*lswo0yn+Y_(&xTW8#o8+^RrXKZHQF@(TP7GubYwW&Uy-Pps%VfqRCiXVaR`S z25x}u6&2+Np062edAai9!Y(8;bv5Yy(~9N5@5$cba`J?d?Jv+zLtA`CVkWk%@HjE> zb?&i>vf<3r>(S#LuC8otu$uFmc%jIzBUr5Mb4zYi47okMl)+#F$je0(ZRmT=Nq z<+C7gcJK-4&{Y+6qP21jz(dp1auhfP`5#i_+dFTOpr)tu8Z}Dl)C$L*HlNwlrd2~E z9IF<?oRHZ^$wgR5&l|9oq5uZvVr$gL$%;>UnJ(@(9f`Q(ppuzj9O`Wx@%s~ z*zCKNOoT+&PsHbDd4**<_k1qG10-f`Jb0eBQ#n+Bi%mh-AvxdW%{9h4nD7un_L?j_ z&2;xs61*C6Lxn5y=s5f4=0@hW8)ZI3%jQq*B$zWHOD?;UiYx29^@eNK&F(^5b9+8# ze}-C2(a9*YLM?58=C^E{9HEP0t?;ber0_}-L#l_T}waO&@Gej7ALQ<=&szsk0)P>+gs3wJECWyJ(x-2oeB%Z`L zz2jmDh}o_c2sxR`-A&WJsXPTrRDdMb+hgbL#%BJF1m^=E9!K3Lj;2G%oOJN&(}Tx6 zYHg#m>s#mu_ITOwgitczv;kGkQL zqCZm5WMackP>Bdo4)g*704t&gCbSB=1vR;z!3`1A&{{lJ#;Pdal~$cD%?y1j>M}ES zn-7NKe?0S6{AR~fzj ziHJ96m*oH6_Q>zFHL%(Jj2KK|ey~_`9!p8hV?A0hYQr=~89Afe0SBP|yTSO(X5ZbG z-2s}}G)j=$v}|M82PL-5+S-KRU?d$)%wqeF;n(|hjqN@~{e}x%^B8%}m-F@#FntO_ z@C=&c0KF4W@y8{MCe!XmAm$m;geW^NsTHH(dH-4LhB{`Gj}hm#Yu~yI_jH$ZZ3YC99t@$HpJ;bvJNkR1Qv9 z#-np*WmP3H79*P6Pr)fEe`c*mNT(Z|uz|5mOS3hG&B$0&NUf9q@`Y-SGLFB_dUdXU znU94fR2X)Ql@+kkDJpy^KOq*O3A!bEcz6IjiE&602lQyE6S<2nuiB&JN6Q>Doy$C8 zqIxCjNnSZU;#+%1`pof|67fH>%u4Erq6!+cWEl}`km&G#^xI6fXvAL>L_{DeJt_mm z&iL4%YKeNHUv6Rc2&cm}EbP+a?F=$#)*+;**~jwb(+4S89PJ+zA%K%pgEa|oAOeTk z>@a*prwpFqO(S+kJ-Pq$@npbO@QDu;V=)}}rQe_VyO70M>3(KFvD3D4O2J=k7J53s zf5vJcn3<7b{iqKZt&l>gC`F^y>46BzYV&=xucH_>*Xjy~op!yN%?t;J${yOmM7Er~ zys5oCnlCyU+TmQ;{cH*TcefFP3uhIhO#to;OmCj%zh6S z{!)iqZL&It(-s?Gwf^dA&H=R!mi;*#Mb=!YXa^iD@BK3c36aDUDU}wET)Mvg-k~2q znQo<&zJSfvXrDYSMUm75kmY!n8z$^@XX2p;3gf!A7k!8K1qB8DE;{!CC&5ayhY-V0 zS?eK2TzafkpBt6p*~P_OzTKNsz9J*XTmW|SEVS<=){~^Ci2?^3fx5W=ONnT`oih%m zkYRxW0$Ifx2 zZM#xQ%X!rny6ko`0o!wfl$4`7t=jgmMLpbayH>q+1t}R8xl(N=lKWY=(HT;-M-T}0 z0Ak?p*UR|Ytrb}fpInOwP=(GjUms$xJKXbU#k*4ja-S2UKioX?56q&OVise_($OaX z#6y|%0cT`B8&|1&KaDu4K51wot~lE(Z0=0ueR*i}69Pc!XGXAvd#;64pw_ch!%(Np zxeMkmv&}z6CF;M@EH-ZdtN-5aF5nGYA_}eil~OGWJL0kK_bgCM^24kbAdEL0{V|+k zR)I_jlwB+Zg6eSxWiEnh$;J6!nXT1@9VOwmKst9|-Z@;&*Vzf}?U{@D5+JKesi>&v z>P9M4oXj?h@mnWDKi!ja0>~Jsz#E#HdV8cgja>FwQ4kSt@6jRP(GH?Gzg4w@x(`&g z`y;_Ke*jUBms65kr_Pqkal%nL!nE&ZpFuP+GxPmw*ded5Py{$Z4Tup1fra?keEK3A zbEOn4r}C^^T zg%OHxxVl0bzh6gSHfnJ2Hd!CiLokip8la29g@5_^yG$I{cB!@(aHkVUr?s}e0$vrV zn3RP^IS&hqq=f~7a>SNzHJShz3N}fpw*kssG zbUXo0!!&*?DS2C0adCgZr3}aqf|eV~@bK_C#4U`Ce$Q}T*4iv2XKMrNJ4(Kg1_zdz z&YdCTSS$9eKKFq4zC@l+DAf!*-!pCJ>b9&MYd_4d4Tj zTU}WR1`+>tSU&_zA)gFLvqn(JIH$z2^o@<78YICWh2Ie-3qPw zJVgas;quCbd5d7r7TURTmd0d*Qaaa)Epck}tMd{iB!J z27SKPYF5omICBsZ60Z5%OUJ7u?oV;Nl|f;wbXw?51hbr?1pNn}JNt0_lwWBHNflg@ z506jd29*z8pg1PQbX{C;`f_= zSIq~EYJ0oGO<_qoNlF>QvLh`X+Y*-c`#>xPsN$__k9yhILZ3?1kJdO7?233U5z&7l zA-}vHM-g&io-Guq(AVN#a17)FqD==U?CzprKq@b3woyoC!!~}C>+#{)8ACNTsgOT# z+@tEx4ImF{jfrMnD-P!qLt)Hwf3|*mae))zIwpm|v|MY2{uX|^{BASTfH<&K=`wzH zN-?<>({O+OH0ymBk|JnU5TDCsn<3(2yl?^7Kst;ZVPo+ni9+pG>n`U;iG4-Kr>9E* z_E5;(b}hj0J}ao(d|caccc>SaDPZ^dp*uW^MsQWb1EqIb0o&Qt`DiWILq%m%%tq7H z!a^zIh2--g5v#PcG<*b18<3T9#pd(Xl-yFJ# z11({<=EI#hvO-ga^+CizChkwQ0%PUn#P2#T2Oat{3$c{lVCq#pk&e&q&tp*mn(v9} ziHIr02V|A){;$9>jOUe=1tUfZ-8y;s`GH}G<#q6J_W|$Q{lt6~3g8zK1V?$^;E8X{ zEc(snpEb?lOo_hdC9!7c0lKJ<527VzUa%t<;Bsl;X15E5p&{R~jZG!ywe4wH-!f)s zaozs9*x(coY>sP9UZ>~n0mDOM$viHimI{^RJF5}fg{zG&+vghXLD=V^_;^D6GRZ%? ziCf5>JVsM+Rjnpn#z#%u@6LsaRJJQeQ{P`5wuuIE;;3C=(G|fW&#>&l03nEviRoc6 z>J4dX19Bc?NqdDyIh@G8j-!6&vwwkK8!r8Z+<;D#h}Xr?%efeVEzSE@k9OBF9kk_o zd$h}A^R}hZLddljsHrrPz(E!x=C@?4EHGED=k#>Lp==pJ80o<6xKUYmDo7^U;?W(( z1LSB89fL|RbB!U6AU{MLxVnkg6tHygcIsj{!>xU9tTt87En~Z`!#Xpy+OaW#muvU- zd6wuh&dL3#V7*)pU4urRd9MV?=Vqi0*^+vU{a2GadVR) zYGUu~JoTg9KOzcPIh|XYn}uAY?e%EQ*O~_2lV>l~oYKF&$)fjO%H%H~13!DgQ zJ&PFzFOC2PIw=v{Xgx~~03`DfPqGwIW1g=S`Y2X7NQ)1i@Uz7x^*@uwuz>Fk=|9nB zYg~85H9UV@!DV(T3K42t+ciVA7^Uyf@WV&50S7PPA-1d4{y>=H`R;sVKt5?$+?Gv# zcDYwvuGm2A;d*=>p2;nE%w}`yug&=L{W<&BXP^|Q)kHS6x4Nk+l{iB_tf|Z7%kWATWW5#hZX3uJnBjr(xyH_q{8{MvLNh2V} zCdI))0mCVBEiCfaBA&oxGD3(#Ua3JD6=*S?`30p{@M9;8@RQ?T6XU{uAKH%-3gN-s3ZCp~xD3>S>k3AMX_6@lOPh zsi?=^hxn_V9cksGrMgqct)5K)I|Mh3MDitBL=dCHGU0O?5R__CZ%hhRAd9aU^)S$?Cf4eoar4tumRyEhvsLXh_ z-d;9qeYVmDso$yBTyHgFZbC%ZV(gL_C*V20MylNL{SgnH2y!4C$V*}QVTLlFE>t1= zu-WVJKCUU?Nt{Dq3pK_#UhmzSk7~2Cx=x%rtcKwwhRF|%3>oV4eJJ)f1o7G}bsqKz z=BYAxMs%|I9ZcviprG1GsC;0&dfN!<1}fga$UuxgPcxf&1fv-?T~Z>6WSte?&tfrWUF; z6+^qqRvx$_M}LII4_Bt(R{Wx^O?3UaOw+?2i1r{n z5(4SNhrVlAU*7mebXZaN3y3rU7pgp`gRyiSaY`7}LVZJJi7rVg5bPJs7D>^yw!1sD z4T;(sYz{hSNf@A2cDj(q)g>NkN?v?Cb}>R8q1RYY#-2ZmJSV`hb4TGa`8#>sLy-Ry z0mEbV^xE#@>C4r6eT8P#74EjDr4>d&nAjC=`F8dOgx0Iq0UnL2A-Ecb7cByWO&_x1 zZ%eT9j{(#3LbGgHK;Vdsyyt$TbYx!N=N{050Omiuly>j(YAP}6A)KZ{N1J)I-UGf0 z;lZ(E&>Z;CayK02uRtg#Ka_SjxgsXBTt*m;vYX{Apj4LQYh^xWq$bt4pUR%;3Ezg| zIfpI6!17SDuylw8svtD>Qt;ARcR-}ml3-yGk&hGim9OHFcG~nudM`#B%EgvrfK#A z8p|&0Xn1JWoKzZmdaIm9LIP4!fvk57At52YujmtwhS<8HZT^t*9U8Gq_{^R(EG#sl zqO5G=6gThj!uSX zo5|xN`k?s!_RRN}a2zXvh8|3`6t?xr^^ZE9j%^k*P2q=BCKHLEKp&g4d%zMNWDxgQ z&i*IDF^AP?Np)BGqG#qRk08)OsZ3Q897I*?wk&ml8pdZut2VHOy(LautNgnGcIuyL zkb}ek+6an`FAO%(XgrN``Nl@m=8+7WLnc_&)opR12(3^5rwTk!%YRxs^ zThxrL`0#y07-j~0c?mEKe#wFi&cyx!Lm&G|>`03>$z@olz%AS9TUIhI?w8K{o2}vhOVX{8Lk{U_{W>l$=XAk`wFUqd8Fd z@e7k~%e%NcA>LF3h;2nTGNt|pCx&xizYsL6R~R(QkHr5jD-=|&36L5g?0xIHwRc25Kxif%a&I-6DjCDjuz}jml8)VNd;Z47TEJX#97~t>dbFI zn{AKglf%LCm~z|IrPbtFBhX9m-y{3VIa__ zV&%Y!7(yl28*hS1zj^&CkCF4MP2IHgqtFtn6+^9(`j_Ff(61zI4#Ssz{cf5 z6Tb@-P=M?XG=Dah+Q44@v`bJlU#J9h$?wdUvj93?puHVE-=75l&&uTpFrUa{W!V|z%1~=*%_p&!??bm z*n^`0_7nv*b*qu%_js-WjTyjCa%H?&6u15HB6xhJ7|153C^2RNc2!p0h7iIqK#_|} zN&+$g6dPgTeE3Pc<8(B1!3=&U$zPgaCsu_El+v-+YQdpD|1+3%B2gJ$O}eg4?! zeyR%SSd`ap&&zIUA~tQBHgoKdZ(JC2;)pQW>i~A?@h}glF=ttS-*72{x>)U9j2hoJ zGc${u+D&Uk;CBb`7^nA+XAzI`OBgyamtlu8@=_M-<$T2*z_y8l3@1d0_>gq!PHYCe zT+Em`IP__m$bfXC?#lGE&&5VukKo4PN1DLu9uy^5gCbp8PzhL=U`VxM?bXD@)uhr4 z;Ko0H23nwepTuWj8dhx__67IgUAINzrMw{Ci6wyX3@Hd%v}@bEu8B~~jb7nN4uP%( zK+lK4kumi@39C(aco97pzkz(_NJg(;ucoGyOH>EY9U@QlP>!ul)OqFsAHP*Ll_7lZ zKLtd&sWU*8m~U4)vZGfyiJz}E5)u)i&BC-CY@GYu{`CO52#e%wce%Z(w*{@OyX21xs!G#q1j~jJNAK zDtRnz=k67|C|YeNl@;an^xf&Z0&pnJ;IH~n2_tjOJs_q=3r?fp$I zcIj)yqsvst;d$#l!@%MshYFA;kz>pR3_bE`{HLIM(XW3JVILof$$h|p%Sga-;yd5( zw)e%t^D20d`u(^V6RkqZ+nG{m0=uBbIc1B}_MeiH=bgcE0E!YgEL6gA2OvkRw3$qYAWnPV9wnuwx@^82 zuKBs1RRD@U@tW<;l$=8#x@PXI{c>AGBSL)|Q-V^M<=M(#T-NG`-F)B_g4&y#fcaMi zX6M@@*spx^gwy-7B&6bt%QY54+dW)g``%tXW_u9T#9j=&s!BbVe@Agyj8e}*;sc(L zCm>89zA3M$_yjmBMy>u#R;k3N=_OxqeG;P}ToHFfs%m0RY#!0+&T#Eqg<7+rJPN}q z3y(D|kx5Iq*DLKv}$ZVTUK+ zHk(tIDypF+hNSNR3M#;V!t(M0t&}0im^56dkZQ(ROuG)0!^sA*7S*QCkoS7l)<5EV z;?fHc+WHkz=TM1+jb9oVHn`h-4_^QWPE`CP1^olCpH!xXqo409c821?ogkyS8igAp z0M!7F?AH1(3WQ)S$OfJsV-Dio;80SXUPB8D>FEU7e2Q_5AT&zijY(ZVljkbP=fQ(Q zCo=p8tTTI}4DlggW5EqhS@Y~LK=7aoQ;$r!;B8z!#QJZ`mq>g}hn{yeHxF|{BWRJe z7gl23SZDWF*bDUe67~Cb`Zf{`O*MDkqMTWL5rQ1KImfW^+@?Hoad~;z^vXH1Ckz=0 z7HU{=n6s*yIj)fYdYp{weDwlX=Nlm^5v1cj9DF$7&5Ed+L?IPy0?ccGgDJjodbGtO zK6Ju$kQuuO2dQz3K?fJ8_BsAl#ne%<2-v)KSuQ^I_4hwM!vi4t#bLS$9`ha;2N^e5 zVCVU#p;;^v^vZmhPW|=8#mMmR&mT>hRyZ02;uv|_K}eMUOHl(20i*U`uNjp;xe`S` zj}$IFae7OIKdbOE8%F*6TUCWD$k+GUi-6U8i&kA1usU@p##MO#_9a8c|X7zWr|4@SkrwdC1<4sZnB zF1CUPa{hf6BM+^6`?6%>irteWaQ^sru!GfE z2#&D-xuP`+nLV5kT@>-*{6C8i zZjkQo25Bh?=?>{`=@PsfeShQo?q!^R80YN0*IsL`IoGV`xkx%1QM#?p)ctjl4xl18 zU4Wte3eDtK(H30RbvD-slCqVq z&J}vXy%5p${5xQ8x1)zXu)2f&&;HQ+Ci=i6KY3=~j|U+mx2e-_m2s}eyV6rffwH@o zm!~KE2h~>q+1lECKE<(0o?5i zH-49wsVjYL?r=b0ISbz@t$`-3*)s+(4W2HwFO7~S@$6=D>HB}c zIC{gLGu~D1W5gv!6eR3VLKBZH^9I=Nng4C~(m8;n*#m=wAz`S6uX?5Z3BY&_5QoBckm zI6LO{2g`_zjGXU5<89SSYwyp8^6}N>CeJ?=aB`G+dA@g++LuK}HPefW7X8(0+`O`1 zI_r<8S-Q}zh%b(AvjehPJlEoSC=Wd441!H~8*Dg?MnVn#JcMY4^mPk-)#{0YV%K%0WG2^XGCbIgE1FHa{@VDgsg|C=m2u&7?jDj1JRUIKN z$Eya)S_R8w*j%<(HJp_?#QdKq=_R=|-vQL50VP^@SeOL)%Aoo>$8{8Ic=(4H0#>(Q z`xf?a3tsn8=*$@l)2$NI}^8V}qyPMtSe$ zv%udR!O6}a+4)jxf}EN|!+G2db<#DMm<>9hJdUgWXrS&dV<#@y`TP!O`7wQYg+v9Y zHRt~TDGMM8nF&X`hsO7BBnM!=#ofK5m|PKSW|z?S@%gcr zMthyJ8VC{wYCXUC%`x&vsvzc@kP#C59$YxzJSLTM1;0KnZbF>0cT0lw-dABE zJO7YfL<*u;FTFdUFI-YIAMir3qmCJP&;3mU6uxPzA>9LP@k8=|c1^l*cek<);G4NV zn+|AmoV$_%(B!u+@9`A9Y=`g8eD%)9qzo$bZ`Yp1243UsU?&PDM=3dSQ6fK+a)0t# z`Cd6!iA+OY7WH#3Y9=I-%cOK^4WMZwJ|Sk{Z~Z>x|Aa%NIGpyb?%`p-Ck zvJ#|V{K=cBT&{4!sbNbmrGG-x=?5tcfBp0{H|)#l18@ME1f#n+ItPs>hUMe4q_w#} zlA;c~tT8g@zw2m3nu4d!>wG zvPs%Q(xvlk1i;beutX8?Y#*iu}aQ86;P2(sE0VjI)04 zSk#Lxva~yvVSMkVgRMb=5uPcIX=EdVB7VYd6lx)vJReV@AFYHvMVQCpu(n%|VMl zzq->ZOozBs+AZ=EcedXhwgpK{DF93)KBp5uQ1%$T|I~*fOzxgu>@;`zJoZZ^E1jgj zcN{J@c(y;?#9c_uO6B;oQ@CLi6@9wg?)gLL25r+l;>+-d3!{KX*R!SbdKSR@ydmND zxV(897xGBgk^+N=0b`NH=76*PEnDc1-O*UbmzZWcqnDSd5|#O_6JI6_Dshq`-XaBP z&KsU_eqrF@f0#%sPMiHa)aMZ8{PXwBH!-b@Y)Pxxf;~RiO>vQqZ+Kgq4gg`(CJcwq$CnGfFZh9t3`akk(EE)9;k5DRPkv)K`{axa-L0pa z8)JKuPiGg2$~Zfxr)im)Ry*h8j7GT7Ziwz)xuR)ev8||Xh_m6wjaZ%utwFMS986e} z1TV*kSVA#TW2{=tghPw~Kig=lt@$`5#O>X9#y@}3wJ{gcLSqJI3XyP;8G%jwbUZK6M@fZ z$q{l`h=B|%8YUi}ZO2=bnx1ZEYx@^gIF?W}j8j@gU_}ju5JU_oo=ZWPwyT;r7_(ho z%2e7TK;UAJLmYO24ZISZ%S=G$%G8sUd9{q@ML0+6+11ust$I18T9oMi5&k>RIj5S<}q>Q$=+=OEC`9 zfR#>7wHCa(r0Jik|ufg}U?G~33D9Ot0;AY$9BE5uPrfn{^_ZyFYaAhJS zBj5aZL-fiOeOv@4M6cu1#!|MDh7=b&w8lG zvZaVesc~gBifUE+1X!q=oh>(C4i{b?D+*?>xqgcUF!$Cvy{Q-&O!jU@^;ZQeA&o{p zB<~ERKV9T+4WRAq?UC?*1sdDh?7Bq$%1Rk4B=eQ{t&=8)^UQV`<8{+LVOE5O1bfC5#0CEfp3}j2Epom(`eg$wslD#5Lm$p0YL^Rg-FO|d#|Uf z2L~M)^KuYZvr6N$SP5(xip%@{b$PNflLPdg0ue0W&QBEUMZi3+alt-U>UJROBgLga zhR%_T82!1VNe!iOxVN{SyjtzU*GeQiIgF_}R8>MzIEuEqgk z+E1`nZ%IIN!xf~kkczkZE|5WZZGA$??ecoWzG$ybVTw^I=@rTC{alsdnPJPBbo0hE z&apC$u{5uwQ9?ay%zO5Ry8uwgX1iKV23A3u3NX&&3VDD#)=xM`BiY=zs>*gz``{N1 z7nfB2+}GBx9|dj@dXzq;%0tx5R~j4HD=C_Ur#Z9zXLy|yaPeRYbSGe!G2Yu#B?Yu@ zQO-PsH^hOVwvy(XJ051Etb&;P@b`nmi|Fs6hnITDcM}+Km9pgs!4Xa-eVk6Xe&0D^ zC%IB*OSp2KQAQJdvMfDGYr(L4l=BxJ?&eqXlV2G1W7KHNf<#l|;;JUi(*n7=JonyC zB9Ne>pmsTI(NA(jE&=sJzlugpj44wCdVAB00g&1km9zWp>QBz`wToudwR#Ybr8s|l zQI0>!zBqw1MaUnkJgr=sMr-ez63VOzOI7X|N=ykae-sx$@}6>>D!FmrP7XKhSmZPW zf~~@`rPJvLZYp`fOc_o6{Zp#6F{s4vX69J{oNY38H}3wuOd;R2duIR)_+;4`Iuiq@ zA4uypnXewMcXzyO?NYk^YyJtxPDBQMWrGc>Jk_1{H*{Unqiw3+G|fxUnA1!$!{OcO z$*Q#^EMai2b3&5~8#EqT3(=N=W?DUQOL-wn-GD-x>2c(h08QF4%nF>EepoLlV1me zWlNPi1z9B45>cTAvGJaMxo5~;#k|zeb5urFi|sc~siIn{C?X-b*MdT&M!oY7j}1jW zDXG1mKN$l8#xn$cZboF3SgL{k!539CgFzZXSB&P&7#;_Z%2jpGO@+>a2QN@cM`#uBn!QruC2Ua;EOOA$9YgqI3YY)K61iAa8ds0h%AiQT zD!kS6po*#NcKI#IX88RsUZyL)lwnF;(9o*FsFk_8m_mWl@y?=;FqsTwms z)rof{c6SPzARWV!Z{3i-dRBf-8vulnb`>cMMi{asrP`2@QoaKZEM<0lIR?f;;Okr0 zg;_YX5+f?3#a>1CdEBH0qorxlRdb4y<;Z>?wN=C>8O3V)G|BEd;}+4tL%+Hr`W!&H zO2w0{{ROiTuqkElClXia*SQ>K5UuUX?{K^+i5j~P0iGp56r^dOrvl_n((iI+4*k;u zm2)MB2l0ABp2FUa+|kXz1Gaf?e;M>39&I04ElZh$%Zaw}3pWlu z|0dHpNhlHMRT-V+lIe%K|425`=7fO)%bMZ8r z8nca@3BWS^cY;iU#%A3SMeFkGnq`%|#wjh`rzFKLNCdK?Ien}pw>4gb^hbN8V_+~c z)1+lstQ0jgA?F{t+T{eJ;zV5I42puKJeHIktSK2b%Mp%&RH$$x!Lskst&33ogt7%J zBeF~_DGY8LqKQ@80(5uM9Pod`n1C?D@ZAO%i!W!xj)bag2c!h zq2@%U#>MX4Qz%RDyOctfe2$ei8{~Ws-)6XIw0sS01>_0~okH1oTi>-s3G#001R_U7iXj&&P7Onig$tA5C*rH)3;adRwY{3kvZcxK0^ZNL_9tFig<9D>Qi2X z8v+|_@+Y_!`ZDU2i!EsgZ0XIvQ?(kNN0=W!_}5xE23O6zGTv4&FN!%uOqL@GtujR; z^2&@zxaD4j536)4Ba4a6sVE0rHmXg*1E_&rGDoD0X9RI7K4-`p@grvz8(2*EpD>8O zks4@(-WZtNe5(kN@(a5U%mH~sG-%VaownvnvcTngVw959#NBk+(90$3&^WQ4^We>C zu7}$o`tuY0a|`G(T?B3%TY7BwGO@z%6*AeMjVnVk=KxhvHx18Y4%C>gg}FqW%BIjP zPgX_FycMq2QC51EE<&NS+@V{P)tAy#ujzmN8a-vBoAPER12A?53K+yinq zV*531VcJ8O)8b(|rriAjDQwtMrg!}<*~yUH+DgOaOPTkyHoB}zWV%QUaDZEa_}`}M z9eF?LiEM$ZaEg1r(fj+#?F$Zkz8F2J=?Qk8YeQ02)NSN2VWO>|#60%&EwT+!0^Ei6 z*L+jE;MjF`AQVmy&=mlhBEM#mP#MYHG-bqc#J3>caQEy$!;jV}C7)BcAS~hBB;tNV zY24JB?I)CNW`4uHqHLPwPDQ$XM0qu^Sr%W`O?qWBj8vL5#2SPYqwp>e2I!7ih`o&> zp0Nu7rSVuG&nC(%@E}@93AQBu@a3A)FPl6F49sI_+YI0>Yg+kHTp0Ex#|}dz)(l`A z1`Lgx-q^T&5q|%oP43e9wvgIcfRo6NFO{BAjy9> zktNa3q&C?rVNpkfQ25tgI@^<lZ;T*_WyhO5z74fTyVCWWwAUM6G&l?r>UE2hB24Ueqkx zi&lzV3R3i$LJEu)$?RK)xw3&Kvb2b1qym3=7LL1~peqOw-Yks6?+)?I(y6BqP5ykP zDm2Y)h9F|e7+KHKBL4(&Gn9`gB>09py#}aMT7nQZUMdW_=ulJ zD4E_7WO^*k$;yU%!dA18sBm&;B5ylzV8>8B!WAt=%yVT6qnJul3dQv2!;+;W3gZgt z0xUyFq1=5>GZ7t|bMD8A)(GH~lK;7jMYXS_47Su0Q51cm`NjWq^b$G_QAfF!2KT1g0vQ=uN@XJ%NsQri_5^4_Dz%^T2Fy7>aR7#brh&Kbh z{KkmjUc+D{c2_VE>ww5{d@*NOhK5DrOJ=xZkQv>7JLG4C42FHlE*+TLB= z>X)B|>MWO>D;P@7fiaQef85xwF+Q(qDN`a|S$>|>^!&JmG4MgP%v>U`Bu<=aHUw_wGDEsWH4LwF1I!`p>PC*$)&sl=xxKe{d>ea4i*?0%4#Ly2^oK zj8Q#czCE+;Lq2N+)taGV=V(koIvB&^-}uk)J6&stVN;ARA&(S?>gt3ok;9Z@+bYx( zd8>m%l5%?%@J1NA`&e8=3``tR}M2nDoJW$bCjo3Wb?M0v_ zd(tS^xPWyz_R+&g8we9Q;`g6ncbY3$hWjm|Z)p|9ixURg)YMl_rM=CS)#!pcTkG+} zLeJ$KQCS{QfiWPJzK;Vxi_y^(I~z&dyBJ4-B?0r6h5KU8!Pory2O@iKtZ#R@Yugeg zhO*2X?~w?|#fIupXr528$eME~PAIuDWw`B=2&jis^~53s#Mr|rkIqWe>BLD!MO7G(bx&%Wd95J(+q+dV+B`RRx>`OZ zlYu*1hcdX9SvFU?2Y)FSP^*LWupyBzBl_$H8y`>o&E#&*l^8|cmRI@j1t9F^kB9s1Ar@cE#y}AtM5P?E4e+ucb)xxdwP{3^ zIu%1-c(}-2&f!{1@`yRW6b*Wai%_08ih+{_TBZy4MBPe2KrbE;x z&0jNTS3>Td0J$uwe%EQQCwk1%+k-)%4Aq}J6z=c7vYbRsuxzm}smn>x&>^DNSEs%` z!8x3xj!k7NAi7UY?!)KN5<(B9#lS6=_`JKfzNWjsJ%rC=s=FBLa)|G5OyEBQyH9I1 zjR9j5KBY7p@dZi0eQ%v9kH)K#%79_QDp+(S?W_StiD5ooW!~wWzf?B#ZN3ufiBa*& z0t8t>EjO(i)G?JC4zEeUB02mBl3$&ah=Kg8eJ3=RDbhWOwMBK0ENu%`!dy@@>p-uB>sc5UR9)-!mZ(vX;J& z%PK|`E|~H5vJm{;Xo|+wXuinZ%iu!1*~I_r2fEU0I!b$ZkEKZwETF#erv$ai_vLho zJA;f4unt|AS34tpV^==A%y-pEQ>HV+SbFfK)@7c}(4EC4Fvh8EFQ&B*ew*Bz zS`WztS%{@S0rZ26iFR2=abup$klL?KJ9;9I{>j1VFml9;$d_|#Y?&AfODpF^zS|B# zXB8vJ#dbd(`#jWx(dYY_Tlw=ZCABGC6e!cXo2ucTQj7Ck0muFMUvDK&@;pR{=zO3k z6^i1Ey?z&I1qG6i<2^Wh#ko6+sM!=GP4nW2v&s=wp(H(_CU2XDGw54N2%kn;>@ggm z>f%UlMKTveKb59ybov~Mc0SFB4DPrU_F~tOP6kCia4}CQ)z8^BLMeghiT* zWkOrn&W3hW9qev7mni1PjGXNR=T0MiN}c0cuzWBp6z*B1^0B%Cx))TH8x5h&RTOK& zLdrhfFPmcs)QSg$2l`zJP~WN!|6W{`b~4#&g1@qm4l3I;`Ftt{U3G_6>X}0^HJj1i z9$We+92R=kAZ90GCO@5@-uE5fgd5*HAv~JuLphm9$?(JVJi*zPnwSvC2vsduzbTwd zrEgqa1@*WQg$UI7n!)sK{DyR!S(sq+0bcasj!_+gL_1}pgZL}uDK~B8!)DX|P8UMa zEEf>9qmwh8>jHPd4~;^&XX2pYl+qJ1=;n(j1=P# zun4HtY%|`;E#5z!Ec}h=aJJb&RF;bd}zYy4{YRl zP>jRhoK#rTw5b}>ac6`VeIUHhlv|X3nnK&AGc3Dpqqrc(RZ`IL7+sJ3ZSb{ynupeu z%}70`tvoqX!{|p2f?J6f#}4y^|7|)DaN>9Tt54~nWLUZA7wgqAYDsd%Gyc3=v1i$0 z$?H~6LC)?8!T)+oS%R$Ld-Z<$`#s1EmcBmmiyB&KFuq=az?}BdmOpl`!_%M{>`YE| zI*t^6Qs%sD8a>#`z=Hgdwfos(6)l6lpdp|`O=YTXD?uneZs<8YYRE+$jQ55rIat|P zPz6H5ohoH6#G0AEo9hegOM0lkHu>Ye^ZPP#L9y~C_d!tqp3L;6S`gdt%*Rxtz@Hz7 zfv5)NI}?eOz$N?uoG*w^*;KywCbKq)~^qg|7nmu@9*(759-hqhmn9+6DiM7d%l zm3rTz;MAfe(H<~8aPY!pCHLZrZ7yyt%+qcgh7m|;cI56f^k8Oc_~5)+Wpxm~-pjd? zc=c;U5neY;|IQ*GR=uRxz-OHN?X*P|n*X`MaZdhRa}-JK=2{J5$tu+kTd6?;$jEcP zuy@9mNwOptWVBUlfR;OXAzxCfS;?YA_CMI2?YbE*arIqY7)vWio>4#{B+f}9spl5* zpES7OR8AO&m55)dI?AZkewNZEVF|)O5DbeKtgMMETQ8(vU-74h1XJk9BoICPDpucN zH7#NhjKAy*!|lRiREpENaQ2L7n7@Ec2cOW-2@EQxk{(o|jBQg*jeb+J41>std%00H zki-y=;xI! ztU{d5f3~X|kdtgj2xZ?jqL}8sQa(vhr?IA#E9JvP5#JG$!adJ`owGd7A(FYHRvGgg zrpU;tt&ncGWU|U*rLK@xAQ{KgaFCfGYC-#m4i%_|}0weAWM;s}K(i)U}C zGj%z@I+9TrbW1wof8ujA@yn#TCvM+LG}dv&S-9I|&N7QurZF!U9S;^O?IMotAg3nn z$g~*3f1w?Q%MsylQ2Q~$D9@5({tj8bh0aK_LBkmv>LWp-*>kj$5evF=jX1;ZfvzKE z)||SuZV>z3QD-QwjXW*^ySbNN=;k>GeR)28Qt`cZ#qvzxUb&_Z7QQQP*7P9!@CrM3SK$cz(jNylDQV^WR?$ zg3YS#bjrSVs=MFbDGZr-9e+q)#5O&niFA>H@F09V_ zgzCPN^+cO{!5*Y7kRc^3Ha*tDlzR}KdRp#Fo=JB1Da%7qAjp;inW=DQ;>K%OLi$V= z+vhUY0dsnGz_4$jg?u1>N3BlRo?q_UxhlX=R;uiTx}DAR(c%-)G1URGj{*AikoD}2 zm(ggZy!5m+x`;QC{_#k0W>7gJkI5f>lDr&88`xJMMPm*XDV>|Nu5BpJP;x%;Tb--CS%<2@ZLUTBn(vPzbk;} z%OLqMYY^0_L9T_z^xT=sC?Xr)7(50t^PAHVA8$YSB~2TxIpR544(2^M1PlK8)Y+$5 zfSP2k3+vR7NJdxa)NNX}WBAF|4%HQ&_Zb6zRc2XY78c;HOeq_)w*PoXEGkGad6lXBg; ze&_tPBC0+Z$rx!kI~HMy-1}{a(0MRS{gNJ>r8y2CM#4nv*L+@~H}4-)W)SVr)l)T? zzS_dn7Dc4<%2{0gviMa{$WVfR^utG?p*k!W*R+N~3*n|Ad6lk)U86}$%);WqzBunW zIm3yb9TJBjj{Yy6%Hf~-y%5ju_IGM!qv;+|*qRMPL&N*p#7Y89^~Gd3UN58I7D(g;t6o6)>qwkyurjJ49LZSsv%)2ab`rJA7=| znl0#Pw@OiSAMcH!#+mY@nqL~d+GP**;aN`v!<<2}>)7Ny{U?r?KL+9Ua%sObpt`hr zOZIRSCp?1p6Mv&s5vsdeL2F?^pYUKSS)s`rr5_h#T$(fFW8;^&S7ywfWT@jP7;X8H z!~RP`wC1As=O#?!tvYI54vuN(pki391VW@bze|(=p>a;9iO=!xe}`{HJGFgs%QI&# z;4DdqoVII-sp}}`Rr=U!#-esfL1a+EQp`=^olP2!dRQH^ckUKJKA!a*t5RNdtOB7D znf;#JQz6RJc_eg0MbNU5X;lzZYT~8-rj}6T%|8ev@QEQ+=4v++ubu*2A3N4lLB@Eo;DxiQ0bDbK;~dX3hpork^chm z*MYN0vk$2jto=%AB#`PpOdYka@g-M*hZ{x52f2V#h*H6K4%gl|WW1Z^6s3;R zOVS;N{2)vacH2P3lp#W~%?fjZ2xge@90E;T^m~UrwUiLZq*Phdm%O<(#ST-y9M~qD8kPwN;I$$cN%z$nsozWLx@tv~ zE=ZJ`ZMaZ6Nal%Ku7FozB3WvBP7nsf@ldjScEWg3h$2Jn^RU$2KTw+1KDGt@ zhQ_5MGi2pSFQ0yEr5j4)=7frRos6}(Ndl*7nW(l#m=l5_^563~S|{VkIO9Kp+~`{) znORi3Ju7p2;;LP9X5Y1>L{C;eYu-56%d26!JnaOc3Q{MA@~3f?%vUFQCcNsDE+Mnl zo#my-J;O!9`oOR0hr9DF)Z^w#>D(9$!vrVals&$&T!3)_dXPu!!V`bvV``tzmdqA0 z6D~Z82oFS~f%0}Ha++qxIrcB+1%%}dIf2H9Cm4&5Drg*veKKEEH81a0ezNn>*DD<| zPWboOmn(l3#MCiT7ugH$WWq;tK)O-KXPEg^C~Qre7+Oz7s`tmVjla1BIe(VTG_skB zWXy?<5tP^1^!n*FhLx`u2w1nYR%Jc~-|LBI>*OBFBhsJxW>>i`MG>H7%E=iOS!5xK z5k{*!O@oSQIP4uSQ_T~^8Qp>Zw0-6~Hd!;S#`4$#h6-K@1=SAztGGw2M8n?Kj!!qa z-e2=^h{fG4v3YZ0lz!w7P9dOi>o<4_7X^(UYj*isX#-IGx`g$rA_c!R_d%z(dveWx zcmXz4JnXVB{EiDdG(a}c7l+=Ly55B|w@vv`TGB?QzC7$y1pLswW>X-op7@0LYmZfQ!1Xdkz?0V2gtrwX4he3K>{lr zhHBxAn|>`6+GD;sWL~Q)3f^UoMs4w+Q5?>9|J5?aE+J8+a{sa@{SHq-qbd``Z9dFI z%jRgsPT#*J!#N%rPxBt}x<_xQ^#dP5XH#hc{Btmm<*)Pt{%Jtvi2w!io%i>OgaTQ}J@Ql<;IO)@ZMBr@T%c?fyKDu+4-dC~aV? z5yGzaCCOm7wx7WLM4!yj^q{OPJ}i~()la%j^smQP`c&F>t{iU8IFJ?pYgwkaU(WCM z1NpINcShSS{z!MzHX7e(%4`%L{LT9Yodpkiw3hdK#h2!+v#1^F;yS-my|eO92@4XP z5YXb}g_(ji*m3j=cM2-R9E6B|q`$i)0%gEneOJRg4B){0TX-4^K|%CwM!Dpx~h=N2~X+Xh9^~pg^5Qj-U{i=Gq(zWG;a~b47Da-Ny+mZ{g&(d3i$} zqLPvMa^J_X`drL67wJ>%XluR~7N}cfKnno535lV_g76&DHjDL3_sA#obq1FiC;M60 z#iQ3Si}dt)p}wq|u~pl%Y_lbOB4vN<+$)|gv^E(h7R7}Lf-OrX1fOx$MHmc^rEEn{ z^2jABX3FkM2#}vIuakiMCV?z6E%qGuf2xr-(ghZxt;yAxRJ@?F!MX$@H z)3+r*J1&ab*Q|VH7nxFha$MB3b>M{~N48*gxNUdwdb=S0DF2-X<~vf&=I>CDs77Yf zM%@FfV=BO*}IJN(^AUKEc%wy<>g)Tv~fr@X?{3xp-Ac9THwg$cXfl6_vWBbZ6QC ze2o!8X*lQvI1~C(hc*=~9*DB%oV35q_0&XzIxRD{()7|~P>;TDPmv%m`z=vlkL+A{ zT3ej5VxS|JJ7hgTZ^jTs%)aWfgwxWvVF_h!^3>1nR&Z^~QSeqp52q)ctLm+wtRYtr zbazT^LU=gxMq5C>iHgxi)yndMK|~=CbO44sEQx<@JNY+P;?Vz5n$uQvw9%_Oh(A|< zlc|ze0zHH)m~#ZKIIJysXtkUn(y&F%G*BE7t+U`pWGc9mq+hi z-Vb}NKyYTIPRvwA9iAyedCZ@VyWdGw%33)ZsaOD+M8~dBr3!1wxI+8PNows;-xvc{(D@?055+EK*{D2}MZ{VBL3$hiK~SO!nk>>>P5ZdvM}prF6AV=YA% zpY@X-rybLXjB%3pqR}P8?TTM?`Lf~J_!sI0{?7z{{h*fPA8fKKzdewlBtFb7PEK#P zXQvIR;8r-`^c9!y$TFa7$lwNr5$p$xS@LGNG|Vgw%`1WCFu0(w=a5Tf)CO|cY?fTs z&)w&%^h`Nvq80n6@;bc{IBn+YN=n{cbU&fskNuUaTC9%jUmk*8w-UvL-50A_miT@* zagoqrQhl_TLG5-_{qrcWO>oeKVX?&|rUQ3Syu)%jh5vc@FKVvGy7 z-v9!2CZD%xK69-cfU~QR-rVl+G(6e+DA3{ka6`7eds^$erf6057Ke}L>jP%_#yOlZ zEerAqp)-*rbwlYkIU-Cy_OBqy+>ZAmWn_>_NxIR;2MWKK6e*w{?E}b-4*5x%ARLRA z8|8$!q6^~SBN-Y1Hr(PS5|n`OS>1eCSd{q*E6OTj35{&AI!fk>4tY$2D5Y0|2e2>TTwKu9C(?no$O`f&m)xBE^4X(i!CLmC$)U0o-&tANK_+apfOGqaBcN|}O> zceMeZTG~g-OL6-zM2SR&mfPF{z6x8}+{RHgBGxSPwLUDVeZmOr9_v;w*CxH*uCq8yB5g%r>o}_EpXWGwA@kkCL}A1YQ;^gCg4(oEq4nNY7_hKEwd7VqD)iV zq27AmH(z(vC2Zp$rp9+5^MGOhO~TAjyv>7^*O^lt?kO%EK2(;ATvjpEI&8 zlIk2CNLt^i3i>@mbqtIgDD{uc|#^Pmg#xHbG!zo%Hr&2#{e zVt*az0~pSbF40Ty@rm?y;Vd99ct*wo6a7}|4d0jcr=(8;H$`;JbzaJ(At-n>jQq32 ztutJx@7Ma@c83uB(l)42U*yh`EuvB6-}RXgNTT`x1Ljcn@Y|fYvRh^(*wI@1F{bhw zSFchT`NX-3KT0>rg^tbm#NnxyT&84LU?OxbJjdFI$bUlpXZpk>*r(OMF26x`X~_=* zIKP+Y0R>3TK*gJ zb!XrA1`3xs5pYGySQD8o@t)>jLV)tG^Esxk(08E?^>%V85-f3PK{m*_>FHK(S_!ei z-Ja;gYUS!;3No@i zJ5XeW2>!SqqIP~_jpKC;P__gyM}AetHgXb8`>+x<-kGj^NZY|2T1F1`oWM2Z6&uod z8nn&JW}6P5(9Uv6^G!I3-)WEk$Ph~ygoIcf^-y9;oF9DjIo(SHr%5(Y%%= zCC<$sGz6>rlEh;P;qxyhBM}O=N2h;gV-(y7AK|nCdDNCKK&Qr_%7uA-6yAxERJ9Tw za9%bc1p3~B1{eVUPyZz_r?iT`>k9aSKH z%i!*JM;f;z6DF1Mc6sU;yxm^i`w^)>hZ6irk(!AW-I~g&CeY`-s28VZXx2qqM5DLX zGN;Yv7Qj@nSOykA`dVkV_6&c02vE1j4@>1w0@!^yoG;xtY6wn7=H?I6TArE|Oa&Fl z=vyC)-5EqHgz#zJ0!aeCM8%eA1-H&v+I^qy120m8!TfYSi{Ii9ySSp%1j)W)mR+RQ zELV>T)V#v-BA|Rvc@sp(NYA!FhZQkv*(e(n!TutZ&M4lx1evd{F7zI&(}$;VKgy?{ zH%(`aVP=i?(i{K@L7x5Ook=d|t7@d1uDAQ&t5`MikIwo~HH<64C4Ueuh6%hWrls5a$JqZ2&$!CjPu2Z~esY-|l?u!;mJBXd` zAm&s>a1C*a2gOCtX@l3~Ql zejmyg{LS>Yl`Kq0*g*uoQUw#JH#pDud4y|9ive4US({%hNux?v09$IpPl{r>|LVi; z>OSgG2@2&$DjdG^JJBHa5FnvZn~)4qsn9UC12-#|mP;bkx}Z^D5YMdFD+wYaA{uhi zD!vi2`u>`<>$s5qWyl3CVoBddYIB2(&GKAH4;w@>j9#Qz&7=YV2+qv>;(Zkdfb0N` zVvttPoC7rNA3lEE6C@S?{~FzA$bli+t+b48U4HRxWn^TO|3G_jbwSK-_-!*8 zFO%2v%ZtGN`*2`b%~LSa$9F#OE@%5(ZN#kkj2`D&cnnktF#OT~_dYLUizN(2MVSU# zn8s7PzOAiwIY>+E`{+B#TRS+giad|6dcTrXxG{R=oAC2k@3z2cvM%>H>bQ_iO*b?= zl$`#wW1!eyM+NFu6ce9vc$M69{8T2_Q`d7oU8eJvd8eXqD?IVWV>|}pD8roGCYQi} zPiyz;@83^VD(Kwp3>c_?|8mWT3WJh=JYJm&)c6WDCTm_l+jTyn{r9cwmL>Vs<+aYm zxu~!2pR23L&3G*6e;3-0%?fi{Zrj`0ooTzS1zbw}y99!u36m?C&*O7JJw^Wam%`;6 zhu(+)qy{d|8}WbF<}|M3tmall7BjHeVH0v1_x)najQ*%qDY^BV_08*obM7xx3i1a# za=to|yz|+4%8y#3@RZ1~`~U+#+P$bkVc>Cn)#U7S9iGB7cESwe zT(h>*o~&XOv(oCJbda$+>zkKH9B7PHL~SKf#kRJfQaJ zyRJ)gTq(4t9+}_`tLL}mI>DC*&5X_prz9oe;^M00hfkH76mM=~nVGzQ#e}*AqU@LV zPUljeVAnD!F`ODoT%`ywTf(=dM!4-+h-jq{+q-ijGlTb0p)(b1}9DElOFEPvZAAGXcLA2$c1$g7_t=Cv#J$OL}8R&5e3<2q~}@=y>ig263zUpZY$q$lh#fcpx2>#m@4zIh$P zeUe7#G=5YnoImicEzN`7#DhFgT%CeuxfC72; zLu^I;dG6Og?wtvv+#0HqeGB}L%o^{ zogoC8FdHY+b{1s#w^QZSJG(oRWY80vlTVzjFlNI_GYwI+yS!2f?)RzT zQy2&<;ahE3*4ONaefDe?10-vVUr)2y$a_bk-=U(cn{)(1qvRXP_-8#mS8R$0y&lDa zc_qxS1xQN^+2Fhh){zlE##ai+njqFSdWJ-@59+-Ow%PHUnNXoRRp4)^M_M&GRPSI^tkpY1<~ zu&YhA*}I@@18WaahW`&`Zy6O=6Z8*G2<`+3!2*Qf?(PAC2X_m>-Q7J{aCdiicXxLP z?(Q(`P4c{Z-n0AP58EF&oS8e_)z#J2Rllmf6~#Kp{H7LJ>7j0P^_B?dqcxqJectvy zyuYK7=GFiUSQsW=Od0nb2NjaM=(|MT;SF|$cl?j-y$~Yi1grh7(lwI6nby#3-T7%gX z6Akd7AVI>!@H<%HwQ)XdOz)zBkB3JC;Cju8w9X&U8-{2HrvBhNW6qcykuJ$@w|y_u{v4_NZDGT<+5TjI>!jLnUEnvPW$A6l*wWoC0Lk8; zH zF!UWw-)IR zFB1YUmZ>b`CbhG#C8ng(`zw|1*$|nNKV0^Mv2stjELtN^1>CP*#;3Z6#JAK5s4p(& zrYtz$Du)r=Jio3*LF4mTFY!i~bFUT~-UX(+9rO(PA)hR_KCf(rta-XWre47|_=>DJ z_~5YX@Skb%EcN%c-gO4r2^=pnc}(xpKxXJiJ1l%DB!Z2sk4Yt#%f5sR?hNC=Max|` zRsC(C8KtLb2X3N0&J`kNNM&Z_Ih(T4`+b5mj#{IO0E+KjcfHNeHJl1xS1QY6Y0}rG zb2*%JE^lr8rD4LFZ2ppGE_Zj&(5)J0&zqC7I5j?RI5Lnq3o}`?Q9<+;tL5pUkA`bs zNUg)|HkQpun1TB9LhN@^=4O!^R3_KCR66b}I}J09JB!sAagt$1l(FWsB?@_1FD{)T zuGb*1WW6!Z>Z#H0RBfg=Mu<&4Ukd%jlO>3iFTVr`BZ88Ei~mE-2z;%^hZv15MFmV5 zp6MhA>kG{%r9jfbGhS^coxP+RPwHmR zQ%Ri7f!yp#r%4C{F|oqx@wS)1eyO%xNoDB{{01D@t>3@P61fi&v9#QN9~kDGakXdY zOh(D%ueuSx0U24yjgf({{69*OSJ6|>JN$~FmaQ<7hXy6}-?raTVq8k&|2A^GD@)~l z&=C+f>C?R9xgEXZTjzZV5It>AzV&?aC0TYj6_Xlv1B8LqY^_XzQtMz?hSGbeqp>$GK>%k)k3V6MEDmi5MYC04q--IU^14Nln1mE@Re^r9p{u*;AUAOkXF zpVQc#t6au~L?gI=KY6vcYa3IPVMBpHY;k5AGbs`b5#eze^#szrXt1D$XQPRtiKAwR z1E)*7sn7$q(&Xf!eMIV*VN)ltj*>k2LM2&&4f*GJlzQe-A+mIZGHb+Fl;B8v8$^fIpOyaOfqtCdbUIRmfOM;G-!&tC`I zA#*z2Hz;uDs8&A88}Y7G-)%Ui)2vjb?)G;l*PT*39=FQ0*c|gzTiS-#9p}gLfvci4 zFgQE=)7@kH#ys3zLUDPXH=V1?Q@Jfni?iA;0bTXp8G8 zoJ$&ykLa8mP{dYxx5eN12ENrPvVHq}9T%L->Y!H%aU%dSdcVr)_(S#fz7TRrYS>z- zc$QK%{6tKRr@)h3VQYMc9uWp@B##)}NeXZOb)lre%NvO!Ow zt@Szl=5+UmFm=OK8QX)2&KJ0bx5`w0{=)uO>rO|5{RFAJ?oIVd)%)Ej`&F@#i8r17 z1n9t_??_rJVOO95HV?OwXdoHac$)R2RJ#7|pB&+&%wiZK2@8q`+T)<;9tmmgd#q*k zYxc-r80)PF4-byUMt2se&S`w937>dk2EmKt`KG0W?#xBH0I_Vz)bc!B2quZ^Md z#j~w0cI_@*`ykKuWrI~Zq{oxH_Bd*{tMkdNBQ}eW@SkoEi=xVf>rV6?L{?!4v7SAfH6SO;}+>*?~R zl(9SuW857PbanKc2ypVjctRfBWUtSmLk#d`A0fSl_qOh|^4p~hu7c%*U2pLpjeCn$ zo*|4U)?Jag?}l(4$D2-OvC^n%sFthD=OxP9sgRah?Dt}~MBMLsMbq6IucPVW+ny%# z%iFJGnCJ4)B}8QXyYnB2 zNBF3!vMItK9I;%a-=!kOi~Jot9OF>AuRlT;#E2mW{C~WSx75WKGWMK|oqJ#gqs2CH z%UYd`1%UPj71yZA*l@*RomgWXD@HYps~ehFfMV!nEbi8LWUzv>NnyV__+O{w@K1!2 z0+h0Y!!44dqw}fy@nPu#X6LxEbG+eoT}1jy>sRpJ;bCCnZABY|db?0TXeUMI-Cf%^Wf%QxG ze7Jdgb&AjC5LzD*krFMH>~Q73MTiOk;y1lKV-k!6y#Zygm=dDKDaW*f`;u$6qS6`R zfw6!#Iy|c+f`Z2xlN2ZN?~{yLz&6<3*_lrFehYnls^5On<4WLWRJ-JI8@gWOi~y(u zpf!+ZWslMxd^e8_3SaM;yk)|%Z^Of}Cg@tMK5cQh-1f`KIbCg{h!s!UBS>*fSIX9G zG{OA$QzJw}go5z7)ZipcE)(a}*gUAH!kuP~qqNXR>#F$N(U&~t*m zEY_#?;Dlc#~fp;?-e_V?9&JEWnFU} z#u*O=aaf6%?noxCq!_|}T@qoFes@;ZGnu|Hd=K~8?~6y(i_!2H8LPG2c4DWd7^j?U z5>uEN&N|2+o3q%LZ}N?vmzCnMf?8s|$n;dsyas;$riPgKN@kI{@8>~6QE9d>&b*AWhQzkgMmgMK zYQ)8*&WvA))v#A<4aZlU@9skD*Bt~cC&==1@HlKnez2nbot9U?*ZL@xciM@1b~j&y z#Tvull;P+7)W8GEHNPbnr~$U<^@m#EegvOPon2kBmaR8L0>5Kkdysc8vS1e;A62~WSFg0(uB{D7jFl{2Sio|` zft>g&Z2E0?i(qpFJl&J5R^&z!D#&YYyb`m4b;JOZj&qVh&lhh`@q4!QL4z96N~__S z{k-JCLZzOK4I>~gMu@I!wlQy7Ke+@Vw#k`sJ0FhutZD~CvL!-*jCOLO z1+r!Yv)U$i1cHSLU)&$~E)OEVL~hJPNdV5%LY|b$*kZNisOQPT1^Uj;&MZvpQ>P$6 zEQ^cJx%oJ97o8J=T zND!bJ7kn1CUe5-IiGgT-{{aoREI+kB6K=L7cp(}cTIVaMsR;u1nR;a15|OWg3KhHw zz5u=AKuzb8gD>UOUB(GJK+6c_y9ZB|{^|Uo#KW+tPLheBWogv@TzoY1<-R&oWm(zC zxU@{F`ACT$=Nr1{WSZ)&<9Sj?b>^ieV|D`*Xv&DV2U{odN=i-U%Xz*qx9@<3LbuB< zMLy^#YH<#!EVEWgl*3&VO|Oj4u*U|*`w89n^dT^o!9*d}rg*ijO|Cb4*w_+Q8YXG*&x#j`j!UX?T&Fcm@9J_ZBqpRc>behJb~NL( zv*H^KQ`UBuviUCP+)FZ?1p-av?31cvqtd8Q9$^O>2sg@*zX9Y!a)hb66$`kmnua;iu~Vuj7lIG{mRyp7twee1Wnbm`O8 zhsU+F4-M681LQo`9V*@UhjHe3M|4OB!w*kvt{J_nN~BdO@en9F7^HWoAY?Rz5n5S!<(lD1rXsps>$$|ol<~vVM1;3jQ)BRyD_ObAx_X(Gg1k59&3?KuxEBR{#UUo?Pqp! z7Mjw&Fkg8&xz?5zW-YB9%$@D_Nuw5Ffdo!_wJ%?Q@>FY)*)UP>(2YE6lccre6|QB9 z?SN2qn}-Ku;)p>tpa!g9SluHT_+&Q;ND`20!wK);D0tf5B|8Dj99s6f6I$uX0zG;`}TNpV!sr$v^jb)Ftde=}Zr0Hk~oc>bA z$a;DvzBrw>blIoV2;GCZ&#<%7M$~Fw^Ib>GTKANF$ug4P#l;RQmRgy9I%kc}bUKAA zUlqF9!OS|Uk2sIN3be1%)VFzr$(n&gb=0FCC!>ii_NM3Ja_#O5J8OTsMH%)O85x^1 zscm(+PT;;cbmY4yWd7_Iu2Z`xE-pUM%jveyFJ~jDb*j3Qn&2dkagWyBKf$kJTtgtC zRHT5pr$=^@*G?5CqD)G)Frp++d`U@$P?L0HCrh%?x!)vK_Kn44@XfCQMP3NY%~Dc` z*FGD@H52LA2e89@dp{+!U7&KshXZY#lW&i8gDMv}6nv4o?2iYN1i!(@@n4c|vVuzC zn32hpr}^F&7y5V^Mi;w76(@?2P2pIlg!Xl*M>x4JlvXWtEol_oPTG9W^mpeJJ2BFT z>6Ko}-7z7=<8qk^o7a}%(sDXYJbb}wxhCo>$4H7U-Ol6!Oa$ukwrnub4J#%HW5Sm_~32u;nu>PUH*trTC#Z@{9A|*?V##iB$oQHXg-E;^6Y2$hMG4x zqD?dJk5GS6)VdX~B~7q+1Vp;-*F1(eWsmhusHluLM4Au@jPtlH4O=4vz#r{x8s;|u zbKhKVzRS`7F~j6DXm`<1|ATQ6-*HT{{&u{YV^J>M%d2B*N-nLQK%Its;ApDW)x;!< zxj-WRw9$DFaQhlC0~YA;Ghe4kEHe1Uz=~nLn@a*9kc^-*_#n1w@~K7pL~>QE@}^HcSYq?Fps24XPCnZ{+hNUyL87qN08m z>^o$OM(|lx9G+g8&XpxO_pdUel9G|wMVw);x9%CHdrE>p9jjk)Z6+@w*u?{~vQb#i zL7nHMmZs9ptD{{m)h5m@MO zMoxhYX#8_-x$JD_Bm61rz~N#!u?oQq;|XWo%vJ$ewE9FNH5wAsB%0koYOcHg!23aPhdX?`>Feb1TBK!b$W!a7swYOP?Yh3V!1jd#zBXdkj4+G8Bbp&p?qs=u1cv{Uyi>4`)J7>J zce0*oR{tC%PDnHbcHCBJ14CCKud2T zHahPQ5)L#gHUO$@cRUjt9w7Yt0ie%GPeO8QD{8;f6!D%0o6muIn8n#O^WU>pgN=Bw&41z; zlr%KOYfMj)?@RF>AMgJ+^(7@G1qHAHxQfXhfmH;a<8k zCu?1*B*5ZYUl=4eowe3xhclvo&us(%&LIkxD3ay$?hclA(9ck-Th&&FlPr>ScprWo zng4&@VYTryF{Qn8mJTr}j7ZtJI%UO-oM>ZFVXc$ZHg^Js-S_ExR*sHCEKJq7@C0-BX_7_48^N(=YnzpxglW0VCI_AHHuiu$~C?qyWHUfjK3^2{S8;vn2OgK4Nh7;FfugD^n}e#IdYN|r zEgW~EPk+19P|L!%ehyM>&z0_fh2=Nb7`1r=1{+l8ipYd4mMBox;)IoHhyoi1anB$^ z_X9xd#v8n4-X{k-6(YNt2?DL&pB$vDy&WVtDiq{Xl@7WtMN7{N*958yvlFB01prmge7L0Rrkef03JM(N1in7o-jvp?8fTRzzb+BG3E0)kQ_eM>%YeOK!a;32v?j8u(> zZ|%dIQ~6fLxEW?1oqXc$z~pto8$A>)LYmkSw47XyCxFZ%z$1o6oR!HHzI*o$i$>Mv za>m$3_J0LV!^m|gXLyyWy@Md$$n?>R*IEK6l99tyr)-n+)tR-_0;XfJubyhI(gD^c zCfeDubquX_G!t!9-sk_X{VxJ|WqC6bQo$kW=*9?6QUy1`?D9$8)j_qE2b9ivVNvar z15I_`Iz!m4^GM$P(GSdJJPEZ&^V0+X7k)3WnG3vFT%0hD<AR@@lp*hjRY)n4?cN+q)+UPOG5?n{QUgizUkcj z>GAoC{`*9Eg%~N#6=G{gV;IkU(f9SuM=v}3>4W&JqjaONu5-P!OFM8?_M8qde2Pev0Ab`+T%pG>jk?{16@tH1Zgt6 z{SqXvX>egtN{mg!)pw>A2%~}%_-*1}A+aZ2ZMfQKy|tU4?wZEyc9iPK_i!BI)78bc zE|Sd%n)pQS*6l2TkF^$*HTvz-9TFL&qb>wsT;^sg8O>};GmYC_yhnxV9Gr#M1w@94 zk+J>$(R8VlicUsid|+sdmWHO)VJN0_*%5FMcefXcE?DEDFY1>1e;GsE56NM`0-`$m zD=9koZug-N@1!Ouv|RFAWft6tmbY&)reLnPqMRC~y|RMB{nOKA_!gOr-NDr0e1AOP z#*#g0l3a6$_BLI998Ubn1XYX%B?xei1rG~#X*xC7a^~u5TVjnO+ybc&;4UU4CC%m9 z<`)cnU6PXUZ*{%N!pC27+;Cp|!H5P>Mx1&l=kkowI21VJAc310Z@KBM|EbOmVBb%6 z3r0EvT#oX-RW^P^<;WUw=^kUI4Zu`bsrU5!E;luTT|a$lzU;VS*L`!=`Z9+w&f99Z z&Rg9J7zIhVtK@k!&7zz5BLfPvE==(2tWeGErQEg&L=7$X&&8+__`~*@ zga?Sjq-hizFRxOSgy@vUJo&*IxOc8uULQ2mjw8hb5B=KcySB3a^G}0^o(s-_Q)DP& zjjFJ6y|q?KLw(G$Iw=`seZHmY3h3JRGHLkN(X$it^ikX$VUdnL@Y6%Vakcbmc)2XS|NLnvkH3 zUY*b32vjM6ZNyJSf)Cfa6V#_o>ok0b;9{=$jAe+yCBB0w{odaM(59^Y!$W2Y&#sp5k(v=z!T>Z8p98`0r>OEjs+|{Jnhr8wIq3c4Em4C68UG7c*?F zF7qZ2fD5!pCC6*K5m7RBEgIe|)rQ2mxmY`n{2JcQ33i4NnOw*0T`c1h-deA#YT@tU z>EXufc7ge#gkgy0B_e0+>r_1H!6koghVvREu9-%zG2-sh8U6ppe1RMn)snj1LBg;G zl+FKBfUUvsQEU_0QT~J3EB}jUC(D+zrN?m`FKsy#JX<9r!GGu-Yyc&aIv2H}&h7 z@fb2m>=zfJ1Rh5xcNPucCg9D{WS>9tfVcnelmFXWZ+f6{GG6+}Yjaqgqq--=#9U5S z;d zpR|e8AbIXyJPq6Lw7=Gd|Lc|cBSa>y(`)wl)KoRI+D%OG)>-3`L-pYEV7vqqV|97U z?c7Mof2`feB1CQSmRbf&ON{5>dp_50<{A>TZq?x$6Xu{q{_`$2+B;bMEDTNpO2Hl; zvXSPP@DBlsd=6(z!M81rx1sY|ZkI(D+4H5(x_5DkXCt0O+|)AhX6=v43f+D<&( z?;g5>mBV9Qkg*pmXX&(2@Mm5l(%ZEvkN@LM8c;lj4gDj|r=MBU%ANFxVg8i4QMcE3 zclg%58v}Kl<3omd#c8X}ZQx7!0P*fR zeAkwl!v+Ti+p{L~3&VATy%dtaT-k6&gXVe>*V zZ3=nQlczm7Go?o!*vu6uNFI)333cXsrxAP)Nw<8dzwp*b?RKw|jXWQJZ8aEbJ2R2( z;r&~j2tWRPXr-KoKqW)CplEP zKeL#PRua&NIu#v@=#wPB`u?Rat)v&K%gc*DdQd^tn9yo%xV$|%E=|le2yRo{MG4`nUlo2!h&<>*jDxlke=F}4R4)*O^-~3gZr?geivz0~?a{BGLWdkkS z6C^&G`gM&?iQ-vhh3Cpcw!MK^XdZ9RY%$Bu*EHw7ZW-e{bd**3JgRlq4f+%C&0RaM zhb8zT8ZNo+7!mRMbsa$_=7{pt^O)n`p7~gw3)zK?&vh^)&Ss$^)eMyJbTC_IbQy6n z{-v~~nMWnX?(W0Ciu{oy?N>t8WBnP1&BNey@YP4>yp|P@)s3;$CvBIT6OL*N_46U7 z8u|v0D~HElf2RSeN#R%d_N(1v)nu=GD(ZWzI8F4RIcu>}c40vQ;YSLlBgwtBZd3dp ztm-C+kH<#@ZD252ts0H6Kz?j7V3zV}uZ!#D_R858ZswVDJt1QMb7wkyYh(Bm(&FS^ zbmPlbtE-rdaB^00N*!|r_*|~<%T`_XZyx0N*Z&6#z!VG4zdK$O%V1S19yKZD;pSQf za+LnGUo}0;(%1WwTy2JDUYADfdK{X0Aw5WZ3_6-h)QtjYNJpiIBlyPwpw-Nt+ zAm4w25+#j=xh>OvGDed;pD39JolgqeGHO6d8%;6Gx+V^p))1>`%KEOs?n@Urtnp0L zTf$5(Tf5S3z0C`!l&+9^)i0Byb)*x=>;Xj7fk>$IjQ;MXU3y|(!AC+bsM+cV@6*lZWap}3_|?$|3ntYTA2+J2VVgVBlt+_j1?DfLf6o`M;Rqn2MmQ=! z8_j0o7QW@zC z8)rYoPH2k!gHavdqkm-vVq%59?*B8=W{s5DKYcH%aCm8jM8q7Yu`_C8Q=RGOWhn0q zvpf>p5HK;VVm>+@E{4z)P~pKnzPFQV%Y#&hh#2O3ai=ls;NXi2;UjE`ip##K%Ud?YT0=n$xW9UFgHe#sjPcV&Fr4h8F%DkjP6O*( zE{?Q+I@9Wb5S0}(0CjzVer?Gr1cEhsmW=fJIr{G!foRboyYd8w-Ae?=DrnuhvW%G2 zvBQP&{-Y4iyW)Tza}g?~Ws{4?M!a!U=bf`ak68BK-&$eU#^6rZ^CF^ARUtrONn`8( z(3l(?DI)tft_E!ZFs|Bh^`VsED{txb{2X5%Y0ue=GH&j&!aqyNYZ%JA0&X%i{`|yX z?MJL@@gYt>RnOf$nW-5xQ{_7F5-P#JBQ?08!Y2QWIW;0(x$PkPLWs5I=@VJG#QKJe zciqb1ia`zk%ZL9cMg%(KU?dXiUYm24u&GNuhrh^%`*u*CiO(zB*+csL^6c2yES*jB z53VwLZ={fCJo%NJvkX~Mt&V4Buf5j%ZELnriAy#u@w!HlBck=TNDRzBE>{T=n%Rum zib5qz#V2Z~CDJV=qwD8%5tiZp8+_M?-zo8Hr;NC#9OM^OV2ZS9FQ3u#ikwaQ7A5T! zvBFsa4>trA@J&H0!CN-IMZTftir+<%2(sTxLPVSCMBivY(PVf{Q( zBp`lcs2bSBGB7|wRj;2sXmD~mANITS_J6X;$P4BLP6~szp8skS$`vb^QPx< zhwK;2X1BxuflLDV0Uk_D1^XH2zrU0U8@3loE-*N$6$r=>1iq*h`okQf1EZ{A<|?A1 zz8%im&`S0+Z+752`@f9)^Bq-XmCx57lEhu9QJ4|&whdmu3Lbg`9V4OR7lD=iK^^+U zsg&Z=sXpdC>uFM`GJrW>(9uc;0%gzw1b!o+fabgT%r*i{hKE8uSnZRtEF zy&dE`{Uu2B69ps{29BNVrYnrC;=9@*hA`t)@9wwlR_LlB#=)xr0-%qXz$9|Hs3#5X3Ft5o&IrVVV}9rHXL zd?<(@Vfw)S<1Gku!%qn5wIr6?zF-P%pTIezVSwsyQYiD8HCy$x6z6tKb73B`f&0UO zprEW9>EMxyKlK-i1xse9H+11=yQAYW^7s>j$9ryDbikS%`U9-1&328{(gmNCotjdX z7T7};uk{QpB#6D>lT~Mag`K$m;EJ1lU49aaar3N@8>Rng)-6NuJQB{shDe2|X8zpY z;P&V0k|qgbCDWLA|%Q(d`->+=ag$7gUHT;o3ssammLncJ2mP$^a zMtw!h*DDZ#wGMdGR9CRbWb@wFxlAOBgO1m~1c)cs~m zo1GKr!~&``6oNl>Z2dx#`(_8)T%p2JvVe4CTuyf;cQ2lFhkXn6g8mj-<>VDDacG>+ zBa5>J{ls`ghL)$cPfc}*J|#$klE;hbhT|17;i`R~Fgg>m64+NO5D-M58POp>JCyoQ z#r)}tgiE+hj>o${!+EsFsgb8p{B&y=-Y$i@OehuFx_#;4qU~v4j&0<6e>8nZAC3lF z^L;NOz=+t#$F&#QA?(_GU1VF>U0U|y;CQMhCCZN}T=vV#$3H+AJES1U$g4(F{lvrY z_4$&Y`QtQ750`<}Km|u!)yknSwSoKouSLX7%Bj~l5#k9P4%01sizFJNPBP#W(o z@p-(noSSC}*I{_POZcQj0%N3!w@%T`n2ohA%HCJRHp}vf<^xX1aKDEGU%nE3yrZra zb*`MesAPd0-m#^5_PbZ($Do?)9U!1~w(U)mEs9t=={eZN_P*_fZ|S0^`%bDI$V5iN zpS<1K`xrf}$uob@MImuuGzeqqc7FS65;ieqG^YG~#_1TkTy-1I=lOKoC6LP5L;7*i z`uq5yo@aRu{`Fq6%vwBjE7L1fA<7F!m0|a|1RpkMfhSM-Cw=};z**K~>{l#k#zahw zE@Xz!?6Xq2PR1!h1Og$m>tldA6sPo~walAfTT08UA(I!WN^ECq z;CyC9h_zpcr#Uk#75L5@lO6fmV>IIYJDq8ZBP87@=h#1c?FrkA_0-A2!^a)dZLJM| z+euQLjD%M|4lc>f)o;H2*kZF%K(p+~;p$>q%M8H-6veE0+Fjm@i;a)&dzz0K-!piM zfw_CVp2ad0QJX1o8XF-!0#aS5Q1d9!e*PA2|Bl2Q(#G`K3*VYKm-*(+o8DfM^43;A zU5?_dU+Ag2d*7lLp9iApCd2m&29C;GO(tQk$kyqBZ0JQb($7Dql1yxZcIxWhYd0Eg zV^h%$aRjsp@QgpZNrP?Ea8F`fDN6F~FxUG9YU2KPk$qjct%wq~80%?@;$&!b z3zGNqOJQ`pS3!bA^(rND=?ugHNs@W^&9_shKx(K>WlXoj|pb*PZIIC`@!C6%n_8m81T~ucONETdpG8)6Zs=!y2vp zIzX@0jStCnX9gUNi{Iq*_JL&LgNT_{>)~4|&zG5Y%W;IanR-MOuTMFyBrkmkG?M-& z&yPRLar9l(sY{miWKSP3Gj^}`2{Fnz%0s~!SEX5IyFLKd{M28}0>ANt@{IRImF)!A*mrdO^y-S^;+h&Gr@S*)i)IVIT zDJ|;pebKg^Um?G0)}%ZztUW(bKA;l!oB@)9w4OVC+?QV;ujVaJQ+cw%@x{7hE&cJY zk1&No8SUD#)^OGWS{}N_iPEKVvy74$ttuCLYmfXKQod&9_QrVwT~l(KmhMPvjFYM` zP#~Frf2x75Y5SfaMMHC*y+lyc7VRfAl>6 z_%w;vACsQIdF+q8tH3i{ay*r*uuh|h4QrJ+9xAQXAGNmTx2vF&+l8wIj zn^SR^aFYnnq(;$Qn7Mphl0?!wDTp&W7Lo&8-Z2TLO?m5G8eyBa>o(CH>85OdwBSL> zB(T@wVq#DDta9cU_@00O>%3VGKA8+zZA3_JZi}Io8okFcY?#Q?eRcVg*{w}k?{}aG z=))P^qtBjsTR-?G0JF4eQ$c-4I$&B+lP!{9|JrZN1df}Me`54}NN_k}dIwdUJKgfI zNX180UU4l8rz@W3(FO3pdcBj_vb9~srLp<}il=#5ouKq)hgap<|9V3B4cNnu)3apl zc{6`qGysO%YP}?A;QYbTogAsvWToH&CUDwWN3~8e-}TV{VYZw;W}-a$BgX#G@Lm$G z7UGb(nJ5w;=|ofE*R`l|0%)&c5Mb?Opyf1)3-?*5MZyuhpjwt;n3JkG>+DcUXAUh_ z=6rIG54zMb{nBo2-r%Y4L1FwkdGa^rVX*|yyj|`G&Q)Xckw(%dOx?0jZ5>i3;*pT5 znC}wing;V}Z<(clZEJN^*G?8nP;0Qd+jN-tbW7miK)~*_4S==AqU~!uOLcJWyOmG5 z;I_6k@!TxEr-aQ|GT$l3E7oM~2qKcedKjb+$_<9CGq5Ffn)A=4R4LJvGQ3?w+e-tt znAmWvk&P^!>4GCTh#`JXSwE)%J~{ZY^(( zV1dN=ya;40X}b9g`p(+n&yub@(-t9d2p%`5`FPIaj@#SX_!hO}Raqk(jb`0tze?ge z0&Y&bZ?6%&4@G4RrL1jg^@_Cd)5r&o1`gpf=^kH#DU1c?s*KG`jiJXZU;8q^dUkMu z&%a^xU%|o5$YGv7qlYOm8|0@9jQ-aZBSUV})pDd;X0uvjg?+=aViX6QsS-3O9HkB^ z=v2yzsYuPP>jIBZ_9s)uXJDagU8S+bV)F{e z&|l_6D-kLdGnZ6>xUqnfj(_VbmzEvMkO4K-!ky;q$toV#9DFRwYIIE!5~LdbK^l7O zgMNzKSaN9@Zy1pcT`t(ndiumx0^x^p>4uB(0QtnWjtm412TX}p^TBk)8n@H-LSl+- z4X@I?5s$0gVx3;Z4E}nnsbG}*NOr%B{lSXvK}5S!*yne#;*Bk?*5Q*Lv?QhQ4z6ZI zrEikpZLSl=rpPf9^J}eF2X?h{0@o^6zj)6ZDkIgx_9)Z7&rds@E`GWj_KXG}F86-1 zKSZs!*qTeC6AcnCWKJB-VbJn8tpu?5b#PG=BmJiqB=^{nkz+McWMt%)jPwtm>y;nP zXyc#u1%8UBeJS!YBbGJ>of{b`K!Y~2=zFlCQfa`pHYDD2L$VT7oQZ=&9P{ZtgZjex zRruo;MVb~#3c0-#qz#W4TAvNPtt3M%v@Rzmx0)`0_7k-1mDl1PVV^`kot@5RCr3ow zm9oj;^SRug6(7^|n5{KUkM}<c+`$Kd4kh5qDA@$4r}}ScZZjt25se4l zU8_ea(-aF}(SZy&UE=asXJ=Er=HAr{h0fDjpT%vw%!v^r^HqY(Bq-Q(v#E{3bl$?sAiU%m-JgXf< zN?Ts@f&ji-qH!E?+Dr@#Bb{yvTV4a}h?UZ@8LsIWV5wyT1t$-esdzt=(=#0*(r`mV z-qU#8VKyZU>aVQ54gBbQslS)N%vq*Y8MB~XQkj=)=yA7Y8N&P0)HfqX)u;nSLC=RTsIJs2&h#{c8Jvj;L;H@?ycNkp$cz2bGYIeFBB$OhI2L5lds6mwx%a zZX4cKW%+ajooFA66kdL=goP)@_psPRYTi33oBp04h#>;03R6{s^P*pyHl<)gOIjAa zGIk|_o3NujCnxUKqorkrTK76OmhWzE8r~##;(U*I`hEd%rxy9A!qr(072A6XBmZyZ zY_LAxhKCtS)+T~jK*LqD`CB_D1X#DNHX{Y73`5$Z_OdQD8z5PjqUYzU-5yyVziJ%>NEnw$fFUgoLcAEB)8kr)G=d@asn zEaTxCT++4Uinnk6!!}aVjK=s^5et%BY^(@{s(&$Lv?)Vj8S~dfr@Zpki;r?;RPFRg zN}e3^vkBp|CKA%4X2)Mr=CFoFXxT7&LO$WIPX;r?ZcGB2Iw|o!;&7Gbk((AYzFm;D zaK>fxcqG->j6Ng4i>UsTRx>4&Eit_MbXj#ZA0L(HkR|BwjcnQqNn3Qk7jBW%!mx0~ z)8pae{%BsK2sK%qE-*y{wgJ^!*GyoRcP*t(_yicvnzV;iWN5aXYR5d?TEt7N#R{2+)f>5#BGskX9&9XS0(Mb zoTcOaCuZ0ZB@uX>w!HE;z%-U>MqT;2A`)N^oCyZ0*I9~ET8kaH~@OPacRv5&>e zK*W2}1<@>m79K;AHmkoKpPN>lu3yo8m}}IWb|Y?C^{M`{}|_L607iegxQ8EqA>o3b@v0dp833Fk&5>m#~E*#G?UG3 zbEXdZ+l`AqOBZOTptOmL=DrG1+;+b%h0WlhNjgwV^uYB#gr*sZA2hD;5>DJS@_aes zlTzBBD-e!w6%;R%5DeBtpw*srq%at?F5?hl+AQ7NwH9FI>(>Nwyj(*wk0NcTs!RQz zKC9b%xJ}uCISyKj@Z}9{^CX#4?l+ujnMjpr&^6l?^-3{HoLW{sR;pQ4r`RzN{#LCv zn|gE7Wf99Lavq6R-SNR$c9cO}dS-bac0?+$^=E0xHO-$3?+`@%Oa+|W<2IG`Ze}Ag zjaHCtRg$=~^K*Lv-_^*zM51O0+kkbKUccfKb;?>)-u-AKwG?ZzlYV{{2Mm`<9_?C7I$hOET$#!l90E_KXiu8Z-370SS$f~ zU*nZF7GJz-eV)n0<5K2^)$TbHtu=PlEQqVaP5U6#wpt(3cw18RsXuXi`^2T%X-*!+ z)kb!pS+l|>H(z_bKnS5>1K?1jz7|tmA}f}ZgN3G$$(cR*FIJkDtfs=cFVURq0h#0{ za0~U9UrZh#aHYM51s;XCUbU-jM&Mtz#@U#l3>L@UNf+UXqlGbBM{#(5;>N}_N<}xm zjSq?C*24@6C9w$ru9&^$JT3gvZ*Lm^fMcBGFW%Vc@mi=uMJ%XBQe0hmJ8L$)%U$ej zTbTl$`LJ{;^TA;(vt2Ui=G_u-TiZLcA486cLsX(25^LeDhT0#3`Io|->v45x%IS`+ z#Y-t0T1(gXRqWZ*h+5r_VYXg=04eigEEQzex9t{{Q0a@1AgS$icl}BeeA(G-v1!6W z@l;E@vg3V{7?nhE@H3RzEf>o#L4;sgI4&6F5#H0-9~*dzv%>lW1$UYXUJ z+rnOvTEDm8C=X&b;>Mkos<*uDqKcgs)i|1r6lEta&N4qt2Z54PNZxC0s|3UuhS2=b z4Y=fQ)G%3qOGMp}v(D$=KO<03Q~Ls1Az1xAIOE~dSOt~{gtOX3W5Q2-MUHWsw;`L`gb24tOpldY;>9g5}C``ETk#fMk*A%9@sit($LiIG4!Tf z&ebyqxa%w^*SNd+LB2w6W?tPVTeqY~koU|OuWDJ{ByP`UC9eCT$u8`{rs{IiVHXd5 zz964UMRqWFoTTuifBr-LfTG24CHM1`C>`u=Hw+8RyS$06(VtaDKW-89Wj{kUDEvuc zd*dyD-Y3KDq%*#3+3IrHQ(0Wz*4Mur4{Bg@jbn;W)BQ07V7@A23JFS2SNQXXe#Je# z7kSjghG?c>8r&0Q4XYz)r7A_oslHAHfZo;Che-j}Rtd5njES)$T&YxaXGkc=#hpmP zKO?GR%wt-PWTtJh(luJZzm%G;(Tst}7R1`hS0|&n*9oFlKiuaN#e`bayCbrlqWKH3h#?dLv7JK#X=vbeI zyt`-b0raI-yg2k-B7_$yF}`KeJ0Oswug=C%UiehZf1#UKRm^rq zW9nof6yuXdHcv41!6{&4o&k}d)11{DFNk?!v91|_9tn4v*Bh8k+(4|DA|J`NyuXwalF+!ipl1a}n=U$zRuK811rM*~^8^?y6ViQn< zly#2f8Jklxi@VjzmM^Z}f|r<$WE)DP#4pI3NKi$gwgx!gQPW@j={JTT%!HKgA0Y1B z#g6nF8yRJmr>DOSJXZbH%74=~7|GLW`$G1I4WUkB(#zv^nLzal`LO3NnU>>aJfg8z zJ0JsIHCoK2@0aN29OGsZZqlax(xfCvA{Mq@<)AhttK`mGR2lar;o=zgX>QE>`SM-8 z=>pN~lbp3?`feIZx(C$FmS?iuj5!N49{cxeucl@Sn^oCRfU*0$t0>Ptk~oY_jW(AD8aw69X);cSBsosOI-tSBV)_o^#$MO7SH&i%>CPu z9=SL&%%Wb<0+8p=*&0LdjnApnObdQ%aq z9>=K*MP;V^ho!A@h8k%tt&hwE9b&Fa&53V&Mc?3<=i!PUKbHjhk=Gt^eMRr-<_Jgn z6W%`V9h>na*L!}|{wV=N9e1V!du`QgwWmz$bo(#Jnb#8dFdN&B{Ege%ap(gQRj=zK zwXI_x{w`UIjy13Rf?T=L2zeG=Y?YR8196<1;w0D~CVNw_E_W`^#?0g1VmS2UP=0(r zX26GKvFaARAR$4#4ERTKbDy?^ouH361Pmz-(P&9r}!PC2_a!LGm}#^P6IEccR~$dvaVtIwL1$da=$o9-3UW^Q zV$~I#_k-Drm}+88%!O;YHL=W2-F`~Bo5Z>Cj9d9xt+^z@f~b?fru0OK{#g0$n#ppnm9 z$BKkwZ@I^S+k87#6^`Pu)k}{2_=CWm2WbRng!Ln(b=SpQtA+1s9+`f-_k>z1@Q?eX zuDm~A=nVozNl5G7*FH{8(0vUIgotWZJ}TW4N?Hi0r5s&oV2*Z&cBH0e|r-Iw{@%c2JDmtxG>ta$A&iB6Z^ElZ*LeW2l zwd!O3cg1DM4K0nL$>|GS*KdS{TqV#eXuN8q&`tps5P}BEKnE@YE0`~ldEU?M$&m^z z_m5SK1aB0+f-3>I4mDIE1<+n(NP(^`_!>Ca{wFzVWf$7ritBl8(kBb_35E4X`dIle zDIx;BuiV5JZa5r{zW3JNozRGSj>2rX{tL2?Q|_7p!@a#HJ#_gpy)~3lKcb?}M-CW6 zr8@(%fpK=;74xFH`vTrkPf{Zw5g6WJJSod)pI%jku$|$#FH~{T;UF&Fk*AmpbXgrQ z-#R$Jl||Wv-FSisMuGMFCoLgufYD@eS(1?H`W~a-#m;D-i5hIO;U*=s0VVBpkL7Cy zQP&mNyY^d_V`VLlj`%D7vGTdy#{7;e3+ct0eD+U61Fo=h0)Lx)5@9b-=c@?0=rT0L zUy$QY-d<&T)NFHleT1^Sym6R=fZL8MCUmOh=gfyXEXI}6Qvd2zf4fv)B=9)-VpnBy z#rYja8y*KyAoR~Rmw%fcTxW3K*4t@JkK>qachk~NuFIAVpQ*KJcE_I8aNc3uHP+CVNpsGG62KctLlYY?;w8w5z)NQ$vYz>+_xx`8#fL@Zjk;}qjvc7 z5ZHiml?M~tHkzxeX>`J#LNN5yRN^&Ke+$aXH-$pLEQaA1YN~j*sn!n&(qg01nv6P= zJ`)BGJ9-~EJbILEm*y&XA4%Z44JmY3E(0dx!MGth=cX!`EgVp-F$kNBhdC@w*CS02 z()}Pw>FLRakunj3g`=Cmd_eNE-6-tJH5)R^oaDUOom}_}gNrs6Ts3l{eEv=8xL5bC zXnY*ieSYtSy-Y$7@;w>$?Cw38uJjq$Uk(B*?Kwa=ccNz8B8iQRJ`KS9m1x4kYz#Fu zeuyy zf9`D%gRrOj27*JXS7%oHUWS;qtPth+y*5J-Fs7~RPzOy3Tw$L0x*h8 zRI__i1zg0fHOznhd}cgRKeN&cKo7-TvpM@SGnecG*-UQm#56D4Rj zSF_U*Mf90BSHlQ26Smf4?)SRjLM?x9njP!NJ-pb%^j=Acxg_@ zl}XeL&&JVR=gh3$S00}uLW3DZeT|)+n<;xEvdCw~e|>nRMN7udKXny0l8Tu1at;3S zDd1OhW-Od3K0X0-&JQhqv`4Iq-;%hH#m>@0=4bD*2UqtF5ZN{&H~nU?a$x( zFa+U;|1HpIq|GL$6Y=c$DhCa67Aa!(cmF5DW+{MqqcT;MLiCV6Lv*}lxiXaN z8O=LQZfmJd+z@;5gbE&8kF;L4jX(Nkxu4Bjw({oaZ{O~z_+V+7%vzojla?l59B_@4 z@V|qaw@U1SqWDi+<>^#3G(&$gJ?2DB7{l($i?Ti(<)YzmA2?MpXP7NF`+UH z9-heXQbE^YjtSp4JDbJj`%ZuIs&aDNhmvCDz8~8tR{k@j_C9Kx{FFMKFSf@Xj!+;f z35!g6*|iM#`5(ND+*wWM^L1YBH_VX=7>~+RN~OAcw;TPXEo-TKj5Q~h^bf7Duk%nX zuk+@g*c}Sp7ja!(>V72`-HvVigfGY#2CS=ptS#t#R#`%!I!Rlc@AZ6-TPvP_bS&9O8AdDFq=2>#^C(_@$a7nCYNJ_R?i>OKUi0kh4&NsK0IU_L)9zoPBinL*@GzO5ZOP!licLrNv)eNC?1FH(D+8^lG(p?f|=o&Lh$d?;Q)#9)%?#oNvDViv!p zN|}~cpT7t`HD78&l@Is)Pfk!HUbBIU6v79Sq#W2?;2{rz33FO55Fq|^babToocC3r zy%HDt$`ZNc@<(!Waw0WtPb*lr+shlia24C>H-8NZnpl?`fTjw%Ig8|sP0KKNzDG5Y zbMB;If14-=KSECzb8wPLT)(?Q{p1OgQo1YUm45|%3l4d2llU_H+9^a0TDE&vH)s-p zU)yW7_AFnC`Oj;V`PE_LPzG-6c^qNRI}@Gb;D`Qxx|Oe=Jb6;q2ppq(G0P4nJsN&| zfbkO(Q84Z8|lux+DX?^(~L8W z6g`@@0WPI?eq}h+N$iMyqsO1g%CFy=T~WBDVtYx|924J*Ma(>Y&%{)+YfQ`V*L(S5TA?1AY}X58o7Eo9k_wMzQ3UQ&3?N3 z%}&?OrI{34g4+s9O14V@YUodm&Ze@3ukMX zv(Iw~x zWld!iNk!{$n%6{xu(3{HTWxq#X*Qgg!dua*kEOf1BvXZ4l6;~0O6g2tuY4~i7bVKS zh1kUZNCLG_$hrakmyARv|JROnQf#sC<^E`1x%nO6Qwwn@7{B673ss0_WfFQr(5uS6}3V z7TznZHkIdmEPSu4P<__BO=Ym6G!jIE<1)GW<-1p`eHhtGZRmsYCOzUV z#%-Jam3P7@Gn+7W9c}{6Ur8H?ZhO}B03IHluLPh;;rFMRnVtgDulb+4%(XhD z3N|=1?KWUF4MiDgk?a?5H$6yW#F_l|aG0A6e5a@r(Sk7pm@qb{y-&tA*1wHr67gJ-6-e(Z?BfcAe^Nw>X4&d4y-p(DK*lA;*1GpuSisgqE{3tt`ajhc z5zGue+T{}pjHA&JqEs{Nflzb+rN_}Kb%(e(mG9vztYdaoR+`_jRiUV2(aF4<@^K<( zh{ze2AS)&+h#})F@^}~i7}sc#Ca*AX`VsyEge&us*^9LkdVD%yY!>O(vmb_8n3~E{ zY@L;CCS}COPlRv?%@nHV8){v&S&yaH;Etimq?T_Uk+UyK5Qi~v|6r)lZZ`6nU{AGc zS@g70K|)j0JTG2F_tmV7z$l}m(>xCxklkfRg9cb>Fv*P;j+mL5`R&d*b>EdJqA0T1 z&z{%P`ydEL*;EFngvEDgLXC?4pL|8spEF);*ZJWY^2cwS59ay1=d^aIK0LN_;!e-b zkb;|oXd%V#h|Vt^;!S#)Rx|`31q=SZ`pLqIk}6NL5LoR;IicHv7I}F&P(L(V;dp#v z)Y*lTa`{Y!>T|HoJFPf+F+Vgv4njqn&Ezp338sQ*aaBt zEoZYgbu6GQBKV;u^TM?H2hjtTMBBgN*)@XCo>?tob`d`-EMNxbD%oQ~7{tH^vRHUy zV*}QHC~s_h8#PyJ!BMJsv@si^sp{F~rp0on1}fpUuACwi8x^&Pw1|kIXKvn|YmUe5 zWot<0qJ{8d@%GI}alCV$K@D%1qK8s_CSdJl+Q-X|nEXnc_^|p}LAUV3IEy)7L9F|{ zH%8xTrG(QW&3Eb5#_Nhr%Z6N%%~&l%QSUAsF~FWRx$P?Dzir;g@x@>xcH?{vk5rQt zYamh~W#-XVI68Z|Ze*#@ zIfx;5B-8(xLFlwS-XOyvz}E~+hJcS>dq(=!91{;kUnT#kAc75bo%iSs{koEKX5Ydl z%*w%m&%C}S$EdGgJ+Z(GmXwTK(Tj!CuH(TJ2QaAhm7YG~lPR#+M;{;KHj{TP#UF=1 zT{)UCXmnZ&%}?{dEs59`AuMyyFYN}`!Al|M(-VIG=w!ax5)ZLRqvF|3aKL~SAS5_A zMHrSQh-4hzqXYF{RTW^thGjX|P|2F82`>+y_kk5Jx78qc{8}iJI;MednDjYO#W@&W0Aj}NYp8#UK#I31rG&L}E2@SrA6mNwgexv0od-$R>ah}y`&y8IwG zxB=~?)>&sOm^#aKG5nM2{#~hNucPg##8@`P&$tv}FOc{%v#`k0qK7jd9}T_{w*Tz= z7qfEekERjM*H*Sq0I_Mq$jZvfv(^cJyfdRb@_-xxaDecrs5!_ka0X9{BDZzUz2;g^ zK&+Uy7M!yZ%AI+n;k>MT}U2?$deCBcH zwA*TUV-F01XMNAqXmi%@tj}Z`UPakWu$zOt5NZLxwUR zc5psD6EDk2>yBvSO1)B;485INnUHYLu`<2Rv@=f;h@j988oPj2cz(1|ZM2*R>CGWU zS1XHVvX1zmQEp?Ht)sxrZa)y9fvv>}+~YX;ElM?6XASc)@wMS^-2ANEGsYq|g*Q5X zfBUO&#<)_Gu?x*(QO|N=qt@8ii1$p0XN1?8a|RxTI6rBEiRx>#T1YjvjYq*&S2fd( zjN5xSTDlx=Zv*6;j317B6`Z8X2zxS9n3-1phBYk4NyRJ|u47)je+zK$=e=O`w0GC- zYNkHq$-SfH6c%1iJrPN89Z}qEMVm`Wb=Tb&jK`i*DvANbr1Lu7z&EpJ@b}$=bEM|& zyG9X4XJ~S{^Co@Dp-rN?a@DfNz~BGD>Ouf zf1l76dIBkR1q-6r$+jG64o`7d+Yo851C}&5YSLRIHmXUq*punrE%@=Sz88CnqrXDK z-20EmyqeQnCNDaPiHH#h&m#+X_H(EW=~j}kTDj_Dlw!s3kA}m@IMWSSzM=PqP=^|9 zNyf10+~~f5iGekDsvPn~lr-+_?_Q5(uJ~|T{-ua_e&EA_o*lblov>$@A>Udj#T+9o z?fEICa&nQ6%rUI%gKAOCHG-d|0EGL=BJ&J2n3!@*U1K4;#h)tZA`g&SW57MZvM`VI zAtEBabLUQ^^o<)gN^p@o-xx*I-h~&ZQRNIYS-;cPa6bOi=(XeZso}u!%I7oW$pqn5 z!f)B1jvqz+)=oogWu2~K72CxiO*t;@1qB5X zLB(@KY?<*Z-m-dpR)OPEgD{=OPb^l0-bJU)+eAvD~%XMH;sCpg2q36rTZ z8i@l7+Q=I|*TeNAI^J;sC;;W0Y|L(LSQr*Ch&(-HZT@zbh~N#n<5@Nc3UAU2a5;AE z+37cF?DC`5k{LYK#UMvn%?Ua>I;~!Kr*uz9?cVXRk`vdHNfvio_V}iK}g#%2e<{qVey5v37?+5Q9F8o*Xxw!Gae+nG8V8Xe~) zfQN7ao@IHrp(HuMCg;r-AEXj3w-s#;g`le?EHWt`V9o*%3!7WZTi)6e7Vlu!Hg3>sM<J%EC1-hQBe85zzUrm zR@Wf3g{3B{|9yz`(a?LTt(&W!h~J`&edD$>Gl4;doL0u#+Ky*u@c*g>{EjP~;hS87F8Gx?e!Z0{G;J6D0uAg4aDKS&z(Ccm4)QpFx!Gv4wX&-z+!%o zGY?Y6)TmQ5teu=r;45(1J$_7lxOLKH0x&7AFCISo&ZTB)ISJ6O0O>qZue_U|=KkXG zV+rfbC;a@4J{WC?l+{Sy(6`0To^Xa$Z+T7Ea0B}U_pw++t#)ttePwwi4_ssZ297;- ztS_3W4YT%TeZ2r;c$up>U!#RcN6>p92Whs(?IWkx@n`Y#mT;5lAznoNNTX9eb-JJH zPRbNj@oIYu(M0ON5O?CBI+Jpa;1X(iDa~V(#Ia3CmUHga)Gy3e1uVj#e$?DNRNvO~Exx-D=WPnkq1kSkK)u zT%W2&)apq}Io7h%8ED!a?Q7eIhNoLIx?d-o=082<98k(X5Te9Fv+)49cP9;Bg zbdFmO#WoqQO8HGIap4*HK3a~KBw|h&3kq1uUpa$R?gZM38D8w1gRlv+JirdqL}#OBlXFLro{RJg#do z)bf~{+xCp3U6^mz03H=>X?X_Hz?vKTbAtAB?r3LV06YJ}%h-Z%%TM!8i6eJuIw)^D zvqR1|Tl+Lph+Md@h0g4x35Tpj`(SQ*gEU?Z4^m?e+)6dro@>@~fD7W0-Hh99Z6BWE z`rX}1G%@v|j(6S{34%Jq0;wD>{5;w|$pEV5(Cl!uN3h z_J)q350q7O*6_c2ZheD>>8@ z$h&DNA?Ka11H?sqrWLtLX~+V(ZN{;O=<*tTR(HBE=5X2mg?pHiEqblguT)1uT|TT2 zak8{TEl(KRni+rWfbjVKPs~*T_VTSgvmf4E1!zlY(XhdiZ3{zhx&AWh`mZB>fq{X; z=l6Ux9pw5tD5!^;-O3K(W9WC`44e?_m1+S<-p))uY+|0h6pr7W^|EWX6$tY~C=zYy zT=cuF8hRfM5uVJ;v>Tt_@@YQwvY_Ln-nY`KH$t!d^1lvv=4TyZTxy7Ui6F8(QD|Dt zId)ZR{_oz}szWCc=k=GMGRGRq)N`?jAt+WF4qvR!WIsKpa4qBBnesbKYxBO*XX2q(ROgS(N%YA5j3C&2L?oamn8MoLA+96GaO)5f!i;3 zS|!DTQj7hdUo4fB)h*amb(zH&=P%E;Hn3uCk-X<#d-#M;n<@Ifr~$H-S6>ysH1VGH zN!S^>(eltXoc`5Cr&I{Ze~P}Aw-Xa}*vd<%Z2+!*rY>r>>F9~#_CZVksy7%%9}R&(o8aRH5izlH zswZbJw;gL|mENuawgzJFIHm748}|7#@S(fi=Qk#^GLA>SV@HC0Q9fZVJ7J&V#4}ai zHk(Py%Ca=Jcw;Xvu!Nki5dekqDd-nBIPJp5v{dh1YHF}I5$J~yQ21Y3YT@2KSY|!` zXY6(UL*|MK3I^5Yv3;gFSeXch-Caf%ZH<@CzkgFQhvSYWZa}1)9WQbk8p6M@ET(>8eQrthvPUd`{kd~BOT`}Fg@Uk)PDD*U_@xg?v z=g)};DJAoCet6Z6fo^e|bMw?;D_Es_YeI7SP?MFNx^*8uuClGHrO}Ca-|6k+vC~VA z@%G$w83CyZ>oayptEZVf0GS6QxH=>zT0yBd5xJOHEyd&g<6L#!c3gQ!$5kX586%H? z022@I$u`cXIf6k@(#zFy-fAhQupQs#jmX)uOx$1TGdWD9qP{yW3ezZ0^N|T-K%!95 zadGEctJ6uQS4!>l9s#-DXA@| z!)@&Q?2T$QwcB{@qvRp$chBEZh@5s6iE4$?zhZj*==_L?Lj!+!)&GlqvM&;Js5d^; zlSCv4Z8YuG-6C$q{f_cat}me``V1bF%+j5?c+4O@V7fdUwLY44Vyw{~EXG3;fBYD& zwn#rcH2}}$wp69=Ng7w>WJ%pn^2AeS=3x=1UEj$1DI5grWesukJ~0sXI^uW#$QVZD zv^JRi@FO;myQvD%7oCqdjY&^OBB_Ug1|uQCwb^YqLxLN!`&~;vIwnO$Rn=f3Wb^2J zd;9JA3tDbaanc8B5s_l9$3tnt5zxNr+T!5g5b21t47I`$gxaPbDNwJB$HtcGpXC2~ zpCEK5ki>t2nyt0PG!cBVY=as69Wtn|fW7R_foAJD&^6N&l;Eefr9$=I4Rz^I`jJ8tY5}}%y=@MS*UE2) zs9UNw<2W(V%~J*+Fp_fp0WII3LRb#HSoE$P6w1kHou%Ht|7g7QtWi(IVTn5`3XVBO z%{7`R$jQAT&-x|y5a^OJtv>FgTtD8yc;P<_A8tg5oRg8CwO^b|D7lWm(^AeX{K4c3 zCSnGXOS%(mtksJnR;}^F6mLzNB;(rJnh<6XpOrn6o1c#eGOu=+sw&TYg*l?z&OJU~ z4{|HjlCOl)URgp#aK zt<^GQ8vmz}^D>tryZ>}dNGA@1hTWBtIu_3oFKgKE*>G&^u&y2DhTZ9{6vZF5KG?H> znj)O1^OZ%Bg1KgcSd81qO-PFOW?{T91|^g9aHF%;%Qc+N>n`f=Yx|uP^UpXY;9LsK zboBJ{k8{x*Tyj7!R(~!-eea(0Neae_=Hq;u{-M#EH<6HB5sw4+MkN`)Ip3{!fndh` z*Iq27B`IkNo#IrB$+9eCebJv!#Qsci6^Rt=4Gj+NhxYwIJOEXLD)S-vLfzscnwJkm z`1rn3Z3(z+idRRYbqY70L3a)hYk+o&`!vhtd+@hBB6g%iQdCZY27WXcn%*P_uS!$W za*2I!X7&`z7Co@Bo3O6kDz+=cQPa>MEAJmOWt?Z$xwrS%dEWWq{Wrzo(*9z!qrz?Z z5fSC1wzV3=C_s(5-~Vq)%$al^p2B_fLc(0npFk$!&2;+Y3e z@-}l+9F`MCocO10EKfPdP!h~h9n)XEJUqzh`D)Iu$HvF&VvUmb>L`Q{?}Rn0(bg!p zSA2?7LqdHPoI6MaM1cZ|lQs-Q077>m!0^APN6)-$>wP351RF_vMnXJ$)<2Aq`xB zIgEhfQvbTp=m<5hx20Qb@o5py{EOt~Kp(%;VglDTyozqX5g`#PKU=>`019rF3&TK) zHSc>eRcA|QMgClnx%#~NQMO6S?5$hyn(4Lv5J#lp>O5!U-TSl9vq-8ZUWenB} z3knU5)mR?A{Cw)D;yec%g?zQoQWt!%Bvik-UO`l_&VH)iPwR+f{agqyvp?%K?A>-6 zO#v1tkZVH+2M_zvE(4@sy~EddZ;Y2F+clb*zkfe`HkXyHlX2I+9Sd<(>!?`F2J5l= z^ULTK=5js?c4>4N;^sX)J%!i2w4DZfZFdX~LhO3o&JRiAjPjLQ`+6th!#ze6>u)lO zOn-O{c%N2^KP<@2DeE{&ivk79{I7rHe!B_ zC}7x4ex76KE6Ml#9GdkB`}gy{gd2jn61!~}oNgYh#Ba%E2<@j~pjtL&y=*))U|OY= z+H|}tT96h+G{>OB`$0Q8HCDu==L2m5;3;!x@A=l-3Dd-Zj?+=k#f2VttZ6p@Q;NX7 z?BJk^qus!9mYY`?JF4*IfvIV(-`Uu0WWblPcN#H%M@K(w%9}Sfm<{vB8cMbY?c2^$ zj^|rE^nqwcSqLXo8jG5(-dgEf9!sx1FTXpIKfk!ks11QsIj)vJd-k2YZ9PwsPCA?( zJvdS>g!%Kh<|@qyI}fX%e-FH-W+mO%KmhPPZOR(q#KRw(m_C(YbJyg%3f4(~H;y3rP)45TP z>GownLdRQjIinR`ZXj`ahZNbhZajj`d>{{(Gx@=C+vX@_oZcexphZQCY{a_dC@@&M zm4!K44oeeQ>Hx``9vCp?(swV*+p^DWf}1c67tO{Sz9@pHYr-_1$HL*qZ^;b|Xd$$= zqoiQY!4t7jJipW5FHhL04Jobv^}eCT2~1ITfaps=@SrSy^aa&GfN(C|uZi4oE^T~> z!18#2L+*H_G%cE{C#^i?VA!RV1Vb=0rTq)|U3rw4ekFhOC!_p|s%mIZsKq2VmB zcY6yO0I4v+T>rF;OciuiJVCL}QbX~(F1%)oYe Op>k46l4auWzx*#!m3NH* literal 82561 zcmeGDRa9J2vjqy{ZUKU82ol@^!7W&V1qkl$jk{}b2n5&Q?oQ+G?(Wh!G~V3Kcg}al z_}~7=`*6qDPpqbQ@4c#O&6=}jxRQc2CK?GE3=9mWtjt$s7#M_V7#KKL6gcQ#x;}u= zVPJ4zWWS23x@Ukg5e@JbNxL@t(|BA6zkSzcC}?hLYa8{44ML5Iwyowc96Rfb{&dc{ zw!WTg5#5;B+9XTz<~RIjm&96G%W4ylJF2U&0ra{a_S)PH_a?p5K3}$#DD$ox1lIY zh=t(%-;F{8(Ww9LMwp2I-;GcL|Nnk9XHu~tOaAC~ffEv3Thr0a z51APo1NE895~nUBh%pC^?k`}L>>liVsBeguLCfyzH}3z>w)*2f^3cir3GCf=KAy8% z`<1IJ&;@`6uuf_fG>ENs_Hk-dpi2+qvEJN^8U-V4t_WsW)kKb!AV6R}g@jSbF zs&&k)jP(^A9bMML@eU-WO0~0x&ENjg!!Hgqxpel!yH7D# zSYs4JWlaZc|3b4WSbZ~`=GT5sSarmYPQA<8?2bx2jn$*7Ey?4l%J<$~*FtXHJEu+O zID3CoRySj(kpi%6wc~3=8R~25>Szo4Cl|v;Sx7!$mCnuf<>M;W)HO6T_|BE(E1~(> zA{5B6^m}l`WDf57+-_(m3`Wm!dE=AI{)p~5?KR(7dPm6uIqBKA&JXS0y;)8j<4F3B za=HBX5&A*sZ`1Gx%Iuw-jM*F<9PA(LqVx`RAL#)`_ZUC990 zVlkPLSzaN2q>Uh|p)%U-^_p$WPWWvG0b>`;p62|>%5WNc93c|k^2v>M z*Bdavud?0uG{#Un;$JL(#)CdIl0ppDCW~%4eRswe7vIOUXt(qAHKqDs$r2@0+aM3dEe_NoO^F_VN6s-PslHlPU&|EqdENbh?}y8}I(+JS;g zULCK;_xsj79O>T*ICmg<9+cj$;GRE#PK`cybg_h&+q}oRLm!i2-C*(DvO1M-TKw+h z;imrA$H&{1(-t#aoZDuh3M&lOM7`e=kMX?h1AW31oUpZ zd$bl6nri$!0166y#I^JlgZ=cLbwa__)z$L&xB!J05lf@^!19>50Nn_~Hd({QWhD3A z%@ufLa+3G6-*69op!(1TG#4bsu7BH-w9|b-zaBZNnp$gVX*Dr1`(B`IB;>r(g=sgR z!OE+y?jCeZObjAX{lJzwmpr(zu&}YgDkPN9&W!~IPfWyUYVsL{_g&ZJWO+8X0p%`` z?i(83r!iuDH6Iw6mi6Q_>N82l5);u8{>iO1EhH!0J`Hg0@E2-*7aaUxSF zb9ol{9a)_)x!lHf^2h4tMH6Kbr-*K=<7=Bf#D7G@>7;Kv=ouMlE|3)`zYIB+qDrIf zacVlDFby6jQ5uWl9!17`Qo>H-`+!U@Y_r3?6Y|dU%}p}$$eri?{e8u( zN31pHmBbo6Y3#7M9)(mnisz5MH{f;Brt^;lJ57|y3zhrQe(GgEjSJhm3GgHWPhKAG zipCCxxeooFt!bZc7mH#jmx>7M9?zw-IVKpW*t`7JigV$gZV{%FT)V#I&j`3(0M=_R zF0T?iuZ#17dRzC|dLo3q$yzS!087ZC`oflLL({#;&+cBh6=|vuM-I1dNv~dz>(4_; zyD#-ZEw2309oZ&JMzt}0v3$Rk(9bE*0!ed z!cMlQopfYu8d=&r1%DB*o9hNrDbr;i`e9ap5)` z`jI@J+{xc6F`0rlqZgwr@uWF+;NxXX<<(|GBB6UCD{kkTXKc#VR%5od;92j7^YpWo zdJ~>a&C75dR{hq8-pb6?-h?bZKaUlVTlB`zWnXa7B-OpJ=S4+fbno#}mL?zHWrtej zGDd8daFbs)#T0}+UCDDBR>b{&95pYjmnMh|Z==-uQx-~a9g8(2kgegxd#OqLX7MeN zu>$XTZyCMw_3>||=6oTstxT>|C$C#lTbXO&`Q=sBW_a@?*;K3K?3Hn-i`8aQNhlRp zCeWVP>h#ZPlZ*of*WOk<9%j!8@Uykm{RH`OeL-R2W<>M--F@e=DINUY;}$G%JNzUB zfkV&Pf->jlX04AKC;`C;uIzUyGEo#yxdt5$N{SGmqiu>k#54MAC0g6Lnbv+K zp97OrG=Tu5KT7zNHwLmPropC^BLCt@7M0a!l=J6wxnVXXUP6o8)&#)z*X>~cFq_}& zs#EuwEftLW?CBLv#$#{)wsDzcIm^gkbETMlCS-3im<1o&Ya2xXxylP0cejPpfdzNb z=(KvjMgn|uy$t`J#tNssnI+I|S58b!d9JnkV5c>|x8|+U=F97ciFq@CWK5i`dLCo+ z*`Gf~s=sMAK)~zIppT!?s^0c@uv5pOhd-pw5vW(YnIhZ*sInV7p#Rq2vUOnBNOEGDQlN!=2 zz?Amon^;UOvb~Y#W17)(pURVjyg_znppOK6-YyEgYmr!0>x5j=T6_j&+7oaKMG~}g zo!-gb5O|mJGB7chv)-EAy2YVyvpulT$B<8u!F>3x@!I-bSaYgQoBq-~spnmMdir9q zBDPtbFA%5kcB;|E2^rqBX@N~4Mj#DU31#-?xQ zxS(m2h#K6ly`W7X^di7PTzI(=(R$DPElO#rT(-Ai)DQk*p-F_rIJ@jHe%81rWgArT z_LY8Q%+E_rDbxR+z*p=73J~m6y@$v!>r_mvr5PzxHT#)2si;VVFwo4ax{aeC2yj;^ zKZ3;0+ua{mp2k;mlW7a5nX*3!w2AmZNDyc|vAr-F8$tq*7)N9E{T3Qj1i%XJ7xE;Z z{U?{=C0d;=+T4%t4G5+Mu6o@z)A(%RykfIs@2&9$ZjYyJS3r+_%)TAMZ!b^es!Y&- zBw+BODpg4520!aV5Qr0B%Q`zL+v-J~VDIX+55bd@pFP-m@x;D9*lc@&JhI|?RR&@y zINZC&+=QDQ&*dGe!4c^;l_ zVpHZgOOSxKtF-9Icc&G}7#8kg`~H(z@X`c65EifVGQ&`W`4Y^Cq=enb&s;ek7pL=> zzkvqR$%6`+^KI_wYU6&-zNjkx#5q1N(@tZT-8n^>yuU`YBl1sR1L=-nrb4;A_}y&U z9y`}P!!Svguvg5@Ze07|pV0SsmhQTAdW4t8`wNU$7cNsd4R6i$Uz=?Y^^9OI^GAIl zB#g-$z7_0_dc zB;F_zc8vN42fTVV*iGB?oB{$Fh0Xd5ib=*;a_(}BgLrM{>?dAAqMD|vi9|bJE^03< z5&ij^++tZTr%5?J;)X9B27WMq$nOx=0~v!{IFX8R;#@qX#SQcN!(U#E6d=knaG{Ce zaz?gwHXg4mv2(RG`FKr?-EpWl&~3GT`sg}Zq8s?R7|T0nPvVES6GU974N0*7UY`}| zhQo{Agi40)GT^J-N~85~lSrrc1kE2}EHOqQA&pli9EXee$1Cw1d0n99SmJ7#m$?eE z-__MKs4+^uQ*XWF`5Lb`v+NBbS%qx?pGnCe2y{V`RFTX45?X;-d%Tu=cB(RfciN(% z(_x8EF|dI!iQ$E3f_0h3nLg#rgMn_!!=?`$#(30tnQo8W@JhGl&k@<%IIk$P5HSz; zKX$oU5m{cLA{|N+d;Nx0DhnU_(oad^wunK#L>RS@&*-%4im)1isx>tt;^*q-*6@OM zv*G!g=eB9JMCE)?A1>na%vTf2tCwxVAeS&hfE0Id|JtOWNIGZSHwCM`2+32g*vNSN zDbBP0UoIdmi)RQ%^5_sCT(_NGPv}2Sp!~N;peCS5;{mdINnLn6*YPMZ9eLm3H3rxbZA zb&wdtO5k;LGSr!$!z(a3{c|r*Y0>X|PiFH@pY}V6S-Q9{(bQfZx!hew1WUR)BnFq# z$yb{e1YoQ&1ZhzBm0niozMv-{96hwKY}n%pnnP$+M*J zk|H=@6{VoucDM8C`+(!2&xJ(xyveGGkVegcxwdFQeS7PiE%1EDAJ8D8f7>yw?f(VR zs~#z@yD%{UP6mjWTL;B19$M}CrgiGYI&YN%s_;Xdb4 z=a?XR;^exvdXVri=<51*smUiZi>Daby?uM6v0q{>xxzVE^$WBEg1rqD6pD&syh@ag zjT}$7y1C-;)M#5^>C*Y?$IA~Q{pioblFpGnWzYNyqH%pi9z;*ES+s?)@$=$r`M@EI!Pak>=%XwNY_L?Gps}FF?AQ zo;l#EK;@aI4*@WlCH1Yp2>oz&!nQeCqg=!qJ^!S?u>YwfVW2uWw&+toUdd;Y)?h+> zd{S+6c>YvG*a(=0H=6Ie@mQ@bG zE1Il}1uV7}==Cm@=yJSw*xderh|7-%E-fmW(0MSpQnvpM5^_&`Me4RfIwv$DS$F)L z-Bttg`KoSBg?u|O5VUD5{JjBQ`y+Qc2tFrfGu++&qT3j#?O!l3c2b+oGLr@^L_B2d z%bkj`n|>PK)G-;lLTT=i?oLfL9xTliGKZJ}*7bEr(wJU)qR0M!|5lLnrx%uIkhPBF z7rt-#Gm^hR)UTDAFZX1j@(;I^g@t4`@1r8I@jJVY(PfgZx}7bOe!wydnIz~SGT`)j zGHKNQ!bwN4wgp9dC>wOE@hEN?U(Q3wJI4xW@_V|e^M1mx#Tsp! zlDnr=3Q(`wgj=T=%6Wahd$8X~x4D@avrI@Sp_%fjZ6m|6+SIQPUiQAoO_}%$*zIc;d71+4Y=n@%w(5cDxF+~Gu2y<%ct-(=O=S3y1I+x!4;!G1n z(dyt1e}sfZE`Uh6)b98M<3-++6*LexGJfCr(sDhSy|nXAko?xwIShGSQB{sv16Cm$ z=zPBO9#}$>L;X=u9`2qBh-$>0$USOT_UF&G6KHRlMF6}8eD>e5fAF@gOsZrTl{}iM zVR0h3ZJ-(GRQNuo^e$(q4u~0qbPIv_`ORz=$&q1J{qY%WNN=>=XhhB;$q)V^fgE(B zr08l=$|6HOp2lsm>3rL#8N~{Ftl0sQE-d!v-Cet6>Lbc=k(L^KY49fhVsmqSy}qXN z%T$yJR(x*;@P>R6T(7iJC|L(3@{C5WpdAbXUc+>}jlGJ@No*&?=k?$$xbLGr;9^E7 zUX`%`{V`Ubg@Wfn5(qCA=CjFZIcss#*ICAsA?UQdy~{vPFFnLs9oY$4`Eo#{6gAim z9_W>pWZ_yuakbUz|Ed7D4HzG&kTMPywUl#z0`>NP2W+$;Bewl;Ct?t^k$snjwXi#8 zB%)yE3cg*wIqtqu5p*e|v#^aZO;ClYeeFwiiYj{7o}6R8(drb- zC#v{|-~Hj^bibORKyEmQ6rnA_S_p}!%=HjVBz=@E#a20bXK%^dCDH4sH42dpQSYm)!9;UdV z^R>b8Y`}JkHe?R<uB|+jh2#Yhf}9Tb2ij` zLYK3hxn0dYDckoMnS0M%c6iGxmERC_NzFEF6RV)l4XSvyh-s=fX#td3n;%sHf1*g_5nEo~tH<;h!ai)@0 z+TZRGD`9U5pupB0veXOW()?PX^Z|c%{`i$whDe7t{cW{ZILFZ4NBIwIfaT6@(=l;T zsHj5VwV>(C(Y)WAqtRPqF8uQe`Q>EkUH#^Jgzi7qyxQ+ST|JaI0VedFpFon$US?_p z_~FWgNQ#QQQ!J^I&*;hQ5IlC&s?~bkk+k^z-TmP)lCy;pcmdg|bk@fM!d5rrciV0H zNoPww{wW4SUBY8-&5%>44fEtOu0ysu`l&!Fz0NELE(FyiS+0#-M5Sgp-6dvNiZftr z$CA?)O0yU~+WzC{Vvfye3%MMoR7j8D?l>a!CA#Ai!+X3Zf6m9qmWk!GneBLdyua{$ z++7)8MKt2b)P)k1Cae8~z7WhjBUl4ZzvoxN-r=oUcnt@&rLKa)3131ItCe>o4sYAf zlx&8SI7)5BCgPEcMds>YMRbwpwA62y6_k6p0_O^42l7Y0x#hpV}wUIh9N(MA|n9RkUV=Xv3gKKj}3+ z3H%|Lqux}W)VfWsTB2_7I8)ed+P-ge*(o{o;=Ho5&?vxWy+s-@8NXY!obKzpF?KIs z_9oZ?C@Yy9_%oV9GpX-6je1OA#++u?cjTKxPbsWsQW&UzCMlVFxOCo!KG`s`vEEjj z04>zS%4A2^7i%8dR!2HI=N$PiSB7PzO=okYQ3&MlM0>wXZB}d4FIZ0Zt@0Qvad6jP z{!S;D*+KwKj}|Ue!>_mF=9`4s6%Lm};aEbU?VU%F6F#A@kb6cW%u#i~Tp4}=D@M+I zJIB4m0lZ?LF#2u5U_6_YTTs&XRikdCK3CZ8NXTB~#qIgSvAT#%lw}LFV+NO}gi%!z zPaGqQ1p}8`@~$4E1;Au5kNz6i#R%qXI%XHJ-QsLfD(Qz{V5XLt?l|w<@&VzuIF=pF3GwRWs&v{4iXwviqRy_6MBs&@aVWx2i)5Wo z7Q$5ym+1rlLVDjIfv>lRP*z5-wX5Cqk7`;SG@Hy@>IQ!W{+(z9oQtU9>+nKBuu#4G z!xp4Uk$7?exAWC|ARSMcBC|xN6T+iuptym{_>_7$P8q7pr>}2(T`q=IaM@2 zc1=_=WRc|^z`DR$``z@5gE7#jZc|F{m(?XX%tm$i#TnthnAD@iWyK%spZ1rpReQSJ zk@<&Ko1fvW<4}Ryv5<}$F)JWbSAwK*e>0PV$NXe z7_?CeWmz|#ubbKans(KZ4hR_=AXhd!JhO5qMt#Pb6Z9;`}5Ine5NHZ99A# zecIgC2uQ}&Wg#_c-_6&eYzpun%!dEU1^rO&|B#Fp0_EV%R0)CX*w_@pJc0)s@ZWb! zH-S|mYpm4U#9p(<1&0Z$NL{&C!@ny-z*}v z08Z}@LL;T}Uiicn^!UGbOp#_fO{3$drfj-bRt+=+qNf8IrVMIh53lND^+%hQc8Mx! z(8%~4R@ei*HEI7KdG|;wW>9>#!2vGk|A1H0%;o&Ob1qhFfgD5dDLBq+tNUo9fuS7n zVVJPv4UBv-5fBB$yV{xnA|JBqz^21}LSx#4a>{DnO-r2uQi~Sil-+h{>4>a2Py(h}5Re|M1xS#+q>;VAOO8E` zr2b>QS&QtR4+al?^dH56ZKPK)`E(Kib^p*gr$3tEd#cYunH4w-*qGGM-@edt*DwA0 zq$1QugO{gBDcxVq!tX4q04#~>gQH9gt7(ZeD(S1XH4U2oFBia~l2ppK0IxcJ+H~P; zv-;$5UGFR6#!EP`=JL1_4b?US+r`V`p=3E?NVVQN75G;Sx_Sh@c-2niN&pT(iml+v z-i>#5?>N<$l2|)|oFCVSn6Kq2fvjB1sl_T+l~FOXMO2X~WM-*e);+9|Mfmf@IRd=bhEFL*Nc`UVb_+_}W4;$>3+gBO zURiWLkSDAme5H(?*Pj9$gWo!_@|H4kEQR=YK?CKLDfvgbS&A9X_}xxg;`zu|>qT}> zPkBR|Pfl!;MzMJ0c+$){7zrcQZ(AHv)WMval`iO7=%&czK+jhKfrmcnY@a&AeE>*y z&fdsfgTWih>>>4>&XYD_lL=!%(6}64fD!a#J!Di!renq8M7?Oq;uz&T7zc|V+3xez#$pC^r|yI=+W=wYfh-Xw$g>{Z)nIC`?49F5JK%<6ZJE6Hv9_fWehMVNqDYq4m=E+`{@OvpH}*?K3=Y?K#!kct`*Fl3wf5Y;z3paPhymfp5u4lelh>m;?{wuVJ9G5dP5S9-H{`L!o$H=w0f&FH-MN z9xAl((<@zH_W1Io7ph2&oc-3f_(AfXe=cvMj9dr()NP@i5iZAf@4$i|3}csf*qt?t z!BY~-g2h4tm2&n5Fd0j=lJO{xZ-yT;Et4zPavJ&uwlPJe47GWei;B$gt;x*Wfn7({ z+B{P5fgDFHjaaFGFSxxu21F>+(Q?m^_kOL;HiKXB5KC%2>IN1DiW-8I2}Aj~3fuW^ zKAX{Nz}l_1xPg22=jx-{+h5jRUgrH?jhglJ)zoT~`0vC6qc%RwX~<~-^EblXyS}o1 zmW&A>5-KS41;(wq-AVsKoJTb%>(g9L+Nl4Z4O0rgK}TEaWu}+&8-4k7l1xDxpQZ+K z7%M*>EEN2W_TeCxhMqo}QWoGk!-u5JW8E`UvUMZk2&&)EG&uF}oBpks44GQ6rVTi(jk($U`ndD`S8lZd?R;v)tY2D9rotw;Wy+o^P<9wpvp zjoq9@=P4r0CxD6;mr~0&IJJ?FrfbsLkv~gV0HURjyUA6{URiy#8DE@)g{e7s*(h7n zHD~D>jpYZ-;NOFjP2V6-hY>MnTKC7x4VHOGp}U$^kB@dbchW_A60)>I9{GZzA=PN2 zRiHiw$sEiHA18iy898zh+nFzo&lKg?VKG_kp7}*ZZRus!W=XzlC_vM^;{)b`2i35b zz!xa&in~69O0-Fo7+U(S3amga*4seLUU6b<>Vy?2snG%D+@~!kY-exEYaU-s%Tg?)nviE(`X||{MkE%nVT!s7e6;=w4>G0 zyfr_k!ef{#rDnB&m&Hu@d*mW~q(MF?l1qyqal@L|u584}pq)6ybjRWrpNOQ|uZFNw zv&QVEE`RS@=A@%{kSGM_>u?v09~eR0!cra)Cv?K@Fe~Viy7ee;z0DQREiRQ{`ey_OSEDs3IAUeloMuTfQm~85xmCPX}QBP3se~YI02VlT<4b zfDIeS@y@x@Dng{TsOH@qq?{pM?MG$90Do$J`Ll&Jv^wsO2iNo`6})azj^6z(aXDLy zslXwL`4N}~H}H+0g6Ls;wcd)ywJ$Dfou2~<5r?@O=G}Wd?QOo;MC5;*YgZupx9MNG z@F_x2J;oRJ-nlo4JQVrzQhg1gTrMKi`06c`Y)g z=4`r#A3y%pKDAujkcidfLVfja-%R02O%$2x2(q3#sxJNp=gbbe^QMobyt^pH3=Iol zpX+EIzB#TlO1RE#X8c2EPNu1Fj1~lx>B|zl*)q(};z8;83f%b%FXW1C;(RH>Ziu?| z>0mrH_mCi+!!}T_+Lt6w0V+-!^$ms^-buQ;da9^&-3mC2p_oz4Mz6YIl6$iCUUt7O z05>HA^ci9UOTY%M0k07NOT(e;sXSC-mF3c#DUAP~op|Td!S0sT0pZQKz%rpZIS` zx3v<3dR8T@7EdXtGoq1Z(WfB@P(+ko)^5?Q1C9{rd~K=C_>J~{4@x%R!J!Faldl3b z0_@+mjlQcmTdXoRD^S(7KuKy(ZscNKC?~mD_jbA5+m>>&b$)`D91=;ofgcmw{Pv6Z z7Pew-C>MSi_A205Wk8;SH)6Tn@uxQJq{ObE9YUScQI~Jo0>E^VHuS?FMUjMas+1n} zWnhlhaY&5!#o01Q-ontLS92`k3m8Hd-LutDYirf;`kI!C;$!)1~;Z;xd~ zOe$(@M@92CtcID#(lP+T-#&fw%yEa-a4&Lk3yC8GRVvpkJ~vchs6^JP57m%iqkAN_ z{YT<%WKx6B4Cl92v)_(pO33S~zs&7Hr=2=YZwB5fH&A?FfOw4H+r@!c5}r6AVT5+3 zY(HZ=c@2J^-OzD?oF@NK@r0|#zecCg&>AhuR)XNelH!tTp$ZKzI1y5TL_uf1P2Wcj zyjjGFm1-AckA0nX@8};W^ikvrhe*x_ivaxaFI$!^9!!6O6!?h+tWan~x)TU~^^dts z=rkjUqt=1%R{?U!?{46%3wq5hDlhDKD(PMi+(nA@GFcpg8QEseK~Jr zwk=G>@rm!mg9!ELcQ@xun^&?mFCZxQmi8>UfC&S)gqxQwn`3(s6V(X4R_T(!k9+aR zg7qmzP6O+&HYV9^nNL_uKp*x9w#h`Gn_2p|vkeJp;=AGH;j76vA0eNKXPbWa^Nhu1 zc+IfrbcGRo0fAK!-$zeyJNB`mcqJ=qHfzRUF&Vd%`X-MgpU{t%M$2a`igbPljJn6p zi{z-3OMY$U^SO@ks!cXSw63O3W*Ro+kJw{=Lai<)xNoP}`WTcVK|gFfT~5aqkZ6pJ zj@hyI6X)Q=6vi@m0wBWP_I?MLfuBy6!s|dZx;hym+T7X87^qk;&d1nebcr7BDsua9 zW%`B3R|C-dnoh#~llsz|Xoyx37#KnHAs@by(oJM^_F=G4uWXMu}a?NaeFzq_k{?&wq^sPlylEb4&lvsF9qwDKuv#|$$kh|Dh z26&5d$o)DlRPjV^OQl_r)LDT%tVVX#?X#e2uqGsWQ0k&FE82P_q@__%sBwG2m0A^n zZSTqXM_$nMf;|TciST7UW_y20CMJJk;Pf0yDFGFJW&x(qDS8PoNleg~VnGL+g33W4 zNqz5|3xG(q8`jApsqhp=fw@U*OUn|mwkKrt;w|}sdHAKDaKg{)9JH9LU#`y^JWWi* zW9L(2bAI7hg}}r%>*|k#B>q^yw783#1r@Ymb-CHM`U9`u&Q{c^g&DwB!E(3lUf2|M zx@ZPSdGEzt-GPznm339{K8KW}i=qcp=~NwWucvN<3IW*{NIIp48b=Qty-Rq2Z=jnk z`PZkx2&gDH2r9wT@LIp#6tzrddSBlZwmy`^a&;9BBK~}-V|aHW92M2-@+_Rlsyw_@ zYkYsIGRQ@fo$Cc3cGCPqBAd2AFkBG_ErhBjAW!t7{5^CCPhwq^G@m9vfcbQPe;ROv zaE+T-$?l&Do|!*|LQ3Nat-E`+u-MF2x3?d_RxzNLv#FT~3EUl42B};<_QQ3rdbV$? zJtXtjGbU*;U%IY}$thT9>XmE9V9?bTI&2B6%Slw~0|Y|HP1ZWj&{eLW)YWLOKW50; z(niH}jl&iQQKcqp1O1szlWVN}e_)xrO3{QKXrV<2h*~-n6U`KpE$FpBz!E=H>2g!p zi*#>Xi)e!^;70j@ZT#Y`9N{XH!!ZEG{c*2@i;z`Xs(%i!0*8y~_Y#n}kl|)R|8O1e$YOy&GR3 z_o7!M>M!qtUEc{HKdoGE#+>kAQ;ZdR^{h#@T65ZBAWy(^o)CG{g{R{?D0(OHdaphuNhv8AbRk(a7{+6@T1j{{cp>E8ht(r! z*0*(@DIn-hwr{{@rPP^KQZQqAo1_f$d3Uf%1TmI@yYrSL_g|SF=iwlQG1j zXE7!J%8;7FniiMCjq}&A<#az*{bw-gY>JSsK*On(O}x=jou4L{tpp{Dj;>ppP4SkF zPp8A|D-qDb;7Yf}%&95nEOp+i>w?2XloA*qAt_zCY9q^>;F8|t28BLC$w&qW`0l0` zzPFDfBEBE?yUu=jMVBz`%zz$K|Hv4e7&;i8HM^t z9ZF-QA2;sp!a}W_BJ=22XceR{OH2HeNh+;9Fv_3vaBwrzZWqc+#^{!uO&vXKg|{Vx z{>udzCwP1~7g?!!?iw*@T zEti1TIdY_|CVWL-2C>So<6;W^x687h*UpdsDZ-RKt+!nhm{I$1B&)K!(zS!pvle+T zJ`XM1a{0g^);m}c1AY|w#2|JaRqYhK&ceiAgzJiTRv0_>p59@4xIJ5Y_`q2qkma83!^D0HHrz+*Y_u7;E(>hL` z301DtD>i2eRJ2`pF}_hPTF^*rFQlBRo+BQ~gL(S5oR zUSTYkw{3yb6DXFB@1oWX>-s{Rya#}f^&q^RyBb4H5gdc ztoessn?YZH8jXU5c8kVqOdQy82#PwcDyJ>gZP50wJt%##)F|^h17ThV+Oh9kAs4ax zQcpM)*C? zD*{nqhPk$V%DC?iwQcUR_w+n?iS1zM^!P+|PQ;q4o!a7fc3=${gJ19HqoG!Z&LUg@ zCBG9zqwsfH^XiW+zyq0DA4rS}a-2g97dv_j2NnY?1gy`)Jg<!uuS9g=<6Z&*&%D#%K zzP4IYX5iv*uI_Qu10suG!kT|SS0-@oKvjZd-+K`*74z-)qVIIC$Js!|xX=UHr~eQQ zH#6v}&ypr+UMv2OuMangnpk=ybMDydbfC=D z>%kzd6k@W<|84R&YOz0*&{4SA!!h1JDQun3lu$iFNbXL=?{THM#Ob%P^0*G!ZP&7s z2;6*(k0;`RpfW{{xJExrkV+GT=D_DNwWL|E>0;yUa<+>9j%0v>&oZ<)WobiVfLvIQ zSscdsWpvSoOvDAN4(k)?6qmF@MvL_(HE@EIQDg{j%1X6)R#uT(l_8PWjxf>ycAW?X z6C_k^=?pYh!r#=aeyF$??yx^j(^F+VIuxxW7d&X_*Y^cHI`&c(vpl_|pTkRug`s(% z;EJT2h~b_H+v7W*c%JM3sU4G|eNQkVDznmZI8{b@tZF5Cefrsb3_9F}(7$0(D;9}F z`H=HNMRaf~!5=y@{4?LqQP3i26VhHTm(>alipSDuCBs~QjyptLr^}<^4hj0NYkn37 zu9qWbz#G`31kX{4^>BA~=QVE60svP=o4e^$@$g&aN=kfgXDdF@pMSIOdO-t;w#>vl zls->)&ygKdk+5F$JHo%1bASoTKmHRaBo;-`9{J_uS^qAu4ae4f04^@66`Kf^;Dd5u z`9`(N(!=qV;ez%o_P0VcsqE+Oh=%kbCfSpKpeMBL;n~9i<=uBD-8+2VYo~*XncuyW z`YOL}-Mc?9pta^3?iS9zWBZ2iG#Efb|JLALvAu({RsV8uxVh=^hSysS3<(Lj@Oz28Z!VpouYc0hZ3dkhJ~wY9Y> zvs~X3v9|FqF_XfubRv4X_8m_b(SMXADs`5Ze)1M(V_?rwW1%d`zjRP7Z!kvR?tpe_ z-tB_h^Fi2WvTG`?m-(DxY`D6ehIqsUaCW-9`3kwJNU5qj-e%~ZI_Vh~c`z9>t74nN z+L)Smv-$?pQ6V~NC?P#74KG_+u8obB#Hpy^lX4Y8@{If?o=eNiP&j%8zV#o04%h3S zFk)0xHWs^kiOs@h8asX~4S}c4$RumvD?%@a+M$q26+EcmhJ_-n;F7>B4l%Ahg z-bn-^il-MTCVU$svk>YLKNT~Dt)S@hDO1HC5)#n|i0D(sahQbEHV1#rlRqDwORN3! z&syNyU&5je2mg{uMsA$lcRIev@{S@_oj%7k6Xq?k{Qt|!K~Zku$b$ub=8 z{qx>c#UN7pwOEc*$~n7G0wy_Vka`C%WkSUNc?4$5xc=AdC&L#YW{R^n52a#O&8M1i z;g^Nv<?JX+-{ zRuHvU@=uRMe@*zx2z0zVxSBnL%FwC-k*-v~8brVM*lCztN1*opqM8pa2a&mEa#FHZ zo+vk%;3nq8Sz#dk5sO0`=wBY5o~CAE8eFk3_~)*HUe~|>e_VvnIi~+zfEtGWU*|)# zN5s-YC@Hu15L8#^)y$R4nGTh>vzX;^5?Tu~4CIr4=DZNAol8TrMv#d~g)u&k@q3xt83J5-Muv^D-O_S+MyBpZT!(OJ zw2(&O4qwOAHP!X|+gpO=@VDsoFDJFh zgNOI$LZZ2<2TA7Ljd3Thjx62R1Y7&Cs8h@RKNW@ZL8d-Y1mw>clj$H)8L zok%Jp9?yDp4JSw&8~q&`75hQfJQt29ZDgEx=j-U=5|*7cFc`pWeXA4qjt}2u_vw8^ z@9`p_T22jRGyJZ(rKLL>cPHD#++531aIMMXMymjbvN~$Cm7`QH0RKuCCP|KO`R*`dCTW=og$TYzSiPpVG)$hTV-mJ zpDKc0k610AUsTXR(%pjr24ev_9xE#Yy6BFXP4@=$F^y0z{#$&ga`(M<|=!qz<=B0d?jI9IgYQnCS{>-0~NR!NJWUkcU*N zKSZi1Of*~Xy-zD1`IeW{d8$HO$5%(&{NAVmop+F?O`qI34>+-3lp=;{{*xlUASIJM z32enl-`CKN#{;iTzs^2JY_78@Nj-qI?LmA8!TR&euM+@^nhy3)`-`Haj0@(4v+JBBFkx&!X$A(tt=S zJBGA)zcZ3Y7u9Q36HlLNy69SM0_e1Oi&e()+O9Z?zU6)pn9h}WK7t34gACs~fC9_r zz_Dt$y?LlAY?^O6+NeWuapC8i*U*-qU&PtseZ6-~O?^ipH1}lR^=EJY_-+-jS(N{= zmrf&SW@e^P-Yd_tjv$#WlfDWqL>%&P1s|CBZdf)v{{Lk!JFi`P0+C{2oF_CrtY@H7Caa0N6ZRZHl`#?DD71 z;i=Ml^|2~#l$QAwdvgs&Qr}P%@vC4KPR|hwk}_D-2Mq z$~LzPmp8~dH1}t`aPjcCNn8eF)+=>d5YA9R4mf;!`v+>%0mh$+|HUOX#fI0z5yk4F zx>8O96Z!+3XPs;U!D5wuZi5(!1tK&8xl{=28|qG%a;;yVFv&pVGQx;Oh|Bej6PwPLUJ6SbGM%QMr{J8(Zw0?^4i)HWg-AKP$6_+IvxTG4 z(a?wmEf6&Ff2;gHnMXg}1Y8ctmx-m@;%p2}DBx2PLVvx27AVJZvjguJjc`Z9R#BUq z-Fk;p(BmiWda)vf^=|v0R1>i=5rWPZu{x}aUTExM<4}+8@$BFkIhj_2U6ERe@9Uk{ z5_CqaNWIE;W5f4b0o)4RrRQ`7u9INcpzUH|URbiamKCE;YL=quFFO=FpUc1bM^SF` zKe7JH1!S5jpYH6e)TlI7KVjxNw39lZztrb{)M~QZtT9%P2l}Ya?Up685jCtvQ@*8r z{jd!5N~gUjP{`a`u;mvG>^Hg!xhdk}nhiJQ226je_yJgJ4i!45>9%H7LdKLVOgj;b@>>`yWxHa3Hp^lAHyzix`0VK<#*|~$}8lhCRcO30pwk? z7*F$iz9Yro$Z;?0p zA9f3xK3uwzn0iG=2!xi&;JFHOHb)s*S$>ae4b?e`-g4G;JP)tOddt;!S>JQxB&CG; z-5~fwcuFlBzEBR+3;Bxy7}P@2U^ZL$+48uAw$aR>;AFZkHaq*RY15mPfUsTev`Ahf z9n0l(K5P+vua5c8&Q*^x&+J=a0T}DR6*Lo)$K~ZXaG}~?-XSSV?O&G1v`g&D2 zsP#24vJll&Np(1Kvpr*D^p}h#rukj4idUhOR z(Uy>JCl{2W+`;`RQ+xE#NlNW6}#T z8`lm&SPUT$7KRMxN_eGCdnyTU*HU-i>8!2g*tQcogng2s1(ym^RDDm*!X)8#c5sM{ zj^?*Fh<&_GXRL7!(CM`6-=d}o3OZr@IpTGHLBZFaJCVG|sJ0z{A<+Tm+DLt+ux}3_+M5ADxVjiQ)Hy zIR8J?on>1c>$0wKx8M#5?gV!jK!D)EgS)$Xa0wpV-GaNj1|7HscXzkbbFDr1bbI+&x}PfDEZ3uP8ERE@Whj)@UhNh~-wi*52IznsIXJHo;O2l60Ir8rJTLTmf z6w$IkKg3$^L}nYe#q`T;X}|6|~mH?0mJ%xuyi>D6x0lmDuJov8|G^!6* z*eV`NX7FI1T)Ha|tu9mJ$H6gEW#5P=tdvbdLb-p2hs%cjz*MU8ZF40d`x}!(0uuDP zb#isKoh>-5>+kP3Lj*FTfQ!teq+e-iqV;0PFM8RRMZ*(k6RTR_g@8-^GuV;MOn&Zq zqwhZ6{4PhXuX{LBOeLtND{iiC_kLN2J%L`{Lc$yzjFbg$i41s$al0EnM=NfMj?!#@ zmuF^>jb|dAo1+?E0~)PI5(r8y{vQ*>`v6g6Tw)?7F`_+GfSs+0iwG<8Ql1Lw^g_YTGScf^@Fk%nJHpi}$YR$TcZ!JGkQ+Wq4RB zCajqXGkm%65TY{;9{tk>&67?b3>`MY`Ff1KqqFU5kzo6!ZpV|b=M97|lCe}ol8s24 zAKTY1ShUcYTqE`2PqCU%P{$&O-cKCm3VEIS3X5+BLoP)P+o@$z0yWNO)04MwH;M+Ij2iV0Kfup_j^kO@Xc!~|z> zE7NbpZE7n9s?wVTR)^$3`0ex*6sS*jk_V9p)Q4inpur+S^On1wLg1NkyC>$9h&`f| zW5ImEj^@zZ03B!XxfF?3`fYn?Gt-Vm%gbLl-@vAG+6kJDiGR;M*{iS$3UFIZH>aWW z@)mzsj zUl0n92o{8W^l;j1C&I_~ew_9=#dkrkA{Bh<_U-#RTIl`>BiwW{3;uMl!^7~ubGcrD+9v)gueT&J@r}zk2X#1Lxp>5jr zd0}Q0*;#Y(dy}nsn`a z6h-*L+;fnP$X{;VH!w$Tp4|mxncXiTR%3P!6ZheW+)E*UQ0{oH;>XP}(8(h z^cgO_0e?xePy&X7gpUW% zJ>8Q*g}_NIR-`X@XzMQ63WCV;CE3yyTsy$krB&lWl@AlAL`a2f?FL584R>v#kaPA& z^-G(ZQQKHqA`{O~vqy`iR<5qpBsfL=x-JhFHJN-a!9(zElqy&yz1pPc0-_61a#I}2 zhKEpyg=LCr2!HI|?qR5~w1;A{Ch5ueU6ocENqPt|9nKrexct83#dSa**x6H%#-LZw zwo5Cr$(%TibN_R|n)3NP?e@C0lEK*6kL)b;lt4nEydcFakuwr|YS6;5q^fY4!r+@5 zoSdGlG#*1BO}~1%b%(zR^bs|Ai%x&6>yUWMxNj(ep z>C$*fJeD^kZ|V(9|MAa8)L#dFnT+}tRdj)KJFoAc*e%rPRT$LYP<`gYvB;rO&u42G z!a~5^{`|;PpJ zXeK;&9DFMd%D26!W3Rh6N#w6N*|O<(a2hQs<@RoF@25><7EadU&^CRj;{-NCh7j9o z_mv$z7%~h3d&l-s~i12PZA@PV0}c8ZfZQQO#}(FQA-jx6@@T((F_mh_v@^|2}f~ zl*n#u>Z*}a!PUT>|mGaAL9i1tr-uLA;(*nD_trQTnzi4tHQ(2)LU*^L8% zbv0cj^jX*p(-ZsQ2B_6640|uPzt}Efj@FTyvXS-kRL8KPt-Za5I3F(e^-P5ms3WG$ zUE%xySE^hW2KWr6H&rQ^gE4}5S#|oTL;EVo<}o-p@^cpf6+(pXaeFdv=`Lp~iiG_` zP~d60O-Hb&7lSc2XU`1Vqw7BK1e7n}K@wuk5J^#d>e^89%5 zC~$`tnWEHurH*{^qtOtWe6~=VS4^iqcC*X1rLJwGT49%0y6eu9`*ZKM^=jj0k$ho` z=j|3aH(&6t0L7oaWbaRT3}f^SN0Y)4gaomNW|OJN4JR8ts~6L4&Zpn2Y-#bY4nJV6 zdfm|Fa$es(0`HvqfJ;SY*P0YX!cEYByzcv) z|N0eqs+O$gAY}6Th+0`pgPQauLc0?*40A<k~)p_4ZH)ZZ^h{Y;?2ZZSTqVYZ3InEl$Uy zl&UO(d$1|XkpA#!XGL1V?t!uU8?(X8JeL*%_g7Y0`%l$<-Qe`j^*y&?Od`Ph_X!@MJ2Sf62W8Xj{Zr{zhav>q|z5G$cu zPg+;awp`(+NJQSv?&OQN6KE!PjK+ysZUH7#9%2O{jKyYSe zW^yrlP1b);jUV@CHYb~%F%%7K1S(B<`?|c7<_>!@ute^QKxGN2WZ0Sb*ih(R&q;qS zgF?~IjXK5bPy* zxZ&WU=2N^E4@r2q%NCR|mgHx2IfnnO@Arcoo77aF(p;$P{cXlzl~!Bt*Ld8&TEHkW z^pvVbY~#SR;m=P9s`d_sQ08-Z%o%1k@zyz>Q#Eh>TZWo>vdqx%34wvopHV4K8tB+_ z-p*FKW)2{85dCuZXG@JvR6pgMfYidoKY1$W#=bsmPoom4eTV>J$Ba5#?EBB~=&PxQ zMKv6_M2RwOf8~fZm|`Ss{ZR_^R14ePe_~YJ15SN4!&=Y*9#Zgz zUtt3XVq~AC%NzOaQ{K!>4q*P0wnlyYv+%V>3ka7+ zUoqk(I;jNAJfg>+f+GW6KROIN)<)JmDa>=S@2?w1W-c^(K%vCO=o zbwL(^rfPSs=w?V}VHKq=*g08JCe6ya3G?X~UF&ozb2+}#w+hq)9fzC?w%!L18Sil{ z@$)<^T)o>p7g5RJGQD0IRl}|z*qFAzQER9oD20R24R&{z>WXs?{gC`HG|RQZBPRFR)tC=Lfnxdca!4z^@u{lT zPv0d>J!A9Pz5UTJ-;Pp!$E(N^d(!?SrRKg)>|So#+59-Yf>WmMns}iHS0q6;K_FAz zl;0i&Po&>}6ZO`1mfZoWH)XlVcBsX^RC8}{tGNmyq~uy?95r*vLw&040Y9wN@@jJ7 zcMtfgLZjkJ@J}i#nwtHK6a^+{fhJoWBN}q5r7)NFMM${(#_nC@*U+^MUqRUpJUbRL z9vzklB_$atXjE(*i`?0{@zoat@_xW1aL{tz%?`1;6CtzPg+GTv3a#i5j37{RVK7rF z-@%cw`EC1D7E{GSGSo{i%{b|qGWp_}YmoCAFbc)TMf*jZzyE2V$D_R0v{R+pE?0%N zyBTAx`f~+du$AOcVY=r6d3RwU4)QZ%Ds+;gX}r$HtMjy0b|=lspm=(&uKT-F>a&y>Y_<`~vR|x@Z&&3p#FFUFeywffs`p)p0ST5`Vjl=+*pH z6?S%akz1DA+}@WNjc`-455);9lQp(h7OcpmSD*a5pmrPuPmxs+G$X3pSt07rC=;VKZ#3$L!(2Gm zbInIcZd9e}XnQ+3F11Hj0#iO{jB=2+reB&;>FyH@K`YUvvkw&|HIrcBgD3?ddq-BO z(1Q21i7JIrCy-WO&#lQ<3ChaJA)?zcVj$>JmwKk)=_^IhKmbJ%t-ISzQGUF##%svi zTSLQ0Vg-}4#MepPF(s}!zQb?JxH$;1v^f6;y4aYNHBImlV zx%uk!IQ(wbr?V|uPuBsSTRU+#^|L`iJf2tu^U%+3F~q!Ro+U`ye%N#$02P>@@RaXc zoPOckM&N)B-@V`OFDx9%dfu}bDf!dv(@-d5>h(2I=KQO8Ofo$U3`(ikV>*-P2HuI5 z;TkJ~i-O!?(_ts#{TvD5GE=84{HP+XYSi_F;>AxuV}XF-?wm&0z(RT-qW! zx&$imOC{{{xdS<&u7YK=PrLQiP}D117S+yvzK)K+lHCw1c}ewh5{`uvls-1$@Wm?` z#A@=vKlDp8(ocFeab-S|V*p(PsGoA8-qvcS)FlnMxdsV{3`LJ@46j!+rNhjn=OJfUpAGRMb)FTz2cZ?Dabg z>Ms@svBvD@O-n}dAGDk5koS-GMSPteNL-%roZ2r!itnB%_;Nd#CqEM9!RKg0PxhpJ9c zxX%PXTM6LiCcym25vhIo)od!N!Rq<@(UB2A0Ipf^cn5t9LLx9@#gsKRdp}8iYlQ_j ziC87Zr`N6b=a91(6cmaOeIQmz?ijI=TJ`Mr1Q12P%Q;!XoY5BhO%8AC(-SyU{P|i{ zqU!V{{IRq=o(?bcph0Z!;u}&Te^mTRj{ZDT>}HSSr+KF1;w~Yx;f=Uhull|bZ%a`B z{oS4E`Q_CW=MOv<7P32IX9bqi!#{Mmk2ff}HY+)^g6ofm74xVS{u>4Cvq8mS1IM3Dp$pI8CpVX+%%)#{Mes!Jr*ebYAM3W-ufw z#NJ&}*?WF^3gmLXIq1QnL7NfwKS#h8RSquhXB&gdz zZLJZ5^ziUZukECvvSN5hEW?YcR1}5JySTbCmDtD3WEi`UVr8c)7GQshCpe(X6hC$? zfE0YQhp-Y&ckj|zTBp%F=FalOUlhl`;r_jzWSc><{g`a{%)395$l05SNBL3?CE zO`NgFtjhNj)=)N|PX=22FobRb&>wF^$1T|{UX@kgW=3;8T`tpUi`Htg2^z6|C<_eS zzB#OOa;%*%e;XNVZIif#{3zQ15)FhGEs zaHiOvSk%31MGJ?)AXENEK`CUjSREb_L1Yt2-RdrLMoLPJBZ`$`*J&(_3;gFn4oQ2Uvp+a2fpIY?BC#%w=n?Aqs9Svf?VnQK`7mG31S3}IqOx?83COuj)(&7?Xc2?2ZN0*jHMQ{}?2gfu0=P=l8dilzl z{@Nm)dqsZF{ciTxc{&8=>IZa+BWCyaID_ zaZynKKLQMoC%QR_-9)Oe3?66%xLF1u)9{M7pj*eVG`UgI#Pi8=70umWfP^wPgw1K# z#VYs~=tLAnNh+RzhK$hQ)@9Z^q3=&uJ&yNO`(*sx1#ON1t##ph9uM|oggy_$Gc4OMBqYl62FYKQrN9cZGJqK3Z%py+_wkYejBu8=NfP)_qW$m)PuT;<1pf~tA(EMEjIy5>E%^Ss=xESdW55ns!oJSkSNFU_e z8Y79Vb;6kj0ol-)ZK*$885na_7IV-(RK;!!lJ!(QR5bkYQ*gQZ4^K}7(LN0@*Gzt-o()RW$%Tc* zoMvCK!SfBatMZJ@tSFjjPitEWNzr7&!5vN;-MKk&5)zg}q+DUR- z;ef&4#F*Mm>e+APj;kGBGZ&T{(4~Odb93m#tjzJKELZ$DtqUpL(K$pC0NseCnKJw{@bXx zkJatj>N(h;>7Yc){P_8X0AV#|bJ#^J=Q6k$u7iqShNzfGTtP9OgHUEQ5Cwc;^SAjZ z>buiWWNN+Xc-Nk?7l4{P-S*Kaq&jTA5V-TB_N70#CxVOVcYl`7nCf*ng7W92Jmf@RlHt ziOI=8hCmYM_cxvTW&$jtIk6c$$G6^rx<^vLI_4`=V^cqj^%j}?lmlA(2Mw(( zITN9WMbO;Ij2pyouwi(C!!|VtLk`!}UVtF8BH zWh(b)>kdv1=inL{2~R>n0*#pMA++FmoN*u2Go!oldZ9G6jqfc zn*|UxOOPgFlmY;4$g1JX{;LIaE4u8Bg`b>Q^L(b{Ng(m4omGH2S!{W{U0Ba}S|$RB zp(H8FoG*clwRY!d`S4oLcX~y)D>X*fW8dxkU7Zg{U=yX6%D;{Rx)%|5#m9L(IM$@+ zhbOEE1H*dF;hODG^7X-}Oz*4NsSIGWYz2A9+vPsX&EhIR@%WEip5A8zkZAyw!_M%f z;uXkvmQH0-Ipr5I&TZ_LJ+-{=>OvLn4QCETjhzF4LZ#YfFb$XQv^)nNgv4~-?s>&f zkgDm1~>DQ0W5* zw4Ams`%OmL)ZAMlXY=*Pu|q70Q0gN{p_2KEWi$dlfRo`B`p8V4BVU$eLqL;7I-jJ$ z1@z*znr!`;w*z4#s4(enruHW?;S3<;Y3B7iA8V`QYT9Q)JNGtv1fk(jR7>YJ_TgXQ z+qV=6Wwa}5`u`(IKw527Z{~NfJo@Ex^6(TW7oROF7PhjAPJR9cmb^I^zM0DR0vZ=O ztv9p{=;PkVIiUs>e_MR6kJ;B?!nd9pMx_%mk2<@4oVpyV;b$Rt+yT-Xe)#OvRHrWYxS1cyA2CwpP!iHQF|HF0t1?>e2|XI~y4het#ht%LDbeoW%61+mcR5L+TbAM;K! zVc%xAb63q?KS||bNkGJR6M6p#_NEk))3|$%k=-`TayeF)O<}bGZ*1zTL)#d|q>J5qJt4>zzz}+CR%~Nqn$x59j!_D5waqSmy zoR2;Qv7qKNm$I_5>-k#@iCQJL2;x$M6;ar;(U4pS?)%Z4%>B8fY_?GP8xFA_W;hSa zq_GvUAyx)j3|usnr-H5g#GahhcM2lTbnXfwK4Z!-PC+kA=&&ze1OtrA64bPUEh=@o zpgi#wKY+|fCnf-_gSM{7$=aHOhoSZU1TZzY;~rRJW(|HuguXj@tFl@k?eIbiyJ&52 zS%I?}tt%aaFCGV!?MOmi-;9Yx4?u~0K)eF00J%X3s{eFKT-@2&S#DEE6{)_tUfbFF zR6b4PIrzJJoDmxFUz0&)hrJsI(qwozb$3p2{t;#mRI9Y{O$zu2C{X<1cy4?Wt~5H? zGCWD+w_ptE;ooyV&lZhCggKo-d*(^^U0i zKqU~k0+hjUTc~RQD3QEtj)do7ob?oI*dTdbjflWa^O9@BcYDyq-90 zw+H6{zJi$3kT`iL*s;yznvZWbv>|wNv+p?&4mFTIrzuGPJD}t=TI+E8{ei+&~t+y>s{lpB?{}#BGxskY5UWyT?ScX&TnD=n5d~#_f%3aoNpdPRD0t-0bUhO{Wo+ zL;xio?A`oY$Fmq`P%)3s6OS_LIP}+mD_-x16#2}y`&rb*m39**bBaOq#@hZPhxfNk zmadm(-PNa@uK{#S%tEeeUn7_Z24t@jf2CO)P#Ad>|F}E(%A5>8VQh|`tK{c!-F2ES zcQ}?rql$hFC2MJUb$hH%zW;pVkR;t8imhyoMz*xSu>q0P%&gP8zrTC>x23@+l=&t> z#p?4q)PaEp!)CDUa&wW)0x5-L)WWXiR^vMtfsS+BLDJ7}^f&;!TG)}Ph$rG|80y?= zzgaI`ak65jubgRwkwuu`tvGS0w>17isWeBtGYV1Z`{v~%^8|D1vOV6Q(N}){Ha=ed zz@Eq~NNdvSw*U{!o1m@D-my(Revv|zx}i99Vh-ZvszMpz8vEhckFoj=Ape?DI_$-Z zoIu9ldB}M~R+VJFucoY-nx}GfKk-$2qoC|&)Xc3cd84Kwp|o?$!K~Os?N)MMl|P8s zLmEuTQ$oSVrIs62jjFd|GwZ(0tqJloYYRY^2E0us=qRSo8P=3mJH~2OL)9?^sO~Pw z?)5>Bn9K)zb12hmC|<8?CB$d8SkA2%MMJFfBOK^?Uw)VhwhBPjq|JuB`vlLrn^v1H zf^_Y8&G?7lRHnvlHa%$&B7m{Yrl#gYfSOSeO4k_H62gWs5i&zqL&bMGO<_29W6~Iw z&!4R(e)Vo_Z2|3-Y=0pOQ?A2NXqWz((V-9)DLI*Wt7JwC3`XOLj7Lm|p0tK9Z}mW# z0^BsGE5O{^`F+-(x*+-wfHYsv8v7{05b#JHobX#NM7FWXZ5Fm}P9xnqZkUu2O5z$) z7h^Vg+QlH2l0U9vx1!!;d;ys}R8X*^n`8)^-V+1qfVa)fj-s$v_FC1DLd;0D7@ZgW z-caq0wyXemC=B9y?U}5&u&7+G>(_w@rU{OzCPY5|y!5aKYn+B* zbf}ky*K===Md_a+LJtK6JDYM-n`%G97_|@}7(#$8DoW6TsgFyN8^=PS1tQ-Y^Y%ee*>nvJkV)1xK25@H!Mh>W+bg}AL)GX96Nnm)?z>5nWQ z(k5w(bc3iWm~Nu!N~zFgJqc_S(WMeozkM>TO*>Gsa_1aHj!Ye68Guexpr5+EB;j|x zy}DU`^|ndDtl^RleGEgRDvAX?D_zNwmeVl%EnBzjS)XG0s~EK0ncG;Y{hW^hRrbwR zB^h(l%8(0k+fB7<Jp+OE7tj zB2k*lT0_SW-yT@;g27{qO-B$nyf!+MuDHw^vb!H4_L~ZOz`@N8e!FSc$f}Hwi>ueq zkB_hED`LEAeZ&W3o79cfIlrTOag(t8ctBNSj)^<1Lb-qklGs|YE1(J=MJn8_L4s8( zg%HkdYYPQ@3~Fj2wx0u&N!B&6>|I=x=~Ycb^vR`mA^)FkPiO>RF#czi)c;reqAI(4 zBFGVxIyROUz3TB#=Ui*;_P?X~S3E{%1*g^!7G|h?u!i9R2sohlRV!01iUu;m&ayJ# zqEOdBxVcTjp%T(iQDKf^grB5y*qrSiX8!smt_vkeQydbNMolBga8A@g+3*70-tJvj z%O*w8(}WiieS|I4A{n3h+M%^jckMh8C8Ii;NR>rfi`Bx@CS9N;%2B{*{}zv*G9*ybXVI--;q0cqy3;~5fq3E|YZchSf#%ndb*y60Cstg^;N8jq& z{PdBb5b=m`S-B5Eu>ewwhbvvEga~+mJP^TRfN=+PDCe%hvwak6c_#g?bMwN%PJrjK zzsjQqPd4-R>DX?rLQ5fks$|YmQ650~bvqnCb8;@dbQ}eT{4uV5Xx30F0s_V5WbyVg zA0Eh&i=r*&>w}dFBcr1^Ttol|N?tx0z^L@x^SK1@J0EzyK2aGGu$y)0#Ag3^zC*oo zboX>eJrLd*yk*$+$xki-X^1H(RF#y37BF7e;LRJZvHEw!t~uKQTMafqHg<1~G>2Qz z%4Uzx@X^Ga6FqC+s2N}cAGZeTBx6Vm5}EJH>XU3&J+t|pK2cCawZoh^+J4yHn9lSm9t}{rR;- zuI5T0o^LIV;#na}0Ld4QMWG64LQiC|+g%qm#E=QIIsq{BvVV&1o|1~A^fdRnx9MAo z&ASc950Nh4!1rf6ii$7VZLWwy{cPIW+8rIau3RuOC-d7Z6AeX^G`n9CPkE&3X1^7QGYWZ~QPRE9Gr+j=HUZ&Q zswaHX?Z5irgPRoZ^@#1w$OYE#E)?ZOf&(9U?^;UP4<4D)j5{g39TxF>fP*QDkh4Mj zBV%zZGvGml61Im96b#c(i?h9bo4_{eqT*iV2)pB*2LU`i;}6vlk|CO!yk! zx9bpe2gF^}+j6_d@yW>;V0RK3p{1n-NYTCP2c6bC7)ewgk;$yr5Hyliz_DazmVA&R z#bf<+v((db>+SQ}?GNpluR>qbPzPl0JJkk8fPw-!A>kQj3V?Q=_XOrFm~v?kd;^7! zPEG(h^Ex}5@F%|b{*WDF?C%n9YCN?Z8Jn>RU?s@QBO66RCLt!?pLpHa8@G}!bOdB@ zU*ET!9DsYtwOv)L(U48=d%h#V#=yAQ`O4uPsuD#i_z!>s>F;8g-G9;C?r|;T=GF#9cTI_@B>2U$18)5fhgdgvI<>h05zA~A?Yml-T!X1E&M#ScM5c`cutI5UD zSxA#f;$JNwIy^it#@pylJ6QBH(5L|ghDiQxWB0miQ?Jwx;t#;QJ!?Y-RBemFz^ z&uDTi1|B}XfWeCj55A+WCT-zrdu&vL<(%zO?d`$m_&lS$t1HKq##0`{od=$%F!!k3*cOnZIdilZ*9@?E9Hq%rTzEbvWE~m>*K>J4NRR0@PXBxkoZBYUB zwYztRdTGa(_&u@EC{pkbbA1teQ#CjNWS_Xf>K_Ls06rfY5<*2ov%kOJBkw-8vty>y zc1I}WE!!8eY|wY(OPY=D!DiUsT)FBgIwAf=Fr*jm`8Xs9FaWu1bQ>s|PERWVKkdy3 zZ_JPp63tX%R8OE}j~;EIq7`=eWd=ZKpSRa_4-5cSh}a~i741&%tb^as38|g0aL-SV zG?e{;{0Tyf<^)cTj{ewaWI{>u9)QZgqQ|F4&I}$E%;!ETl+n09>+tb@+1fJ1K{^d{ z=rCmUxH^vg=HbD6F$1tauijQug_RcZxPZLV>2@AHRa>B*zrCHUp|Mx7+lTQ58h*P| zUMQ63inEw)j2AIC_GG0COV@n5u0KP&f}nS3#ci+Ljai3Jk_@6k8juYEMa!U`j=el` zVxki43b@b|(m0?A8O#-AIic*8-5s1mhHhMN>_OApTa8s*mXSB3aLeHpk{vi3i8dWs(J?m!p&&Fc9c`$7o*EOhlm6@ik)EA z@d~s>Lqb0I8pi0EFXYyhm9<56@AdzofC0r?!h`}3$f#=eT-52sR#?o4miLXU^z^o< zsQD!XLz6mFrk?j|r~RbZ)>{y6Xy)_X8BF+HPv0{USLOUyn%M^jWx%9kaQ)Jv3=M|{ z+XIUh$l%wGPmI~>GqlIwWarRA?B*7OM$&e&+;F;)+X2`}Ql6H}trnPB^e*9(`$%~7 zd8UpkmyKwnUDIMa0)ag{I><;@3*|alsi}Yo$@u3qm-gM=o$IF$-iTIU4D5x|74G-P z{YlG2%h_QE@)l7-W5*@HjN^K;h%$s=6BIqcEY3Y_*&AfR%xtUGSkCY5P8Qb4S9k+k zqmXxZwhF1Y;TKO6lj`ZJjE?+iyFYve6#u>EC5EPEDknH?M7M*;-^g_<$Rc`TBE6|x z(9v9&63E5?F0Sse1)*%>;rHy%e5cPButk9KS01d%rj_> zM9@)GyuTu`>2(#_l4f;dYz)iR@{#8w%OmE%vAy66YvB9ka*Pg7p7<9-4;((-mRT*ljpp(Ihb(kORtl{VF`C;}P-%{1xmMH5$r4r$ z7!wPBt;Fpda$Kuk$kX{9(4J0DNxC;mjVWt1v zTTCq1^O$@3wmVij{7zk-o}Oa8dHJ6}a+Yg(I2~A|GdlMO99&#bKn`LEPDuS7>AW(R zFPROrB}s(pCsimm&dzuprfQGSpMFEYL_iE6bhVd{ zXusu-(UPk&Xb=Ka`0a;j15_Ab(6UT@clk9xe`X?I4`7P|@mJoU}r--Udp+X4>oYL%h$_Y@=fDL@FI2*1^Hu&5t%1j?GP z{?-u@Atzzl1AM}Pp?6?Bz;ZYGcwKC+uCz51)N;-U+y}!mOJh~8y3{{(TCS=-Pz+9_ z-J;R@qm!+5IX*(QEBC1{)fg=x@5|570RjvifHI700Z@BApu3VY&rrx?We>{q8)6tA zHhJ$)RA_Nm^(Oa8BCTiM=f>BuBY^i=G9YgeRpb`Yr@gDzUc==h6s>0O&sJfFR{z9H z0=Ohn(>%jIW64BjE8w3?BUQ`%l`-4kWtR(XtX7nMFIOsc4p%oZE)tsx7R=;#gI^#J zOm;bLyd`7N+|py{R(ZXuBH1|IzP{zoo{ERdCHGTaHJ>N=8y29g+2w1ZGE4E`0*I(& zOfSTh5;391c+mD=gXeARLt?xA0M`ARuCCCAHB9Ko*|^oVMt_!3mG7U*30Et$?2t#> z8y=RrZ{^@(EIrPKY@pvdW4b^(TWS+4$->9eLC9e0LrIB_Z47 z4n`0W0(`pZ2R~_zEsp0;Q@m=9Yp3pQ>oGi^2$uf%1_1_Fz*m%~bC9#L-h#~Z8;Hv2O_Lom&!Uro}WG%xlL3=NktX>mZw5L!>*+B_b-CX z$MzrR%)?8-10TW|4u47Ow$I)jhkyGRP$fJ-0cEJjN)CMc}MPB zE8o07CDl0=jC!48b91QPet^+9h4c6n&|3+k4R3F6+ngVX>rye^`zICBuP;oOo)t29 z#=I5en0{^`jbJYFh`XZYU<5pWE|Hfy1^CkqxwJ<6>IEU8Fkm*;At0W&dfo5kOQMsP zgE zySspW5B7C_=?}qA7<0v<5zID;bh>MnKkNj>(H7kL8U24TXlO;B(NEw^WEtc9h`7pq z`@DdPY?M`%07u&9RoW}dzlac$Z1`wD72U6MNDU#i)%v&1XH5hd##DXftrDY?Vg*_dH{f~z?nOM+qXfQAeVdEq3izwb&;^SBp(??qy zHMKo}J>FmkQqR+6H!c+f4*}oHgfjP5?XB)~Rl*;2gtrkVGXj$%K;F-7@%LrzwJj%X za8Mku7oxZZu)Yq}k`aY)m#*h*S^>@1bLZU=;Y4Zy8O5X=ng={H?&?bl9 zr`eTFA}6 zppY*iHyGu0I$Q&NbHO?yvH_i1s*vq};atx?KJ}+G$?g=`i7KnJG+USMslgMOwISK} z=6gM;YM55A(+|uTIKEM>utwFzQ~I0Q2Z0o+RA@exCwGEnlM_)gEO{||Mw;byZ!vVj zY55n$nVve#d!e!;at&?za9f>${>kARpO01IoU~6&W5%Tg)(CqT8xk$@?vz*LkS8p< zYT5j0qqXkn$ycc`bYx^Ru2l8ZgoK0b{h%oeY^pECLkan0<%;F;GLm`6uE>nE-du*F)A$BcQZb6GU4drwi_;A!y$?F zT?_(G)_iC$d@5>cditjiSS8^S&(l$fpY74^FTB4|noH6~J;I8T8{}YkdOcH9Q-jfu zcTb(B#j#^k4u*1Xdxd9JV%b%W5E3VD|uri z@ftXEVSd~OK7ZIym;5JGsPI+bfPkqje=sw9V|~IpNj4GYbgIl8UK?M9=6eH5M!9S* zBEGlxw-s?`xT_U6J2VomW#IfSSewvkWqG@?;wu1YYFYx!6TD<>SxK=1782F;$3RKH z4cP3Shn2%kNMr)1l}@`8p7* z12$U^76z!j)* zy}>+ZOGP2G#zq1iowZIMG3ZKzRe0jVLjnuHv@SC#o6N97vg`(|7w^@wo%;k1X>0i&dt{fES<+xq@Zy1gPO386Zh`&czII<_pmTQNnfudY@ z1lrr0+dd0m6>l?;hVh0>Fxcxuu_$WsxK8W(Arcy`(`}~!=!Qi(p+ImF@uJ7bD3d-r zr_niAjMuq*ldeOq!~8Y*YNbmA!{>5>L>5ckvwKC!j9cFWhf-V=;>X>G5G?s=EJ~0o(!euqlWiL0?y0rm@XJzetJH~<0O}Y7ha#E#e*p< z+#fHA+H{kn0N($1)OoY_P{;?aqYKzmrRO_FjH0xf#yJX*Kg|c*EOerRzAqQu{+R7W zMOmu{ebh8mXUh!@Zs-1eAyFYHyq-5A>z%l;`iitWR%-9;u49zVSQL6MSC(Qhi6#&B6<5R#4E0g~aXHr4?eU}XGWk^Lr+4OepOZzp9QSVi zX>$(XXA*EZ;(29wR$XiM1%^|Z^!AqmvwR3r54D=CrRF^l>{eUPPEPm$Ibji#kG9CZ z8HVkh39f8@I&bd~A>Sx2D6-Y)`jxW#bSpE#JP}VVkGHfESPOi6`Q>o4-;n_4X?as} zna%vB`6=<_!R|U?!4zZE4;c^}9oE{>bM1w^lA2A&g^M`mN>YYDN6R}u5abcXqMVWQ z1Z!v~N$cz}8R*fU>yn!JRbZ#ru z0n%EBFxS-3N=0|TVnI3|1}uA`t1Dm=&E#B05P{TfxrzaL8IKS1d_H^b06i9jSVbWB z`#;sd-rg=gQ>t~hPjW`a4tcFv5naH!!}0vB&l4IUuOTr{RF!Uv``KTMK4Tb|`A7X8 zL(}VnslXGuW^k{c&qom=SK2h-$t;)qPxm>1JPg)PON|M0V^V-=9K*x#hj8D5Kfr+K zaY}&EasM5g7BH*$&(OI4j8F?8mqIOMqYqjxS;Gz$`*t&;jBPRiHUIBlk|)BT<~(N; zzE!=x{Z6)Gj*~9H5Ph@fA!Z)g*p8i zh2#gM2GIZX|3H>#IdAqY?ISrbs_mk~yLnt;|9<>&(K0N@=7PLZsbCE6f)dn!zoioU z)L~HjPZB=OIHZs8Ijleb_ZxF!S4gPj#&I&7tT`Kh`}aB^Coma`_JtOox3 zp~#7Uf65HPaW|TU11{r#`;qkwEU`#=TlH)xR6u_C-xtV>U>2hfL()niHH!H6BLtY2 z0@Z3O!vFrGRPb*JoP&?zK$7r3oy`J?|EtRhqQ^zK4gcpARpEf?g6isc9hpWFD0i1v zlvTv&=uI^gp(?$q!3;LNz<<+!Oc-wZT#E1@AoQOX>u1~R5-j_1ZVOBW`#(&*bySpX z*FH?AbSd2oEeJ?TcMk(dD%hHzYFj8`PS$1FPC$jx%au_ z*!$?dF&!)c7~{70>(VZjv24;audV^YBmKHlFwpqK;KT;#M>#$+Qh4KZyuicFePj<| zsR~XUCAR%o_g!{5lQShf0SEK?dVY+I>t5}E4gUQ6JPa{G`eoQ5k|3}s6C5dL_{%(@ z8J`30$CCMd0Fud;8w#A$yLW|3ADY$2A5p|0JGlheaPCL zkZs4!;L=uyQkr1B6Hp!@RRSUB^GSVPZIL#`4C%3Y+4e11%t~osnlW;U{g~FF!H!N|QOF7c9d@SH=NAt{{D5gIS9A)_eu!7I!t9{IPfEf^a0Q&9}cfKIHjf zzXmHKBYhYB`RbtIj}+F^@o;oyCf`JjC4&b;ddqdzUp--0lxG*9`?G9*CD6auegE{D z#*@eUZ14ft>mQB-ovWVpz0Th5EnNlNWd| z`X3uC|F#G8uggOe4zr)z(zr3BBYc}p|HkThVr_2`ncYbDN4@6H)|!?A+PQ~+n;Ro| zVJ%fPe&ywU%d1TvZ)*Idi)i7y`mpOlcDr|o(pTor?d22pb|(viX(^0~wPy4*pqnPp zk)J@)72`XG@3UTAZ2MIka@Dqw zu35zvE=eUDn?J!A+B~+KM@L(qY_TW=1a!c-7(KO_wX#OiZ#_Hx2+0ZMm;+TglrN(? z9Ioguo|dk3o;+OV_47q{f%%yg$UmT@EZU=@Vvx+hw;e3jY*zX{KFw1k>b5!fG8?&n-siKJ+WvuPI8PG=51G%eCKRdTy9py)m+V|c)#7P7Jm*1iLNV> z7IecW*NE04CKrqs6@^_-4LDJXiUK%_>?U{Ccr38_y<~h1ISi|Mdq$yA)bO|WZYjd6 zOMz4<{xb@tHDbH#8YCYyGN>KS@*2JTXY4>x0Xi_E*atXEEf>}j2qkBsx)!SGoM{R6 zN81?yn*cx&*!>y)p_x~p$8=>k=VT}Sz~m=jUDSonP_asA%g#@ z_0iu{meJeO`Qp?r??&+3H=DJ>0$)~P74$r#tWGlD=lkiHZ2oL^43d?T`=*|&eWaA; z;TGFbt0mr*$;r>%$s27guB&x7Zx03C_gg2VZU;jy*5?2*@%eZ^INfXd-o~=0&FlH> zQ2^)0^_cH=RN0rs>o03Z+$DeTLo~H0Y$l&IwIEKu^?ica2Tnusa zL{|Vq=PH7`GyYcbadGTU@Hfd9FNUwB!g8iS5+TBw}a*m8-3pM z>qB$&4_f!9stF}6UJuh0!rm8)v+s}#72n)!HT-R{yCXhSP#jObyY~Mv5%UMR{^-dz zvOL4?XF4L-eB*1okU+g{+vW1#E)HZ&iVp9aF3l`&2^Ja-j=cj4ABy)LD4U(a9lj0T zH$E5rI?%VkBMnx2u@+?|J9nThG*K>C+otp%ebC%c=-kbTXXQsNL`ntMshqCH$Gvvln2Sz6itks_T*!v;?7 zULTZm-0WXB{^@K>%*YUDfjN$YLaiIu<}^@j6t4C3al>#^pm0)S3>)uF2QC0aly-D&F2392u)MhPdt2Z+neQHnXaTOw{{jSb^HC}uq| zqBpd1hlHsxTqk;th8GO1Oj7j!XHf+E0}=B7GfQMJ5dVWzxl6va+)@pk!j4>(^8iE; z<$VADelwIr+z>_mc^{3~5 zv49sX%$GYUD?>?A=Njkgq87wg+?1f<<^kz!O70VirJFF5WV$2-x)nn|EjEQJA<|}g zIXCB;i3*HftW|w3!-cQ~#q|G)kjP!p{lhT|^sTJ{vL!FmhTxc`pd_@N9LW1y%?e`1 zmV&~<7aU|A8PxHlXI~3%k>J!bWn2dG18am*T*rh2Q=1qln=}S6b!udkQGO7MgUCpR z@y$ClE~{Cilwr`y&=nfui`i0i0PTws22DEt-uGO1&xFPrk#LzA8yUr4vH;jyJycZt z5uhB}X>XyVEHA)FN$DC$?^YVLC;(D`o$ru08JUU;-q&MA?`2}zX*wcp7Mwt(V=52a z?43@xG?=&v2Pw;%bD1ehW9M7I@UMD0qX{&Fc<~WkDi9c@OL8~~h^ayXLuZ^NO9!9B z8%~yg$~O^nxv6{bZbI?qvNzHfFF;CFI`3`ga(zDDX8!v7bdq#}){6*$pBcs(AdSsU zH+w?s_jU5ZUVpdRLIs{4uLXTKEe9se&#gV9n&9XQK4hJUx+&a( z&k85Nrb2NtEOw|S4lT7N=Cq(C>xIL7h)ig;qQAx=IjKV$cGdVRx`DcvSL!tl9Bw)^{qnqO@wod9#2|pcx68@O_Qu|9zU;gEc_2q4 ztcr)?Ei~Mo8L$^=zOVQ+_)-RArY^0y`f! zZxX}vbiGVudKgI{X-YSOPE9Qgd`!uHCt*;o6dI{xQV1s)Sf=sp43v6Yle@Yqn=IVD3qLQT@j(i1c)o)Df%7dspxmX1QaF z3n-~@PIFEhuJtZ($CNV1k!b(&B+5Y&bd6%elc;v0E>&yoHT zXtiyNfRvIh)%fV>m+O4f{FbJc9wHr`*o+Ji4b4-Z^4^}(*q9*sPiX+DSR(gUj#3?A z!#?!*@63fkS<3fpRNN9tUHRzxEenQKY=2HX;3)9TD%W0t5=v`Vg@?z764ZnPWu147 z9yDZ2@GYQeX`cE@RRg(Ae0(+)$Ct`Wiis7;X=yzhb^so0dU{%}m2+)QbyGXy6f9#9V(PPX(w_~Zdz2m|2-J9^j8U-4udTwQDi1JVR>wzsgZO{ zDu$$elUEK#i9}l%Eki-O^=AR10tu50VOjBF2E+Y4%DslFdLB$0K|j`#*w7kWS(5^= z6CZBaf)vR3#7{8{VksXZ(|7sB$A|nP-lk(Fj+r?hG49aVki%tcHR$dQF*J^n`%QSY z>sd|dCM8N_5 zNYEuQZ9xm(Pr6^vkm;#M=_=+pC$h0xV9d^|-C46Rj*KImpo=oSfr?R$jL5j!*=?S= z$(h8`Zyz4UrNp8c0D&8b9JD50{->iaOSEQl;LW0}amejUc^G+JX{ zcebmleY@bmjfLLWe3&aN6DhI$)$9j`sFSrR-M-@P4q)!^gC?`rx|9&-YHnU$_oA3h(BPF1xo0z_c_Qe;?pAsJ@ zgyk?4$E*4}wV8}rNv)|C7PNBXZ~P>*Y3Fs7Y>Lac{T0({qYFQWx=+`m7f8G!lqjf- zn|fr8ZIUa_`&4Y%;n5~6{jS%}%@sRRH#;azha^(vnJQ)N{wnj8lJZb;(YPURV`SR_ zAeEPu$XG|-)E zxbVlbYpCA^eWNmV>M6=pXa%kjVqkM4@A=Jr!r&(mZ4nmPMcsOzsyDXP5^OLsORe|HDpaV~e&$oJU))N$oFiUeM=dVC z2^Lnx@Bh_9Bg6{*_$@%k`Zg`gJm?zaa2CH=U=Bg^pD_lE(=uP5FM^`j%z%Abh7S%> zP{7Whg?)PN{$y@xs$RGf{szs@rIgWCBKwVoUX|!3SUxT-SA$UmfOxU_7Dcfv)++5X zK&;*97m|$wf*k|do!0H{IPD2VEqKI`u?iG&6Eux&yLpj*-R>I@%b+{^RpO5`_$=FT z!O6v!U~?kxN!%d(ZA%%sfYAXBe4IbbyX4gOI0q$ran>AN5?8wd$) z^G9|K^OxX?OrO}^;n3&neKDx>uMP#t{V>SoHv%v;he4ct5c@IhlbU9JbK&lKrgy4MD^asGbgI0z-~Z{_oeLMTlll&nx{)#b}yF&53JU>vE>(l&-` zu+AMSZD6(7Bo+M*6ezY>E?v1}^3_p?M}{hhkG5LgnEI<|QlduXmr~)|ejSKlH^UKO z!nkBP#vBVD7eh0m(|y+^xlzP`d;EL7dyFl-X7O=4MXL70Dc$PGY>*v;j0r+gyCil# zQkR)G9FcF1<=03{)<`+6?HEV0trO456o zWM~?R{ukTc0zazf@k-#2Ab`9Mz(whb_p9~Ea&sdd9TaB1=ft<)E@`uxiSAq3>A;18 zr9hXtlmxGg(r!ycH`=@+TBT(!P(#0pPob5F z#M0}_=?!i_$5XPX49&{X)r-j><{hvm1rRe&?2pQf5o&`OhB#R9lestkXCWfOx>T5V_b>`De=RBPiHhCu8Jc3m zN#v-{YpWh~_)3r(m~QXFMl&I>!#+Wj1mf8$TkMd6;qo7&Tg;3UB+hyQvs>vzXgTo8 z7|%bmGEpX%v{SnH0Od4;pUnVg9rIOKMdQSO+prbXnbd z8ZwfwKU2S)@}SRocQ#g|6b;9A0fsV26@=A~GxI7PT9RdJPnEnPotS5uI2E>w45Ng97H?Ik}NBk=ZG73CCl@mb_eUi)Q+mY zte$<1Q(ClEr>JSiaSa0oa`X=%ilCpVJd!$=pwgPOxvxQaQn8w^^wz1*n{*G-K9fXk zyR4!M#pRX``d81A|C2;ivrWG7+v{UGa_KWGlmZ<=B9kASVhpd$@`fK{DbwBOm3pUd zmrGl%XV|>z$fS>`R36aOE4-W~$Qy3dL-5~(To(fmH{pcnHT>ZWm|1$QgnLO=m-hs% z1UtLWN4IlJr;^SN1U&YHtYrZ~4f|h*l>iHxMn$f=PL4c8er9`<6Vm{s=q(Y$DM{3* z&TQB#$4g@yPRw5{E?N2$lL@v!Hqr5T%ki^sdaX>b$QmWMZQ>T4I~&8|JE=3Jnp!03 z?oTvJgmbwZacjV~F&;!snkr@?Z+dmzDu}qFM4mMDNnXG21zr(j5xPOiyb$lSubC7K z&nuXdf|r9l-0rIqpj28NoIAfd?SUpMk*R^Y5(~bZz5ujDEsaSu4}XG2y2*3f?cOXt zPb3d-2;c!-wV|mK!%FYyGc+c?KQEmm;N4$-y?nASuV$P`DC|ly`dcO^Ri!c677G9G zsuGOAO`ImtV59yDaC2d+{?4nVA=VW?9B1mPn-+z8$Ia=%Ha}Ogz8Be@Aia@V>3QG-aNBg@Kyh(&q}er@$62SW$VO$))g?V z)6LZwYJ?NOTvK{v`Mp?0!C6ref6L=y&Y8gA^eJoD^)!K2SVrh(+>sCWp~fMU+YqRG4O!Vnpq=DS4b)U)VN_qa&H9vmW%mvm%+4<-V}k05*!luN=Y5VU^i3IDFaNXK^#TfY8D_P3(!xlsFndSXWM!!7{d&-i^?);x=; zU;0iw2s;NLq@eK`bXK>W0GCB*B;kV}6Iw~- zlTrg&Z)i>I9PFR`m5mQ*jU^@xR4%%aaVj1Cd-Q^NZhAB&c0n;en4~0NPwe=GXY^;g zwJ?Iz27;(dT?ot`15;b>R)V4r$4oB_+TkT-`WORWif3Xh{FEqOy>5l%Y4~AHjQEB7 zJ2lXY8*8EdE_xrY^rURLZ7P}FwT^)TF36Ea`*xlk6?pp~ke;lhQB^}Nk zzD1i2s9Yw(_z0+-&l5P(evmg;viW3(XG9jJn_?wVadv+Sv`?ZcvZpGuwT`FZiU|cDSxn7k_xOR z{|OW0DVka^C;LLVa-M8)m9JO2K!vHm;kJ)^rK2l>yj+-EQOLki?Zly(PLr#n57iOB zH$}7hZcRM{1f;S#{?3)qT_e{-w9FBhm)B>|9kcXn~y{P zEV*TaQ%x69iQR3&<>Nm~nELeFyb>+NP)QMJ1O@3v(jk|jZgtNGE|jqMe4x7DFQ^EbWb3>02-FLPJ2nF| z;5*ESx|C)aSu2^GVGzh8l!$~wKFrCOU(<`0^1ft&@yo{)#F*XGhrkF#DiDT-rk zf>RGLZzKQC|7D$?c3ga1r3H*=TH~g-k`PG-`Fzc^kd;U2df^WE(+uzGFr~tK=8HFZ zNzg@OG?D%}k%XDjDxlP)>|ELV%)RW4T$z<0rKp$S~ z6?Oih{yac(y-p$+HWp|US_$NN4ZyjRXs8uO3SWOIR2MjhoGp$wV0j;MunHmIRv8?Z=VoR-m{qz zY(iS+cL}H8?)1h~a~KY`fG?J6AWmohHmOwhTcFX9uuMOdOT~fZoz;g(?ITEkS#QZ; z!iU$t&;Y);e~&xU>(PsBjL6kpBFO zMG)kXMa-y5q99=tpLUgp+Y+S&@`EP$M|T({H=+lNiKQvKDR3NmFaSSbQW*L7kc^oU}mQ)bH6`Mru3U&z(#!Mb-br@$BpA%z7Pp zCaV})4f+U0i3X@oHpTkG8>5qxXWgV3ID@Im3ZhWmFuGF&QcQ-EGg$}{Gv7GP$vb`g z(Foh(H*>E+dZ=#9*>BbKyJK;oE~+8N73b4hf`DZU-yKBCokcC&i^ga;B~TP$Bv(yA zH+3NU5iVqSG|Wq?KN?7GQL4lKyGSZdj>u&2f=`}YP-s2|r_7YYGHc2JFY4ZGjPaYQ zrdZQPmod6bHm7rUk58zg*15PBVvBh)6j*nU{8O`$eSf*}POP;~y}bvGg|)16Qlo~V zZL8?ljq)n!X3SWywGSKff5uqms(3;N0<$JkDz9rNU$sziuDzNFH?*-r3$ZL| z?ZrcL_yY#OJ2JEX%D$O!`NO>qqWfj09a)9Z9|KFpByTTQC1Y3iR4$3*d`!x@wsJZr zhiAe7k8~FYzs9@Wb z z0h|w0@6^{1qY-R50htDgUcYxPNpym2V0Ck7LTu%umvRB)CQ?ZHle|4I2gw)EMckgM z8A3_=t_HdJaq-GDc!Y)+8Vtnbdzs`EK~1Sc>JAS0>AHxf#Yu*a?>O3c=}H4L%b|Awd|63! zRXgaEPZA8U?8R*t>U|VF*+e;xHoYMNg6)9CN{$!H^z)ybSQb5>5q}DLLQ#BvN!swCB0YNJm=vOy- ztr3{ZC8;V~Tb(VJSoxXND)Rml9rEf#^c5XrP{PmZ_lQ2-u$keXbm+R?ip~;Xs9Ny$ zPq|06#)aUzp$~2wQ+eWltr#QueyEJUghJwD|)P zbTaClnPuijx#27zzOTsQ?QyC4I_Wvivg9|w<ojk^3)3B_k3y1ei+2Q?-(C$jhy_T>cOuvnFDE zF-xOY1f`6=I+qw4rn@<-h)fop+z%`TJR=e42G;$vr;JC$xKV=6jRE`1W*f&`Y3J9t zL>RmU;cx`+#8vE9Uq&E$0XNA6ls&=iIL3qhraWM86Ve!u06PXi%Z4 z;ooHmBT$2_FLguJ<||`w-WDsQay(wV0&>O9U~YRAKSDx=t(CDqHcGg~wW--sMWvtP zRq`g^p_wtz3mSw;G(`5ni(dWDDbW4(9g@E!RtW=3I=tCav|!5{ghj>PJt(HS0ZvtkZLY|W zs{kVR>90|5?qJ!eC#X@x0$!5B1Ys4Pg}70x=O;_arnY%ZwV!f0#H>gUkDnc77-J_R z;_&jdvuBaHH>;9|8_uCi4s{l^qR!d$*?60KE_r5NFjr#8lroA^YFtB|;@SGxv`4$8 zLvd=Hb-~iAaG3VnmY7A39ii1Uk{XD*WC)`AU-K-1_=8L`%qbC3<<4oNmp~e>a@CcS zhs)W#DAuvpp(;^Iyvvp8%xQ0 zN$z5P1M}*;`<+CnqNw=1ON$=7VP3lK^*ltv`ZsPDZ!n^;%i24KpXihZ{$P_tNnU%p z`t2uuB&iDT#`UA*#w+Itr4A6>($j^SSL4zy8^3XN$rgl2AXAAkQ2Cqviz!0T%rdZ> zcW!&$$FU1JpjRGgcO)9cyjQi%H*eyb9Mp=&dw24aBfv}Lx-t=;0A5V7!?GbamR3hJ z;fpk)CmMvP+Fd{mk(j@m)&m76KUmZywk1mXhdH5b_`q1>r-=@ij-Gmvaak(w@WEfvfd41V)7VN+yBJ^NDFY{ zg(bOGc~UK#T2BnEBMgV~B409a6O65RA8%)Vznw+HfU^Dhb`fhWhnr1?DvI)gKFSA+ zs(!!4=HH5W-t9)5h)^?{*L%uw?gsJ-Kp?B-PFFf}X;|urUS-*#wk*aIZRUg6(KeAP zPnPuJhiE!KL_m-x-~J$0yZ4>_QcJ6MCD6zOwO}^ZCc!k{`aZ?OjUXLKjRGthp(J5< z@CN?^x#Z2;h|kb!OSi5x;f;bS<9X{Be63 zyZP_zF>uVTKXnpi0EYDRqB`FMv+pzf@&jTkQ#qbkOAOPWM$~^%@LUF}9S|Qe+Mgjx zl(B!nVJ0C62VfvdP4=zkk~XFqS)U&)mEfWC;X>PaPzGio6>;z$GnE+md0xg&1*N(1 zXkEMsEPqZJW6a9kV075R;w?-*p%no)<{xxZMp3*c_=k5i3OFc7zB(pvb*^FSHtb~`*H7?7`pS>JUV=!CxDR>utBrnHN0) zCbeWKxr)m#w|y#oG|Lxy5%9-sTD3VjreG@W*O4}gV~I*k>gMt`IpGKkrLs_r zYOxRsB+Z@mgKJF(Hy+Z+5Rs!1O>|yjov5EdmBEUq7|%)FSvXWvO0-he*rFf*`Hk55 z_qo%^Kuk9?RG#peb8BVt!#T~~BLZVVSCgd7tY1_lWKEw{9a=q8ib|&$uPx@BTRQ}J zp4q%J?DpMA*F!%sOTQ?Uis)d3Od;X7LHk>hN9Cc8;UhvH+i?EV)s!;al0;wf0wTJj z^4L_gxTxvtVp8!`8Ucc`F*Sc`8nF5Vp(d3kjcQ{1#FwEGY=K!d#>)LfPkt4q*N(?K z7v%V-^|T$gvLsF+WAqS2oL-p!5YQ0Q?j!XmFL|s)%i=K`rla>5a*;AnKk^;M?9g?( zUViCzdhxE3{jE%Lfl~CSJZHqIzppDPnF3G@|05`m*+*ridv6aZVVOJbY>26(^{Z#n z{XtwJHX$yRCBs|4GL@twO&-Rw4oPzuHw%_TkIrQar7V208i1EIa{?}#HAE&sL#D|c zEj+5i4+!m8qjtoP6C`Tn;MsBUp#gS)T{f5hjL3fiRmxCn_G&1bbsmjs-ml1A`=j?G ztxTlIdsjB7tsp6#<#p_-V539fe|<+ zF{OwJ^ULn73W$h=7=}XRJgFVHftMb^K;nz*e?wj7Q)MjoPN^Y`Khg4rV3%l{Ic-P3 zr?#@x5nf;Hmuh2%X0iM+17VESiN_HCpXR7TG+utEERsTLY;*B)JUMo`KkJegU>rV_jMC$RYj0DYJE{hxN!*-gHvp+05WI|7bk|qdY&<~&{G=txDdf9zoSQ*+BjDiQv4doEbJA<59J)pR}Yb7C5Op`dw%3 z)nV-p+&TG7)Ef387eMAaBy7eOX^2P&H*;Wf`h&y_m;RetL&hz^ClMgf$9k@Rz{vSh??)vI~rF8|W61KK$s{$yS1@5ubZak?>cuNW4e8Ec2Lp>uHkc zZ@FLW#OE5=;IFL8j0aR`A2SD!S!=U?OSCi0U-35Y!vqnJM3|6^Vv|8RieDuSAFUMh zH#suQS{x;qdB%c-gz!cq{e5PrH+7Uk$~c1fyZfyitig$^qiO(Ep8~P_tdmt2-7+qPeFS+ZG_93$VwX zAZ-A&)8`bz9(H?EeE-%qC(i6sOF4oYqRLs1KhU`d8}!B@0#k`Of3-~Qb_^b8KY(D+ z(X#)dQZ1gXA;8XP`6;4X)kJ@q$G$&3c~Y&g2Z`H@}I&mzkz zah7}sX7xksB2ke1@6r|XNan&q@j$UO*74e`ji66VY$CtU@0oZ8Od>+UL^tjN1-Le% zs@FOU%1yk|c^bI+NO?IRPgC1};5d}S^jWzSFdLoB=+C)HdHS)r;_{Cat#aAmL2*fo zlp!=^wamD{rLvQ4B8wlYbugKcp*iEseIMxg>Em)i2EukYioqQnePdQHuaeNf0I@jZoyv)j~Nh&8Dd<9OCn`{F1$BCtLk3;2j5` zCYV^~7z5eZ#*UWjGq`-xBO)RK0@|{&C>RAmm_?RJ|!NBLUFYmhY`r>O9Anpv<3?b-fz}@MbOZzlG5P|m zsB&_+7XPZjz^W7IOV)CPUl>8sHmDqsUv5K|e68qwk3{GjFVR&fmOO!2Y#XQ}Tg0{~ zQl%KeM1ZE75Bsu83xwJ)?~f{EL;-Bipj#d}oO(jSje9~l4-aD$(F=66_hTvcYO8aZ z=*OoY_c}>nk@?v z`O5}%GMPWVrvpmjW&XPq016HGB7ewRp6fwB@sN(E{W6jUo95=`Ru0Et8RMO;t*5@Y zf`9uTLmc1&Q*5KZK4rE{s5C^SAAV1*&Z+~%lI_8mW?=_esJ{zzdHt#X)zv{*=Tlrvh%ZUw{;pYJ;dhyop-_~gU+QQj zKsH$hLt=+(T8!X{CY8u>r_}Nz!97k1QM7h#H$reMBbiqY{QsczIQ(uP6Wjt)&-m0J zF)a@MP=bizIn-Al@B2(g3@`JfBfYGrXfYc$i>F`Y1{^u*F|=Di;1sw}^3XV}!_P$K zCg}-Etp7Or)pYs;d7nt8FvrXP{iaK-s&T;G`)KLPz}H}|K~wpk{^pL~@#$)SNAtCybpCA)xJH?BO(|$LU@C^{ zuW9so_imsDE@r!(GDbbGvHg9fmnzFrP$E+x&~DG;b~jc|-DIQLLa;M{yd9+sT{k%BEj>LC zEpKuUql;I-kNY=oQWnRNV_$4SnNiqF(TS4}n2fLL)dJ>Bw+F;97^-4w912H3|BD3_ zK?6Tv{r9b{E^yGPQm-Cn(#2JhyqLJSH3H8u`LVahmB&swpXpc@(Af&u<(_P4GjI!e z#)bq7=rR}y7;=1y*}FoSUj{7bp33G#MErg8;dXa;{KL~#5yh_11kv$QjcwnM`hWXR zoN(RIW;kvl_IeW#|YH2fv*hyrH(Ax4!iJN4f+N6Mqh2T}=Y zI?lL^5Y-&&Bff-dxMIAQts#geK%>eUw_W|Nm8QXJeRR0+(DAS#{ofeKzrK#&QxkyP zG%^@eaj;12FfC>fkiWhB4hje84*=iZk@&R#`!_Dh1MNFzppPuDbVbR$3=x#23>Op8 zGNny$;P|8&QJDx95YR&b^ZkHO9nRI&1O)Wdrr55#slEtdF#LVCR9+aqBcI0oomCU9 z_V*0O(EWd_kp1Dbh#D*7bh9e-B9JA4W=`w-HUnir;rNb+O>S$Nx*+=_`$F$`6E`6k!1Su)Muj|1mCSGT|fma z@G}`39$ES3(~GO})GKK~NjGQnbbcy6AVNfoJoGY1E}e^~6B(QmOc-LQ4L&2XT>Qja zFrSgDlLG(@(G{V3<6BA>nwub&>WU@%*_|40PRrVh~nvT!s9jpb04v)?2U{`4{UfPak(I?6V|KO_Zf1D%C!|@bpsti znjw2(((OrtEs6($Kp8BmoIvDTy)ms=%GXC_193^&c>X+QJygDQC=p&uA9(?#9W&%t zfVOcLkz|-D`0ssy(atCI2)@74rW{r>_Pl}`{+_-XO{npW$XcCn$5B0$8o~h?>BL!Y zT5y&mk~)b+qH-(usxLbXhLTWQPZp?JG%Ez)SU1)4?9Dk9 z64MO#O@nxs+qOO)-LJj=jKrwy9X6C#gU^v+!lqr|ilNTc-}sz(VbehaMO zTGA-Oupkh1(t0U?gks1G1!iOTJj{ZzTpNrSl#mc-{e;}z?|{Cq=~wrcYkzNF?tUw$ z)3C?#12-3gRv7_x$4_m|Z&Crz01so1x>&~6#yCP+N2Ue3Zng@P_Rj$KIL+ErFzUZ| zCT1w4fV}$@=`&Cmx6^*}d>@%sU0rJ=?ReoJN^tnAh9N}f_)H8qrORwLu}E$6ea%R; zN3Nx7j9Jl50uiAQK?5{8QFDfA6-j1eT#O6m7wH9bwgyU14~39VdPYXiYRV}WxnKxT zSom*SVEM^W7K2ihLG2SW9Hf5b)a)sgfK<&rtY?)603-=U%;CVY0eP*0P<@MzuGDKu z)UPDeX6JhME&`P&fby^D$p@IU;%DBk9W96Sd-_gOiERz97R3`b5O4^vRUwcSc>9yK zowdjT6>QbuL#l+f9M)e2p3#S!$ItA1#1?xpeq&z3f0uk6TxHn@V_Qz1R@;y#5PDB? z!V^BO8Wvl>S+!siXwmBgUBLiC>0erPnQP6}>&nh3wyDo{7BOUVc9lsTy+WU_Tk-a-{FEaTj{}b|=A^ zgvaZ7YhG|XJ~&?1BtO>tiF)@WJonW+Mea2!cft;OI`pf6(Z_OIH%V1~zr=m7o2OTJ zZ;dN7LpIX2#+OH&n3&am;5L=e3R(Uk88eZ=16w@ul}%dA;n~?*WW5#FvLQ=qf}4>k zoT7Og!Js#A@!_Tt^%tb(F>O*YAwWo2VlAax6vuFg@>Z}({$ z3=fVL2qVB`Yw^AN>^^lIzWM7nmqYu5FVN`T7Y62QVF3ix-6ICjx?Lf``_1*tiX_Kp zz4w>_!NE)zLaxt!eNcEkcb>0V20rDZ!oZlpFG&a6>efJ7tvS<_)58V4E_=h*o<~O6 zi*KrawK$*8c>FvTme4JnHN!mCH~8~il!oEy@=98-#i`Y8E+AT{!Q{cF0DSjMG9FFL z0ilkaXglxf=_mYfXDDmV=r`9nD+2JNv`^AVDWo|4YQ`X8Kgtz_`TFx~a89|riI?dZ z7A|9YN8~sTOtzG5`;C^3e!Em&yT@*}fa!{$)oJJQ4%-jygC>c)d)axXy_px= zRcMOrEU(evV@-& z{O4DnlXcB`(vGrN1$vk9enx%LnsZNb_#$u1K33P_ow}GgQ7Ku!aWfD0!c-iz8GM$|% zdzM7#Jb0W`Fe1Nx|HcDU(?Jor_xcgm2cn3;0Ho`dgQz`kPgmijlp5!UX z|Lu%DQ?p@aWoDQlZX`y*i;vsIf`x%;MDTC3w>=#|783JtD0Eo0pf;4t&kx_-qbPWO zer-C34(9<62U9#e=6l2H(d2G_y#2Y|TcS6YfdX6tp#Vy% zTPVEmI)!TTmZSy$4lUmWNumWk?)>#uAoARAng_a`u6bSr<;H>2r3X7!<(UM+5MjO&PiCvCebUh7F z?kJpA&f>GI`nc+QAxVs>Hq9v~Cm-2u)?%ZGtYn0GX@HlkhU|WVN)4NoRALiBP7$zo{T8R0xxo*_~`_5zEOA+72NS z%vQCW`>qG*IT(w6CP5*As8(K#a5MT%I&ftL-q6^iIIsj4CMfu5LsdmZ!gd;MBV65= z&yN2IXO$_yP+34xF?-}E>yq|*@vfMSS zsMC|O{@x`Sj5j(H12>>Jeu^DTeN$`%8w@QK_dp|Rrh3^vr<18LlSRCuKfBi2Q8n7- zV_H0EPo-i_y81^{g5%@f>u1Om1`5N$H^01h#(Xs}DELE_gvpXluUzFbgt?UChQL z*}2{Y2X54{oQ14^F1Qqy7IZ7IS@G=zq|ZFt?C2wq*s0NUS&AYrH{-#*({xAP&;g;I zwAsz+VecT4)w0?V!IL_>UosqDIAb@LSw|baVNnvKEiK@7Az#UUi_L)$ zX--^pyu0hd&|iz<69T9U^t)T#@^DRMwM5y3!YmF~I31neBl$1eo;c*v&U+Uw=;E7s zo3}^OJ6U7C$4mJlwB;E%eTF-kWtXl-eI^#HSP={~fPC zMbxaL11vj{Jm1;RZp%r_sU{Iit$RdtCFvv{-At(LBw+UuD&zQ?%Cgs$f#7Wk|lY+^`${) z(VY3uH9Sk))v6f=2_yW#N}I_st!?d&KkDT(Dr^<*g)l8}h7 zus9A5FhQ*4mMtVx__Jmvj`Nk5UtsdiHRgNH!KdjuvooFTj&G2&EWkp71V5YiWJ{Kw zt}4~DOJGcN<}v)?PK+fAxrR3L-`nP)Co+E0sRV3zV)^Kv&oUnW&TX{ES9 zto5yRZlI%`pI-nq!yY|6<{4>w0Fncgp!&0!FLF!v+pYbWv^tp$!c z6Mc$Y?>JdK>jC+x3?AYds!uq8;52u;g6fEynp(7NWMga-AZBLfP_2rj41GjJP zQamFxjVxwt!eJ3}^TC~2Mo37+-rqw(Uvqm68hMHFD>Wc*fyYkB1XAJLK)q8#!0LyPA%K2Va~Zh;g4ytsC5&7M#IF!G%KIN?}_ni&XlFdHq7~K2}$7Tl|(KA)sU3B-PzoZlK`vf16qy z9PI5)`PY7`9}JE)`0}!ZSVGACPm zY|ht*u@rLWAI1ND)N*H6BFMF>nW|HQUMMLXDe3~U*_svamxK#+{lcKRkU=}HZS%zx zdgO2_mvW(^_jA8E?`S@d%H(q1?B5b+;$O=NmtJrDz|p{Wni!CfFFw*wiX%kP^f-Rq12;Y)DvTm(z;*;z{mH(>E>1 z`Q`BESX5L-!eh+;>`I`8&!D+tX>2k>ZM3xMX3{l&(?$WK=o7|oLEMku+EcUoD_z^PnNSY=K7)0mB@eZ*M+HF4e=_QnwV#ecIZi^DoX{R zV_XXcTfvnQsza`|JU#3oTd48#KLdHG|4pkFZja*hL#@Pwv{fYvyhR-Xu4A1HF&uGMk|F5!I{VP(sdkRZVRgO-PPRh9^Q^c0U) zrv&geKR^FT>&46-iJ98n*NHW_M*z zWGg$(!BS0e%I;A(?#upI`Nhxj*&@x#nAeYj1P_QlzSGjx!f0e$=okf(A6>Db+J~nRS!%$M} zI1`@PNR!ppfSz0i_u9^m{7yoAJZB;;4ULKBOmA;(CSfZ$Kc1SEdX#XTS&dlH^3iU; zeRy8LE68Fj0g}*=%t4Jhe}xxn3Ab~iIi&UWE5s^yQ2d1dXEkDyRq$>}fQbpSi8eUJ zE_M1B!xQi4dT#Y+r?7&=gv!${YDO5`=yFlAtX@OEIHH4#T!NfG!B2dOeevmU4=gWw z5_&Y!#kHsb9aT)MCvP|fNQknW+e$Az*f(a*3gF4fz-e_Oa8F6e7)7sNJ^v##v_F^I z;a0z6zHTX%){Te%xvI^PL_93lcT9h3k_sWW;urI`iN(v>c5|mK3)hr+OG`oBc!{SX zQ@k}Uh%aI3o-zILNDB@iQRZBVQBBUm5Fht!SV#6sIg~OE-=OF|l_F6~HnygRPv+Xj z)C^$DZEG73TVa+`{YsOReiAcY4-*e@qg&CrV63 zMYV06TAL_uV8S=HxY%pVy8o5lQ~f-3+SHzk`-k~ma%J+KH`Ee#=Y+ysCLOFGX?&q; z3N$Vg@HSLN92r=< zZ5jvc(3fT0l+FjBsoctwH6U%#U?(`O*;9n_Q@F44Fa{IfV`6?1 z6${5^SzFDz$Q~_SDf>2Z=x#yCn}RZ%ZFZ8*=jMIUI{KHf-_8!_Z>4_wNlV^^~xlZVDY1gO7t1yrbwoGo7e+hcDxVyR)_p;u6R~23(;f z#77FQ9Upq(INml?{N3(1q|sLXI+`VQIXR#08rMilMn(q{C3A7A7w)oW4sez>>i z;pQ^^!!MY1Zuz=}V+Q!3z&%E%I<-4nC_w62r2A2Eg2wS=HBu zrxDiX#2=^(`R)$}m zcul(f18PHxB>|NhI0ty_*-)MvIoP?_;z3o+*FP~`I~TVrQgOC5okR`f4&&YlR~|A;P=Vkt_tkD# zGGxcekT>oe*XwzlpEU@m*G7iDMMiYDUK6etmIG;#_JYtwvB$EG2VJatm;=38MeO|k zF3?g$p>I@Pd#$dw&23p$`NK`9^1vXC+w~!nH+UH;Gtgn`5i3MN+Et-TGptC(m-8=u zUjQ9t4yeHwe0mxfDNt^_R|=?vHQ6RxCK+DOTLJbP8j{&Jm#RY=Ry>Io zk1~N&uVx_dEJY3ambu^D1_Z>!O#DG-8Ofa-$V$wocw=5bt#mW}wYgBVL|TA!ZpfSo z;r0F3@BK_q$45m~E7p4+N#n-Ez~JFYd1LPDQdzQlDwV{ZnU@qcLBZ;N(YQ|+)FkoULS%ADIFZGUm#bpFZx&mXOw_ z{=$74id`IBViWP46?GFiFVicw<1IQKGv{YeM9 zzmzosgE$U?Nobe@gq)fGj;IFEaz1g@*I#}IRpF1!WL{emb)Ie`^xD|t|1}Qmv;p(T zeP`Heg}>t1_cc-0X%g zV6T)ajOfp?;$$RcsioUjZo7}wOYO5kyRV*;#X|_^ zBb*!hE2+ObrUhNd`MqopTR3z*w@#z8fu!d+#d+?vU*5cV222KE4$Zw-_uz|!7uRYS zih`Jxg%Hw=>HBjg{exD-l?`1d=pV{R{_bA-sJ?2U5?b-mmJ@JgeooHR%4n(CD85uK zGd=zD1w$+E*Za5iq-biwI7uCyq0co~xC(7?g}eZxpt`(b6z{1^VQjd@n!2tt-*!4R zUASwZy{~(zvfZJB8GWao@?)ZCiz}ROGg#%A=#)LZ63xywqB-Q9JGwb#h)R1I+K46T zDmVv$Z8=`21Kc&Q8-#KeMEHzm%VJJGS#joZama=tkaZ~+`EE}fSawJxr1YAO8o z=PH1E!lTQSDNIIF`~W8Dt9erhk$^c_T!(YX`QCG_hH3O8y?E7H?)zUk`r*>{B2?!6@ zdm^?@hhxzjoU#4D{!EU4B6cN?OV!bKvSy=Emr_YDk|9%cMNPo4&YN2K!2Las5ZFdG z-^V`LAS&6pjZEX|gai(~V?i7#F>gDnUoduh@r68thj%nxz8g5OJ)EnzI{RhjcD*}Z zE+TXI{A3!M^h&#Up`tiBZ~07=r_`iZR0mC5`z&;+;}8m$15s1OE;*W+>*sQ-=v^7$aRR6Tx?mgi@DPqHl_ zF7bYc_}wdC6CbZNSk}QDt3=A*f@*|;5_bC;pB^n_D?vJO1k`O@EcXXQ z5b4Mi1!ro8-!dNX%crnx5o`f$+il!S&KL`0Vik{=rwWlnlJ+etII1L>Z==hOt+$&S zU%g8Jrx7;(hwHhHK!?Cope3UW86a!M#NfNY{of*zRM}R8!6nUF63h}%hotT8)5G8Y z&x0EnJFo@`E(;f7avkiX3sjoEYH&xjcJM@;fUXO08xAUd%jp?K0_{_E-5me3sHW?D zeug)f!4jt$@$1Y1-`$M43@ay+727JZ`z;d&qD1ve`#+BD*VldWdU9}b?YTorho%*< z^3cH2id-dC;vp!9fOWgKw}*;qFiqdLw^ZP$sAw;_<@47;3mOYJ>_6TTjTfWt`*{zk z8|CLtqG(X9#~(~{w8jf4m*Ymg@lw<40E8705hIn06K>kKT8OV@_mbXR{#ST{l9%q7 z5}=rm%D5Fq+$Hu00t}>%BX#CipdE0CQDNFc{%;TNN?ZCDJYywn3uCQTxckOml$}@K ze|rieZQVhbk*$tksU6_zD#poya3l)Oc8bIbHRaoy3p_jCn$$zsl7Ntilr1$whKd-- zSHW=$pp<$2%DM&)dnjHp5>!C2C{*~3^k@`15e=XTwV!6;Z-%no4d`?S8YE=~{yn@| z^*?qTrlz`|cxQgLv-|#zi93y+iAplRQwaHWIt3kn?M37@A>h_O{k*?MnH*9`$&Z_z zhhspg51`bnXw*1w!!{|1x7XnaW`uiLQzFdKmQ_)4FHlMN`SWLNinMpv9_^qmi`OF{ zxc{@BGUh$#A9FkOD&9+1#^a-9RVH5~AV|Vor}py``(!f)Q^K+R7C;HAvbwp(Ero@; zom?c#%{}qH9nb0%-#r4r(cu(#^@lzQDB-{Bz1_|EOu@NpvuU0rN+@HZ{>zX>5yGzg z9&#;|fp%u^O5pL*QPF;UQ!+N)XQP_Wa!^oP{ygP zjGKg0CFz2@J3E-z_|K2h3*BOr3R=c zfJZ)JBLbg?9?X^`o4>I>I5+^lTNO#)_-90cs z+m$-gf{QOoQVZKQjr`9`;9w3Q$N4-3J}m>y%f?Cgzqd2jP~fzEkp-T}xyU{jE6S#^ zuC+9tu%-;@dDfD$(HX5H)#m4Bko+Si5z4m=W3FOj<<)<@G1|(}T(?_3?<|w;8wLIg zxWB@OzE*JBNk}O*4=(QF0pAC?N8F&Wj5-U>AN; z&c3@SA;^2LyEE+v=jW%37yfis`&3T*T~)}L=j+b0cE}fOZFfjv&wN&xkUnjynt&kKq9h0ssld7wht9G6GNIv@HH80|#?A~FTU$FxZ?~@4 z!C0`qk6u;NFpU%^@W{RW6r;PMDZblFYyb?_XsdRQp|5Y!NQ` z66~=xKQ{L5a;I25osN!X#rYI2?fHSgdxiC?0XJW_aM~OQ&v>~V7C&h{BmURa+!sSw zPfsAW-9zld+%s9^BXe%qv$kQ zY%H4h?HY{|L9rgx(`UsvAsxWn`4>4;MID;%K}pPe*z!Fip+AQ@UMkFQ+Qr=!I}9c? zH@>-QyS+?n<@CodCx|gYsc-=^HmzaK+^yp&euJP29p741pP(qJ>LO9o8B8FWM#{!# z*t$%!jKiKmP5a{G@7g$otPH)ju*KuB<%g5w<;BM*2-{Ypkt>d+3d1`z?m?04aaEPa z`fu@$2L8I@ndE8>c$zepsauEs@BeNx!6_vOgI>9`K`Ci=TP<;0Gt<12g?gKzIrw$j za4l^uH8kgKQeUl9Uc8K0ZNa}yKsdj0p*kKPo4nae9Km!{^h(sKN(Dovm6lWcKYk!U zR6kffnf?=KV}f*KD0!q5YR!w)+L4!GDk<4`#ah{HyI+lZWOiqo6Xn#SQM}37zj6Fq zHlM*sv|gga=|bC4uxINX5GYWhRlGNoJv&YUTWhU#LV(3;y2Qrc;&m`vY`;~ICx($0%UCjpuB$*f4I)JpEwJ>OXI7y3Sl6Y_d@lu+tY$GDd4&iI5ry zHrD7V6yb3!0ROZ-Sq}s!O2xC{p_xlUb@aji#a(~FI$mmj@9ddW8&h_@Rt_k7LTMI4 zj_c#&O!QXYGl^n{OEOrBcAU82sS*V1#CM#&uz+3+FRcYbK#;-Z zJS^Z$PeR8=#>pxA7(KYvHU4I=?rc`SdBa>d2RJz)T}c-#B{pr;*Y;_$nWu&Y zXc! zCdRO`{<+0i^goKAf1S(`x+UdkEAuSMgdk<&TW@!I+rm=iH$m?|O3P|Abb56YK^RE4 z{cmApk7PhYn%=F=p`dO|{vm5-n?(rWctd4tTi9Yb5&h3!-4Rd_aHRfMXuz9Ulm^_w zqaRmB6NGfR{pQV_5rf0S-}sPfpmd{!5EvreFk>l=+x=hNcINaEthdH-{aw@XFa^Ao zo6{rY@FMbCSyF(?VC zW{S6-?LY(puCL&nMwgr+6edghhTD4~L1G|e;DZb-vU{PY`iCvJ#nC~p8syJ|fI5on zBRCg4FmX;=ao>0pO=*0eiqYpw_<4p|@cz>!=6qQ+f!Uz2t{*rkmLTF7Jh9~(TXm z1|$w2Y37hI-!-)mf7yGdMto*XK&sl$=hx)P9kg440OW^Tsr@e&0KW6uo2T{7sI@1N4K~FnrT*fT3-BXe+LPWA`H)v4!J2=08EAQET$ckPY!~M}LVKyA1xzL zUZM1M5)OHy=C!1!=Bq?8c6Fd*OZ#oslzVk#D{iZ!#fuN*`jLDf2H-xy<$ry{mLKQG zMVWXyxkWHtynUh@VLn@bVd3`GTWYVWSGYF`iXnV`+f(E9K&MN0an1ZLsxhgTUkT5O zG}DCD`9XW$>alp#nQ`GEh3M4`QNa8+ly=<%_uSZmt^o}w$>q=wT4az-b-udn>%Y;^ zfN9erL;*Uiaxyd;FKw4*@Vt{r_9 zm?bdTPD{22y z(8dIS2ig8gQ?H2DfBv7w2aV8?(*H*SSsgM9aM3eWejmtvJMcPn{|HNnNVtXS9rI&I zRfQS2b4xS}0kV+chhCJ`Vt#T3SjY0aARrgV5av=m5sV$P4!_H|{;VijgX1F={{5r% zISQu}8CbFA|EzC2Q`aMb>p8MA#%H!v*g?sc#ojPKj4Gvw4=DL~<8ccbVqww1(pCeE z>f;8t6H?pNDEb%)5=5`~&f}h>FsA|336XbHV}Msa8^`}h2MxmXfzdEFxYj&0l7)@zeWYiJX+ybRQU6wSAF6_Ic)_0n`9gkH2K~+Z|eNjSId(X z>Xm{XFZ21Eaa>+MPiryhYS+|mQ|doh57Q96?Z16S-U5Vzv`7Fp`W=ok#vq2iZN~D` zDt-9kerf%SV=vS1SLOy)$pH+ZfG%#!1dL)6M>TDbDXX&qwF>(+hyut51NSBq#t>4W z^tb+qVX+Yt*wV;EDMXl@5Y?{{EtRAkZiGo|%nt|n5Yi3F>yXn0Q zkF#aWmxmBmx?aWSJ50JsnDLaEa~-!uL+Zq`TV0zCEH`pds^;U_{DD!7(+{|24UCkC z&4ZIbF3*9!+>KW4I?uJ)o~4NaD~FW!J5H%FbUP&aFPDY-(3NevKL_Uo8q^53M>9H6 z7J3Tb`1JTdVuEhIE>__lXK)-m$+z5_9ncdnl=c!d-0ax$V*?c*&(9GYS|lh*OGBzP ziy^ijOHYGQa-h6M@&!e z#Pb1};q%>h$Ggoqr->@+6vP;x_DTT5?jvH&Frwi=+yMN<5Yo;%ZVl`;*_wwsO1xdvJcI1B~MCarjtZ7kdY2}#J z9{f7%fJ)ucHPD9@W2lduD82qQM#MzPT|g*@S*4bGdi_h3PG4=Aj4Trt2;_^><{6H| zba{fL-SXlJ-JBL5f3dm|>TIjYa=zd0UA*Gq1>fUE8Q1b0#Z+1~d0EJ2@Eo+Hl$(+8 zK~QNHFHEmNL}z;dpSt^4fE0Xseg2Wha!x?|n@ZZj&W0oP@gV>)NwC=CONoJR-_Zc> zAeP`Z^j^}hn?8NtmM7E&i0Fehifj6w%inq3U7tJNmBc;0y~`ERO|IVp%Xh~xTX=Nb z%|sfPCGu@LO*%LPjUipi1d_`29@9s|cP)>qbC$lO;+QARjQmM}W2H#qB=78VfDj1< zU|9YIe-1mf&^BH-F;#05JOVu{x7k!Z<(q&MmU`6 zUSTITm@nz!fBvS6DE=~#l7`fdpi4v6)P`l_r!){iWuyK8H@T=uz@NPyh?o^zhi)cq zOGe|dJh@7Sfk0Q3^fH(U9aiozJLYZdHH6pM|M%q zdv8yXBxvkAP;RA^-%~nQh4-;*UB314Iq#sJmd64RlxglWNi&R*f zInUcf%BmMh|9DPiXcA@jZyzP0oERr(7$Zb>euq3^*Svup^9zqm9Xom4z)kdYS7v@Z zCc-#U<)tXgRjXO*(SX-BFp??``=tN0C48umRvX|v`c_bn_313yZV~1no;J&(?05_5 zV{A^M4187dM7zFsuU4mkB@<9}Uf;Dmn@VzYpXgjdpu4S?l0e1(E(mHzSwF|~t@uF0f*mU8}het)8M_87e%I$E~sYki4t15w; zr9wfHuxwlGH|N_mGIR*7EYl&GbPO>wEzT%M(hEBiP4ApHVDE zm}bv#k~~6q3zPwS^O*K(JDnh2ohA!sdFj(x@j%Gl#pHSzHxZz%V^* zhMw@h-1k0vy1m0=wp$2~wQY7CIJ}Dt?mx0t`tfuAcg)1Z=H}&>G#&ut?c5A;svaB| zneFX~#KXo~6}Nr6yGT>@d8{&0fKu+}e1iJ@c?8o@cJ_k0%5K+A4rfHR&{REy8G3A5 zt$c4<_*>*c_rMp8MuVMnPy4fw%j=pncg~a}aO1Igg&<|j?jx?HB%UD%ZuC7++%ASlg*LJz*~;PF_=6_wIBMec}rYj?3@9e zXgKFQt_27X%DAkU>P*=IurlnBI2P5o)On2$wUXvBEt2NKJa#~Gh`aknGVNzw`$42w zy^seVxmt~4Rm{r&0?t2E)ifn^{sPWJ6-6?p83(Dy8nl7iVg+$qE+sXsBNvzv#n3I! z8;7MW+Dcc)t?tMp$y~DJv0BCK6Dls15m)cF!r1ga^$f%b*GF?U+z6Vz*uX#00x6%= znhUa!H@W2U6xno3nGMmfj;xTBPPt2-N%c21F$P!O8*8RfILG3fb}>vecFIVpQC8~? zbTiMBe^Z35AO&-T4O#+>8^H?{=TOalIWC%IaIfwNN}qN}O7>DPjI zj@maE$D;ryp!}!B-i`$1c+5sEAPGlc^#Ej-LM{Xjd3a#Nrh=97U3bQBm?BE$fa^FHv8)|qaJ9<9R zk(hBWb{sfpXDSmG1BAsXLaUPcH{qWO`9%KsHjU4hQMPw7m5GOvc*T)CaiQ<*FW5uy z46I!1Qro8t1NKdjoCzq|x6($wm#3vVZVNuOdI6$JVt4Wir?g}?7(Rd%BDP;8YqIsc zBNn)-mnRfzTEpeMw9u+oTD<5Sia3^YvrRJ^6jYNfVB36dZJ~MB;xE4YdNKJSuV8kW z^Gd5&Y?7b(vDcR0?Ij#@&*@&!R{jXMWCxtlH)m(MkCz>?-gNn+sWAyH&M)oGy4p+i zMyzSitB+ilHRJlENM`KXQc_ZSI+s|(zL#HG`T6!LiyzErEQDFN;=654HGbAEF&Eg(gKsHaDYOtLh;HJw+6Hr4JvC)!u=0WW&vP~vLYv6`jw>H z(n^&A$yrj$hw>jp)%HH&49AA*@p+Q3J(TkEerR-EiD*4ebAOl|Z7nxj5wbu2A76;; z#exYSqd)N87jf!89XG@3dR#5LR+`~2HG4hnlYQzY3)OXt<#eM=YdzgdUvB*Jxt00b z;({(;1GsY49%pU0=sSTo5f6p)-c82j9*0ai=Sx=pIRZjInd@~R$Hh4!;C6GfUNP08 z%dt2G+;?QM5_XrYKO4>s)bf6*F8!$_MF>|ZkkmZOB!r{8;wGunas3{vp$Yj%rbs8* zS+q%^raYpddncZ;Bmp+W$PSuBd~>HU!}?+O#>SgrWU0(bz$&vgdIUeUxpFOGwkHO4 zSMm+L>PneBYiFO6isNkqgq`DmVbL&F=6{}%kUoi zWx3vY0bnme=vg_67c5ZN-f%QjJ%zh9|BxXM5BWjMFl(~2%|$UGtS9O+=ZP#JyYokd znh8NoB=iSNcDrH0^GSE7$R{ge?&g*Srx4L!YVK3FDoDACg`J%XWaRlFe97$91J|lq z=oym%S8Fm`er?jK;QOMDBwXFX22d5*NXK)|N;;kVD7s5x#INRGe(Wn3N%c9x)^n0L zzH|I{wO-sTy+3sd-}Pe=N%YI}W`6x%GrVr~=njCx!b2SgUwHyN+uZ6V2vl6y5KlXl(W|bJP z!cL45O-6L|LOCr8FRgT_W+qGqGhyVNo8lReq7{Tyve&Jkn%|`!)gm{Q|2iR=#XXlg z@C+W%!)>+B&sCw&-4VwLeM&-ivQ|erF(g!xlKAl)Q<`Vx85g!nx8XRP=sqMe_-DEL zV?PnyMNwpg(ItCOo@b#|xM25oDfEW~p;#EuO%QoDXxMK}27@`=n-8%Z4Qxm#%S>*- z2Hea$5panUr0YTs$b%dynlwwx5x9iRb;iI4(;kTR&L4{n1we}{uP0c~R$Uak8$=SG zqdI^bwfE-G!OGF@_dyUNqtp24n|mY@cG8` zSE_1RNlU+j7^m(RZ@yrE(3f|aXLc{LH>wE9B9ru?FZog^8knO560VqM{JA zH`Lr2{GXbiXl2(*?XKcsC<98FH|b8@BtYhd=FA9V);}Vi*Jf`_N7OFtzJ;lbXkYd$ zajL)L>-rED@trHE4{ec!C>5k*5!QV1&b^q%aWUlZb3QC>GX7Gle{M&crZq&b^n2!d zG1d3pUv@}Vu+F8@67gchwj~OT+o>W?-1a8JucpfU-anG$tLC68g$s^2=Zor{Pb&5l zdekfAMw#mSz4wc!D1h01aQxwE!c2VKFr}iWa5-uh(9cmRmB4Y9_!U1j0al)?$HUea zNesL9=9#O~*tN#PA+vf+X_V$F+Wz3e&D5I>-xSCA$Z~;`p2_@`fs=tx}pF`^v zYc|AK@BDjM!QqUVIb3hyBofuWt#HBFJK`CQI^*K8QhlVYZ5I?SXw6B@2w}GMuBjqS z4uHRRF@R;Tc6Ia8v@Frzz!oD;PqO&suion{6%^um8Fs$$CC`z3Mr~5nm=?7|^-((H zhY`jE4LB{VlF=nT@v8Yto;3;6++6I|wT6-biIUi9Of)8U(oN!muAB;2@;LjRMWGW3 z?ynr*D4lm)R<=g-hSHYnA?>!KR2^|kAbI*(rGo-rK`!qG;d900i8C#&r7s<>n`v3PDuFHzkC?khj<7vf+1Tnz8uzKGi#8nK8k!Kfoac#51BL zO>hvq?s-^!YVnjhYUJB#;T7&4ep~G{JLZq)?hMg8Fn7Wd?>2(Z0}WG-vWmdt%{Y)M z;6NumOd~(+__1>#Qf1E@AKc#KpzcoGOnmsHi?$XgSF_6cpy(`8p1mT0ZpIKX{_%I* z9PaxXW%$SpMSS{1_?_9S9}XVVIvci{SSFHF<*ndgv>6WY|Cyea2=DhoL@B<0&O{)8I&LPE9LT@o+)j|qq!T|F{wvON0C_g(2d@y&D}(*rMpNRZb&Vo zTToRn!o&0mu)eXePW~6AiZ3!-hqM##u-Wl4{}Nl+RAl;D`B?VQpv^veY=CxH$voHN zw^Xp&1*`KX7r1UUQ|FM&tjT@r*sk(bjAtp7$PHNc#q=I7Nq(l$k4{+|d|tGn>f zL0^W@A&W0r#r=e&7TgvU3;doBsts+|zqquxGD{p_eYe{!iw5#7Zxvk8($USA8x3ap zKj^r=?;#41o`TvyQFG9J*x~$PoQvPW?wQ4z4HjWPd)$RY?SX$&(~+DE15H&|I)QP3 zuo=!_{6Qsq>UNTZ>%o=mzX?Q^HwIW^EcNhx9r#<%9hlcV&@X8ozpg%px(BIr9cZ)% zo+m|y>7~=X(5Zu2zv7>j>JXUMGfc#oG{nhJngmE7dtLI|80lfU^#i0_W$+;qG2!dv z{1|2APS)Wr9Y!CtqYHCXkabhhX`gXXt!%H0@??|sKbT4iGb48^T>&uq+Jf9A7< zBQl*hb!h|0hh4r%1tDHUU{kXzsMxQv71JR%M9&D3Sf(FF%#O||PGjB*vN&C-rT^W| zp8Kg;3-4f;ixk7wAfsr%0ABuk+uBDIrTI6^B**eGZx3gLEq?i8P(Qd}?8GGWlB9IN7 zW7fPTq!>dK^W*Yb0+uv9MhTl1OUVQK1Qp(h|sD%9A8MNDXoC8kY>!)vj1-c+Hv zdq(9JwInouGG%61qD4vFhpJ-+2n!3@b6FKEx}c}S0IuP+HIF~JBGaAF6`Gm#p|a~x zRVlAlXn@RVFRdhm15k4HzW{llFeNCE7`~N$!pXq$Gyj$1B8ZoT=3zqbNi(r;`8#Gs zh0vWrb=ufJxQ?zJ+z^kc z4|`rBKAU8F!Bz6EMC-ZveK*D4z`y}%Gz>1Prz6jAuIA!rj-+%qeQaVg7Y{;^x~O!d zs+ZUErL*RIpyI;7fRtv!)!7*xWnbydnpAj(XdWqjtON#gAPhz75ZRg+1Yg!Ct5 z*@5#ItOq#sE8D^kLQ#xNtAAqDSQJ=pq)nzPVj`^eU`5 z$avV&$OlK$LZxk3Y7LAt4OGEAWf(lkZnMn?A=^WLGg?MsuoWn4+WG%l;sXm@KmL?BENXp830;oqp zf!~oWLN!8CkENjAxX0FHv)<5J6ll2-vSJ-A(@@0?{P^KukI;PdQ`b9k2;&nrl9c~@ zGZu6Zv7H*aMfH3Gy8M z3dwM_7P&NqT9b`slVCzTt+C>J1SSkA=T(&|9?Ggxf^av70eBif%RG@_B2p<6KT^l0 zj!TXtkrv6UWnCpMRC=P9n}OEzuzim6(VSqyMStITwEiHYf|)ss3c}m2CgwFG0?Fp) z!gc^*6C*vY?sxCiW+0oJJ!QnQb=B7fI<*}>z}9`NlE5l8L!`aJcR$l4zfKLofC9?d zegtKr2B+aqLON9k?0z0$u0l!7G~-Iit-qc!TzpW*K)Y9NrrW&(8y!->ekId{N`~^B zrF-k%;5RMBm7<{TZ6TKas7#nUIH~+%-iv5=r*+R305Q6=v-9m}rjsl8KNs@aj^eeG zS}#n}CmI^WPUE8i4dG)iyEz;ZD7*{oyMw&5NgYzS!h%nmVPmL#*tyQOIPNhGPNRX{ z<9vBhz26X>GgrDC_U8FCT{2RUp=4OP^_*U)Z>LJphB1NQhW)wK;K=7$eO!hzTEgvS zZ?;nW1h|4r4my_QRnUy5$)`|;++xd!MOJmNI?r*J%6aS71L-41;?;Oa6`xkLu<^;^ zzgR$&c`p4G7<-$D*QuwIpVbiY*3{wM?Z(F>UqMHdl{*>qMowLfk|NjEyaYI@gdu@F zMJ{6okj@*ldiG#eSa=$H`~Ha-!-$=p$i{Dnbs?zHirVb9!J#~#KuhU!ITmIiO(_W{ zIoldRS?HX;ie(VXCP9?0S^|&TsY=$vKq@F4%*tjB_TscpJ#aJJrKE4w>=@Ei^RUQ7 z+v=%rksq!J;pRW|bD_Jt|DNtZLA?*DanUQtbU-M&Y$Ap$B&6|hhg1O(|-rHM2_dX?S@NGF7#NN*wnO7Fdh z^p>cghzOx~2+|UIODG8>XYqdD*?WvL_Bdl->~nM06{PX-WR+R|^EXis%HfY5Jq}_& z75r>3sBfK|@NFj9C}-p;UCxxUjNyBmm&j)+S;LcBEfQe`KzSMataVNE_xnvKvkd2x zSnFs$qg$pw<;MuMhp|w;3hLv(RkDKcL%MYR!N1hAT3_M8sD+l}LkPla9lnrxUNnBhLK9${dLGdz5ewx856ARgWKVD@n_9|we zpY!SO3x^Coe;w)AIwQ>LXKk*7G$tZgS4_nx{m4s`XRPk=WGKiTr>NZ26Da<=aTP>a zb|q%Ce*2E>WDJ{rh#K@P_1O!FF z3xQDK-bwELWQd65ZX65=sZ#TDPcYD%Z?B>9HQlRE62vnkmP#xTLv*KvLtMR0`56tMHmQXbHdqCy&uTd}7Oeg>b*X~~?l>)q(fNi-FZ_g+y1b2j7Gm-pkI8~1*p$cLCs zz?xQ$V@b=(GBA9nl^vVQQy6Y0yMhu$R#_m$#Cf29K2qK2w6)2<;&N#t7UMr~W2(68 zu58F;ixxSSl_cdWAS=q{I9P}`pkaB|y{(r*MIGPh{WHVsQ;5W8=e+-{;RolNxZ6=# zW23{LG3Qc#`oWwKMP8F8D^+uIP4gfD*@s2CB~KLEslLMhvQl0^2xY%6L^Td7RFLMP z>CXQBo`4QMW#A|>F$^!ifA5}5z`|@o`}wL*4yTHrO*YHKt@X={=TI zjv@Ay^723URdRk2`-R$9t0eP8i+7k-=X%3*gR557!G##*>rdp7!Qa1s=cRv}JNHDM zOvc@R?=KL-KB)(N>F2l5)9`1k9=6NFf%JR{UyEV(9k9CdTK{Q*4%DxhnwmQO#l?#k zTU*--2FU5m_XGvID~`hOJ3F6*@jn;xYid8Me-`@BTWw|SYoHeLAq)O!81}+a`}NHW zxW>>K_8@%mOdFxXwCMxG^<|uq=Z2+`m&m{hm<^v|l&v710_y_>&?<8ndAdy@Jv8=I zY9_1Rh2x_mM>E9Ahc5##3MzO2=Zy}c&Z|o_27@#I{$IW|P zzwTuPM3g?aEG7NNLG8L10S!N{_9QaAp4}M|o2%|*I6NS*&=739Q}PQrBk8M!9u0ct zY1&HptnemwX&N|I4pxz~0ac@h%7amG3IX$a?k-&&9T63^6LfrzmR)M9vopTZ$J~v! zbNoW>6P-^kzux@~r(G`!cbL1zVS#*2c^-mB{BVL)Y;%Mvg@)5^Xg!OeqjeMCp19W; z$5C%w0j8ev!B2`n>qiQ)><>w}#2*|t`G{hDJO7lF@-58xU+4t~Af`SOKoR6V(SVR`x98P*V}bET&3)ZNSH&Yleq zQ@$bBik7rpD6Z~`q{7_bF)l3$@Us_7qhp^kFxO|fd|BdEhhZwm+??hob?(y@v0W~? zSvO_RJ~uVx)QpU9KvydiRhWpGw?065+SNn~Fmu4|#zHYT*v zB?z4>*Ed^tFY;MJ_cye%ujis`aA!8)6}!a24fbGZlJ4+ zUGxs>{`Qjs|6O@bA8dCig5u&WYAKbMJW0hd8)*)qG)@z($jA5;4g_H*FOE$* zJXGbfLywHi&@M)P&1A8(;OLD^flk5myBEMNP3v~;!u*_;ZKhh5%p$i|MM``;>cc&8 z9?p(v{(9)%+C$H!^a+7KY!7|DCu$)tFJP&@qU0Y{KJR<}INRI!zc!iT5<1+F)&j>9 zTzi>M2F}wp1GQC0)x)yMk z;lSca6NhBm(__n9HWBb%v)y49{#)wqEzjfTA8?vE+dsfM{%qMUG@+ynVNRyG=p^nj zxA5Tur!hY#n1hlJ{%x`d4=l980wD;0Wj{Yog^xqE3urhoDbwLr+|O|@3)0{7kklTG z$xh!&Akf*%A0L>_F)8@$d|-v*))XLP6%4W&@-Em51$+>q{<@kG6Xqt4g8ZA&`fT$XeM_m zbANosb-v^%3~jMyMzdzR#E(%_Jo{pnjE754D!M}>bR@y8*-=?h@l5A-b5s;^w1S~Q zz+o7Bn70aPEC5PsoZpyLbw_6w$+HKN{CnflXQ4*^vYuzDA<(l=67*NQw$uGsenuuQ zmC;^-3N=&oTnbPLemfEulu9~vYA%UYk4Jy<+y`E|fQ_%fN;qe_$r!$6JT*VsDeQOyc%dWY=KpZh$l zJKrVepRIg9h}FI`siOQoZ1pnzV2z6JfKKcUAsH#|)Y=CrM{X?wcO`yJPPVLkh+~t& zVP{^+-=heY0(qoh=>exYEt$PvDEUBQPf8X)c@Z!r@k(ep*i-6keds0T!0X2v#`Wye z&FF^^0Y_EiSzgJWPu})fa&4-VT?=1|vI3`U>fa@=^iJB-Vkc8fGX>0~$aQwZYeo%; zeEB(>^%Dvj2D267=4BomEs!=<3{Z7#3-o+N3LI`vRv`>`=6x-K_B>iDIEgI zO6hRGRf^z|6CvkR{~bE{F{A}LHm&}CAZ{CD5y&OYg9*iCvIP#`9GH3&62d6EtLwg~ z&D~hD%A+N~$ET*Q&gnW4@z-DyZUGZ@)CY)G8Y-$1v4rG|7O&-nR<*Lac!5BN_MaLx z8>#!)c`d5byAx%CLYEZm7cq;SKRQwALZFPURtzMoXMGAaGesO`gSF&{bECH1FLVY$ zy*+8i8ImqX?>iY|LiH$G$PR}+K2Q5zEY^0sES%P;EGs=I&L}F?F=E&y+#QB~UBO+y0u+2_K>>E~By$jcm`Hmg zV!8n>8H#Bv=uSb)S%+m+L9{)DJU*P}HuH6j-L_@wm&DsLVEW5IOJW1+OlY+~HPALHGCe@0BfXw^7?f zI(AdSs&L!PBYWBXKZ71rRApqtJ0$$ZbTfY=8Nrhk0w7{0KHYHGTqxw?}2D&?PKK#omydLZbxl-0{Sx ziuQ_URUS?m)kt+HrX;5vx7Z7sF>k}T zo10hYo?~a+^rX+4doPQa3$}x5UfL%@RaNzPZWXcwRsd>dESBt2x_8OA0&@i2+!$|@ zK9G;H|B`7kcL2uA6#nT2JSH*P6JkOcv2qSoObP2j5r7F4s_7sJ*S-w0rx| zBRF}`W7{GK?`^4_uzOas+0CkxmJLV)5Qlp$rF^EHDn{X%vQR&N{|Y0b!KS&a3N1%w zdiq?tUgr{zo3#abbCS5xRHP8@K)6z6g7iR>C>M0lu1=l3(_SG&+vO@5RP+nONE=-7 z6=~mfc1f#72`)*l#j)!7-tYw*OjOpj>37)J7e!jmA}g9~(j7eMT%dB`e6RJX;K@iA z4v!`E!1jJP&=dcQ7f=`J;jtZU5vZyj(%-L*5;W(urpL6S0gJ#-D6=c&Q$f*8r z7CelOm4LlN7hQ9jlCo7s(9}6!mlvBOe=H{#bkuQ9<$ew_h2O$*(FM0hxx2|G?sGr( zMt5>mOk|vSE7md8RNHm&`YTCTSSN@7TjhR11R;#O*vR)!>t902+!Ua|V6vcL>`Ot% z$V~tf1AG<4E=5Y^!kRQ>e`hb>V`FYVKQ84_SHYKXj=CJ#!9CuFC!TSl5PzX|$@@k1MHHVbnAFMX+?T;`G zve)W_{{+8ybWO9@@)6h+($otDktBQ3N=(vo8T2~U^E$BY)qjBN)D_J9!cvD+_f=Und>zM+o}&X7~;? z?1MGR$fzmryYY#KgHe#OEGo<-HQPK0kH4d`F8Dx5s2apYaG#c43A$yZgIx?4G96yl z!);uVt7K5qh5Ko;8Sz8Qn2(m0mPI5R&EZTcC)8HLvuhiTO?952nYZ2k*%c+ce{*s` zHp3a@l(WKnCw#Gee1dXX^B0eQt;~D}ECpvstE2FwyjR@3BL!(^6@p!Fu|q{kr)`1H zkBa}@(_?bbcY3>`%oV?KoDU9|R-0v=PH6^d`w_P379UwQMvQ58T#r`Sa1ePOC6aTc zlVOM8EQl7h9~ndmAZmTngqdQ!*nkMn6WeF$@w?F@k9_P)mtj8td+__xnvt*Dk-ev- zWsawnzE6ETV}jdf!)A&z&b)HiP}{tAX&w`{c(C{0Sn8DxG?Ii?)m39*bCE9nDW=llGB@3~nhFXz_q!b~M$i{`g?JBE=ucgdS z!{{|L;r$jJPZqy!fP zCqFP1`%Dwu#XQKTpk>Hk5f2DL{c|51zdwHcxc2vGg)`$^YQ(P#EC9WN*m7H&z*iYp z&{9$5P;m+iAE9NT=`{{-L`Q3dJ|rb2rKF(6tOLkFu;rS?MMjtD`V7Rv)6PUl)c^&) zWfLB_d>wi|$?MpWC10=uHJYMt>xjRP%PlQj<+xjMYp`L+zoCA=G>#`<4MfqNsfXyA zyUiRCIj&I#PMi46jVlV#=-`;K-8mem+t{uT2kmdj_Nw$VExDK(O|A@Az|jbHc3z<; zn5D%538~yK@Ayisck|$>mBD^#l~>KG_tqy9)nE-uY4tA503unG7qB@$FPm+F)JSU> zYUG^$2uw^HBggT0`D0V>X3e51Hw4%eUGIzzGlO5kC)Q{BM4cR--g{At`)O4xbh z<6^0V85%ljdWgyWkp)x zukP~y4U^7nomjdx;2=W1k%eyXg&5zgGjhe*nzzP)?9?-36TE80?>lfz;QJls zM}P2cf2G=HB&Qudk9Hb{;rDm*3h<@zx%PVLqYNYU0r{GFD&s9LbGSUV1^BxLvZ|2B zz5ZbZaXs;*<Y%N2K;8zw1&0&G!CuUIaPPN$mt6~2Vl?TPN%asnjNShA(c-%K-%5r5^YrbwwQ2k9S7H-@_Rd!&F*{ z-@-k%sm_I1F1bEbjw_b?@aAOs*!GX+NBZ0xab+@sH3Am1a`=8{;qTTfdhQmnRf0UzEtZVeR>Lliq63T!k>oy_5>^tLGibuOo616GZ zrO(rGu#3J@+^;k%(@ghR#i=mLsh64KEl4sebD}7e6A^Vbpc+-ujym$|FI}Wb8)=@@(U9YZGusg;RLcyLc`D)sI`=dUpv8`6(THNT$?wabJ^)@zkRH2IV%n~$vTK<^_~VdseZKF1 zlH%n^>PP-7pOk6`3`Lbjs}K9-H4MHsah`mKdfoW^*W^ION#C_gG*nGNJNdJ-UtK4C z&0D{EkD5PRZg#c^T93uW#6)8P0vx-4cVD1*RQ3jhLEJE-i+rf`Na#*iC)P6S_>mYl zA=BTpDofaZn8Ue)A(GFal10wtZpGHy@NgQI2eVzB9I`$~ex^@OJ7#BR&w>q$rxZpc z9TVh^7v@V!O04L(>`|HPeYaBewI}MF#b0%7c~~n;bG*+?C>dPlF_YGAWmrAfrl+!} zXO!Q+Bs)`3CS|*mzmE!Z#w)e{Y#b)z<%=U2D9Y-)1V+ydz7_ zDscQ>Zs#iI^&em?;Ar1XSrO;6=g-n1ocg382sx*r!?JGnM)HjIo$%3ZMh))4EN3!e zwbIC1CBE=%+4g+rv)1=MvIwt@28@jU>|e}0*SGs zSk~)ImuKhaJ=aC)aZ<9dL%Zn1eLeV}=3O5`Ev|xH+~?1m^BeT^FP)560O5W)@*F|M z9@X6DS;QjM=5`LUM{(bVCJ*!Y46F-$c-|3S!LIwi>MKg93 z!#&uxyOCS=n0&*GtJw-!W$xLvSx?_XK2OJGopUM4bp$i0Erji~Hml`1gZBd!i@}w=o?pd+7apVG@$| zm=A*MBZtt*`}okjauodL4SLrA+ZPy?Z&32b5$qjg+~CzDoqMEhfJ{4g_WZOcu@|^m z@p9`4)}3<4YiirtX!2H~;_XketD5pEFTV0m)*o-Sa75&VV(LasJy&#Pt*?@{22oc_ z7J8-a35NuRYYe30eW?J3N9juPec>FiR^KXfkwkL!5^4-a1Xw2wBa-#d{}9qfB(*kq zV(lL*+~F_b57_3aewf3h22bj?eQ2*<KYpW^aVX@0UH?O^R{WAPZoJ-Y znzT>AixlWWjS0r4M~@t{zxd(?ZYjH@7*0ax)eedwgTK7{6 zMvdbK>fH&oj~~CTl~d->(=|QX=E)KS4)Vds@V0gwv$XQsLJhBR;Rf1jSrQbt|0@?xW<3=M~y>%J~Et!?Fl@c-Vsj(<|1dDv+Xi5@L=l-;0TufN8B6yQtm(%$KI}o z_0p4$t9q8#-$&_SZxBVZ)49nfM?5M6j849!XUVodD7Q-&UP)#w-niSE>cm$3`D$cJ zk-aMa=#V!^S&Z~;ATWrdH>WE`5=j+Kt{6qHJGAdeIbAMjXtNH!A z+xZgbuwB2^XbS}C-?`RQ*9IB(3kc9R`2x6^N%`aPy-xel7LpLXKVC zPSv`r(uxb~>l)P0MDIiDq{5?%HKsX=u^=e$;FAVUS7|u1uFb$(=GO56w$QEI@!&1k z>S2McYR{fIf6{~DgDkdJi5#%oaVi5hfu}qOY4ZPm;IX@ew%GqI(X``cfslq|jkoxs zcrpV3+ROB6Sp_L>kcE;L1T+zAl<=a0;p#_i)11w@^)4?xir2M=n+Nk%#st`V&8K>M zl8n5|#Ah0F;O#)x(J&QNd4pI~JEh(m9G_2%m!$0&t<2l5ATtHpqcuk}{zmZe9?bh}ZKpAe2 zHq+#V(OEQD-f_$bJPxE|m%DG?J@FjeIWAjUW=wBFvQ1z1Yn3=qY;*J_SCHaS%_Cs3 zoQz<*6IuB`qc8tILiM^eu?kqLc+9_^a5Z@tmS=b4Bl`(<0E}$OJne`27XUl-pW$YB z>)Km2W-RkRzh%dIbaLm(R07Z8WPA?Kvnxv$wGVx#lJKv0PqES{=O2og+_Yf|!~FX% zey#A>V7?HU_aN%_i3$VgDEOhAP9xilsShs?+`iT6!SwFmZ@wJ+L@e`8PpKqjPK5IY zpmm&#?8ttRr`^{1O``CdwVhUJ*S-HG!f>aza{cVj8Xc7?e_03T`q@Jax@z>#JA%bx z&vCH8pLXAkR8@_(oG&l``iv>r(nL}HqC(V-}nYT5_QpUG-nSyzFQUl$aoPTT*h;j^S(1EDM+x?!MXy zLP|hSI{lg*VHPn^Z zNg_c-`SUc=A4qZ+(hF{>s@juDL^_YIjcI5s8#ERPf2r!Y{a1nLyFRfLxD)(!wky%? z0(JDSKYzxKH}?)22fdH_Y1yQvy#`&qHe#4-{xIwkH`~lz_44whFChceiNy}7Q*j*f zUtA}-4a}ctYVJi`$>X}Zywt;bwKubA`6sisw$Kly(AAL&RrP0=vvOd%>c+-jcIm(g zV?9^*LZ;3fH%2Nu#u2iAuRUE%}6 z^Cd4jm-+a4?-xTeO61yz@$3)H%kb)&nuOzhiyk}r3<>+awrwi=8>1wAY4Kf?wt%zU zUtSbVt*v6RlDyr1L|K4DC)VoQ<8^M0LWm0|OxMclD9hPkyj(x@$>CPJR^_oXzBs|U zz5TWYeD|<)GcybRmF*$m!FlqTtpoYR3vM*iRhGihWw24!dQZxqMgd%eyV}*J`Vx~R zz}=wE(J9%;Ab&5$-+@4fo7#(p0A;c#E> z&$tV`ig@SVp$m!~26TrYvKPd{ztLoWism)h#n?jMSlITrhKL*pQv^UkK1`w!Ml5cF z`t~3sVB~RD9C@L#%E)}^00LQ_VU12oN)X^6^ zW15spbh&tVMvkur;8#rajhij!Edtttet%$fT$(~cP;nV1N0g^KV=_K_^7K@mhV4Ys z7iCI#h)7F68)|X=d4G4gYKcTcYMI$I3C<;;fol=9yfjVccFk9>kTU$HarhEkH z$6%3GUwF(7(R$CKO0VOF%`x?dd&S!G;ENc693&!!Wuf@!sHnmk5RPr9&dNP>U!Txh zUM>kaZ?TSHyvzYN=1$Z|^@0JF5a}4|yE&n6YN{CG5DcIF7E&!j(~HF90E67lKbI$mp_X3Bb@!qJu3!1g*75$fAoF_W08jFpy zx{>SSf_{Y?yA=>pm@=C^o1>uIcUh;Cqo=2*@C85xVqfT)3|kt+Khu0>iX>@mk2@S( zqK+@$=lU3O>jEvUkNyQc-uQymA#lHz@#>=RVXfms zTj8c*rxMqUdy#L=n&ucYK`KKnZ@Jb2d8*goS@*~Dv}*<84jwzpMHZ%YPRu>;a2trR z>G@7GHON-$Ru|7E2x#_;sAY-WycwbV4AbuTUdGGaVz{9fg(A;D+4F6`9uxQbS`Y{l zZ^6n?3H_q2Ee)Z>tVb7#&BSeapZ%^ZDR$?@Lf>i3uC;%YLelEahwCluc* z4V+Hd73g_trCv%LzsQ&7rc+yPpzlNCeH`-b5nRnyn9u^Vsnx9Vtf%JZ<-G=2^z|Pd z#~*NV#vqD=j?ed(=4n?k+T*d~lY@h5t*&*(KSZA1_@*Ub;82(O<40&+#lxhmES16R z-COM{F+KwN&zpMf65FmgE{A%LsM+qdiNS6j-z4BR7^eg}+S}(#%K&~Q?-dAY&bTsO zZFc`VD6$dd*he}#dszEH3Qkh6pVJ(`OTtzf`yDvJ46N+yzDp5IeE&2btZJ#yc?qfQ zBeuuxn)XJ?n8xU=oEMl%kW%&<6_CXR95;C~4`;?UjF-yPXxvU3CaXAqduE6zT%9t7 zQ`JQ}aU^CY>v+qS_12BM4E(n5?&&I|dU15YH{{xRT(DNWORM z%*>0eYSi!B0`RVVYQ-AF;*Uc^3%sZsiWV7uX$W&;_G<>jg)VEyDK9@D zMxf=Gu-NAi{S?E+r35y(VJF$hJXOq^(Z0{>NaBo~`^|cL(2&s|Ths3#W#4;LTue(7 zKqZS(oYKpdNW#Rz)2Dp>fg(5B<^aQm5?dyJR-yL5ghBShfCqpE&Fzc z$#QfR;!BH*z_f8UH~}Dkg@FYhwBMd0ZZ3cPsmNl%UrtnZv1JqKWq#LCPf6);Up~;k zJ$5k5e}&69Rt@!(xZO$aUp8rQyW}>y(Z$0sXgrP>lP%KBC?*tp!>ckSJ?rD+EmT#v zQ$>+H+&@Q&Dh)2Bsd?pH$;O8!lfU_r?qZ6%6GFWT7iA1=0Bgzob%tc^G7l^MhU*M?9# zxy&KuzjLrM+`SsKGcj@~+%JCEmvij7*_@U1xwgKEh-PaKi!0uI{Qb+Ky!^rJe8Wrt z<~iPT0!{HwDWCKa;L<6P$gmu9>JipsXXnmao~uu-M7}Nk z8PgV~zm6d8?K%xQ;?+xZS=eAG-TA*cg$4cRj&s2EzUte_Hm!WXgj>*Y6ftfyhue+< zXqz$H2|O{l(Ojp-@IIr~wexSPG-3e`MSdm5gt0S;MS;cfms&9|0m*n-ib}>}lB4jGjcu!GY;$AVZqV4aZQHhOJ89fTZNt6I_ji8x{&}BscCdSP zXFmJPyyu;HpOF@0hk^JAGBs6I@qcqv2nc97BX@HyRwiyHRu-23KK!2v_};(&|E@~g zn>+klLHqYogMbh>hJZytM#I3y#w8#orJ$guV`O1x=i(LkDk3f}C99yUrlzT_XJ}$# zZe{D>?Ckc<+czLMI4m+IJ}D_RBRj9Ku%x`IwxOZ9t+S_pXlQg|dTw!XWqoV+;NbYj z#r3aW_rITB!5Oi7R8Nm2@Add_wE1B{GVqSf;MYa4pLYM z@|4@(1b37*+VHWyV$V)7nDM2dGC>EU-W0t<^= z=CSivOd1Eye}3`+^FjDui2d8(Hb0~gh#1cR{@?F0d7lLy{)-=7on&NWs*$*co?yZ+ z`-qwm5#bRP<>5)o{t)mGVm{(9B90t-Mxx|)OeWy_z+pnRt@k9N2t!DJHqI3>8ZmPa zk$SBzk%BrOhJ+qPj;GpJMn6KlD&1-7>QK9CP7HFud3UFa>weAWJgO(j%5P)oyUsN+%Spu($0mPr z=1L@Z?%`DI1*?^t$d0i~W-^r1!j#9c(5S^|<44o^SMH6RJf)7`CN-8bdOLX&RankI z<&%WXh-07Uq#&V%zq)RgHw?t^ zGL#*`B2!E%xww205T2@Gn^Q~t32TX7iV*R{3hghkP@d4NjNZ_yKqJo)+X2nQcpPw> zG+!oy^aWF+r!6z6$mciG3^jp)-`N(m!_IXH??c1cQN2-Y z!XJyfjfS;myT!`L2{Vu-&$kn5lCDSNT}nIsUK#lRK=fOQP4Kcy1^ZMaLekViw-RaU zjuZ=KMksUm!OYwI@CWddTfe|xONkgjYHq3{f#LlW9zbj>*gJf|;2Apxzj{J=bNOa5 zidl$6Qug52#9^Se$ zfSBG4!!eg;4l^T%UY5bRPwX0~_{9u`9;m4qQN znHb_ZUs3`29bfvnE$pc$@4CbkuZ8T1t}iq796GsBW=BA5}C zO}P)!b(Ec$i#^w zX;3;PZo?;pN`n5ZqA0vHl0Dq!I9XVmQ3!tuxDOB|3#NAN{{8$LR1jnJcr>c(~yJ@UG-aBk&k3Nq({~4@;^M z|1UaZWN{&+cmN3p5aCD1-zJ#E+*56T3<=Sx|1!j0j1fz6g>vSBRggS;5VMAg*97jo zm9cfKBFLTsYZmM}=h1L-7P%&tvgkmqq@C&*{V}wa*Bw}*W6D1eg6Bc?Ja*z>*H#5r z!qxBO+ApU&5_HMu3>JZm>@tZ{CRaAytH{2K96VqaV?gTcc0vnr-HJ2w`k0JIc>2A` zm$q<&!Ql5DA{(v3)#m7yJ!7Yu=hyA(5wUdDRw_*75s`&DxcG%|X+4z=h?it< zTP!Q5df8>Bpvq(z4*N^{u}SI(OZi)%nxuUN->lnYWr{ z#45D?{eHpzDvA^Y9q`vuEOfMfSr{pZqVjD>l%$Rv>B&X2;C3mtK3j<%f2UC|R-oMH zCuyH)@!a~+py3o$V_IBTf*@yJwn|1s&SEqgyE*sY63%Yr7b8Ld6r8}|i9a&>eq2T> zImbYM4N-r+tIDMoFo=4>iRy$YtI{!>OX&kd%wP&^^pVPNY(E$ibego<}xmpa- z(p5R}-UHL&DL*NJ<%YpOI~$8uN=1%6<0rL37+r02J1znn7TGy#$t9-ev*!f0RUGZT z;{`HLhXX+u3fevL@YK@5`uax}WTfH5*byd#OvAnlPs)i0kE;wCF-JpCMrL@KOMeq! zQt3oo}XkHHW(3G<6SM(_^R=vhe-{kW2jSLl3Gc_`BTM zlDnn6uZr_sGtGT)w{wTY!FA@|TXpaDzeA3NS{GyV_1luZ5xfF9r4zlAQh0pSBOE*g zdcIVFNv1^(OVYXi4n1Dk;NDcmba;XyW7<6}DPG#Oz3D;Z=gS?W`=Q^j-)C}HMqjm(x|KEIbfy0Pkj1}yOr$Lu0Uk8S1~!y$E|?=SGx(mMHqFfP}BqmLE5&F~Bhv{F8v6 z5*jfoSE5=@xSc#+Zwm{z<@h)jv-|-|`XfUcREwFH#9rcTQdFO@%Q#>GrH&f5N?a7X zvRnr%#Y!TkZq6<9kMz7)lf(KwCl)>#W*&G3*EVHAZJ^)JWWqraXps11_{QP`O23@q zJ1Z}l3t;&mQbjR#=R^9l(+z1Y;R8-r2^W&sD5j;H7eZo@1lZuB<4qx05mMX8i+|kY zfn2`n7dMz(DE0l?{)#LP^6A))1`Xa2z5~km*MGF`O1R{}yOk{tcU14Xe&O0d09LD! zH?jM%Mq7ZNWHv*cr$f8unZdn{{0Wyfjq?S_5$#j|J!LXhj7hBpkyzWC=K?D-PrHx}x#mG?DqO$FQ z3$35wWUh{acUbPc2SM%4B*lQ(&q@!}l4$MPOkoiN^~n^CwhM1KA}`v*Rwrq>O;y|9 z(LUX!hakfoOPU-sjakNsm7aXc_oagR0A-xdV&%lwv_(?YNQgrD`Z;5J?p)J;I4NY>^FCP>ke(k&(^C@ zHylYY_+b{{2tUv%T`UL`|JE8RCA#YCnk%n)L{pb@(82wue8Z~ktqr_iZ8PfSlLgwQ z9oxCLDO8$^KA2gtN-GRvCVVi=#tI3m018>|BY1^Bm|eVWC4(aSELQr!V4c{A@u-q( zGBS?aMDJ|>%+*$JGNpsc{(eM%5}FPJI0=iWGl?0ByTS^10If?y3A3+|vT#XZz(kDO zIOoCdcMt4uhBPGkQtq4#y)>a5&>x^3nj^SO!KD$7T8V5bBw1}QBROD;E}B8f3*HasB}h-tol$~ueRWF4@8}Myy6$*#nl2BMN^;m8XLdCu z&(^ckJJEdY-~35pP@SCDe=t48s6>0s((J7)Mu6|1FSB@H z$uQj(VVIICiTcJ;-{Zp?awPQ66K0FhuL6dK@bR;IN;2^7$W#RO@}|oi{j)X7Jcne3 zfm$;yslDrRiSp(Ht0(2Z4UOdmKR|D+{FNce8C-(>u4h>o^k$)HvW3$eHF|->|SpfmUirx0-XJe@#jbSB96hiG3 z%^ACe8*Vv$hsSrpdlhEzQ`gfY_?`Q2t@2<5E{sS>Axu+1SWrNE{VU0>A9L_$O!ac_d zqh46lL>_3}*f5_20m+eCP|W!OdSw+o{jRh%##80&*bhB0z;d$bMd+ZF10ypB1uR>r zDsU7qPCwn)`tv6K$WeSRK$~~fkP}Y6Xrm%OR$@ncQ)8_WQR6=CYgb$g0%lfoJPlSU zx~r=g?EY?5eD#fV`Gkst8Frsu67V9e_>kD;cwAD`^7rtkd*v=#N@f+#E5V5{ ztIv*NLea$`fEMQACgY`yjRug^h6Bjh?}-$dRV`9X8EDH9xe_dM5opIOT-s)|gb>2{ zL^$oM)tG{`2)|W-hm=ae_9H!`jbKaGf_vn zMoHTVH>~fmXn*?icbxs}EnaRLJ6@ibDrWSuvH2#?m;6%3$MqQs&%c?`+7&(x?sN+p z$5*4Q>t4|Q`jU6k3h0v7o*y6|aUWnH%Rm$1u*+Gw?xYmi4*$dXSJ?!;ayF1NbnCOg z)2oYLj zbW3qKOVa3IVsk-Kc(q_tjApeI7O3Pv_t_=~-_|XYczl|hcpaT>jJnXslLYrtd+osW zALxQpz!%*qI~%1v(c-^JA0RqfIgCMFPt{a{=c<^p)*m}Q8%Z9;vf@>)Ij_qVY6olH zYK4O%q%ep|aRucg>=vYYcyiI_YYpFBJ1Oy!YR}3wQd7)}xSy)tqcds6qK_hwSMyDW zM#MCHsuP+K@>mw|0RB4YP{RSYJ=vwPW`LsP9cos{`|{}Ue2#DTs8~4UxB&Gnhnm(c z@_0fW02UiH5ghC+5#M4M;Kce2=g#R$23P6*Fwsf2>P&06Odu#6fuwphR=k|zy43om zEMam~HL3uan&p@@l<>d4oD_92a8y@XHTB#@t2weFC6?+U%q8zm$zE~G)r_xq@}?Mo zOACJQ6IIiLrg`Yg<$Gz}21h#Ad7uff10L8-WdtZxeyniLdZtA(&{*Y{ z8>;h0Ma!JrsiqewQtFPpSLwcRSKeM!HZK|IV*9afbIvHWRLOBKarNAPazcBRcNwy6 zs3aVYtX%umyaoneO@Cy_j#>bIpq*yz6P%&n%X08@gC&@Ue zN@O;x1}Q74P@;>6gya@fWZIm#Jz*YGGzPE6KCY+#)1 z&%mfz3Do7nu`I8N%;3!myc^=K&nbHq&pzz?xIq9<_EX&#zUi?DmV)7caD|Q1E*(Fe z?mgwaOm@;YqXRp*AHq->Q-?DNi#_J&mkO6x5Sj?)}51C=d^2iH7*( zh6AM8X-N5GPfa0lAxA;byrIj=k5yXMX*)@Zdb`-Z&C6B%spr>2N+?pavYO3p=Puc+ zxI(`e9Tw&)-SnfVIn3oMT(}q>M0hb!9Q~CZpojX&%)rb#$H+9Oj;0d+A=S_JU+mh> zo9jGXTi#C7mK15Y`6{>=poAF%$fnapGUqsL66^O+#gKS1DgcigZei8{v5eKPswaV1 zK*OZ!USfYagAb4hpB!*dFWnOgkS>IV0p+7YJ)7{2Ke-ZWbN^1m^3EFRH!OeBHa#s0 zExvolI?4LDX&4j;P$;Om81W)lSIb^GfuZ*`$ZI$V$hD{aef3?i0>3K;^b?Ph*yC?= zXIyoA1>Nr?u~u#weMS$8@A5y{Uxwn2T6t)Qn^rAdXTH8nTaQl>FwQDn!agMUUVm|c zY-}#6>m$y42m6A-krsW?BS&K`~*5|i118|P& zw^7D>88CRqF8aF#UW?6nmVxv4%eS9=;GeH5oweXGu-D!w>-876x3)|J&+o^r*PYLA zHtI8tV}0$t7_}4@9Jhf}?xqoTUe!*d@|DLr(-ey2J@A;v9d<*cBJ(&Q*T}M87|6?! zDR32Lo#jfnLeG6jDQ^g2F3v8Ed6d$v5Ed;qN;7is+fUwA$fk(trGy2KvHEyO4K58z{)$(6aJ84BZa48Hb#Q8;UAfS+_B!%x}4_7^8> zePVdVEPs{(s@{LkqvKupiG4vrFn-f&3AdeE9_o*%q5sm-J8+!; zSvm38l;9)g^y!W@^bt^+TZrT)aCfJI31y8Kqj5C+VQ>HT(~Dv8Z|K)5-6WdoT0K$Jl_w?6zKZFaCfqXW(#Y`;+&J{U5 z%+G^3HY1fUzK`E1O2>ipb|d}{MXZl8BDgxcO?9fjy}o#acLwQlvgeq^I6fr@|50g3TNEvOk}l7k_$HM*xb!Vn89HaJ>1i0gVVj=erwNQ z@QZKq2Pl$;elgYkX!flD_=m4r>N0aLwrqZh|Ga(2A(dt?mC+vx1`!|WH@%|wmB=(H z!w;CmCr+#CpM=RrnRRo^p2|kAZ(xFv4wR#=?e7{>p7Qcjt`3+beJ^{D+Hxa5cb=P@ z5LivVxg@w}YwINcX)Ml@RmS~7Q3!I<4IBT(N*^r^%O?mTJ(9nabbJSfW8XYN=YOyz~gD^kRCf>{r{4XHE2D zRS*fzjQzsizUni*_%P0uw)(RyP-JOHieMbFi0X%7zKN%nl;XK4Ebi$NCPNXb$Mg%< z2dJME;V{*H5!bPM_r>*3$B->UwJWIpmW0GxL|EExcBR4qDu{b;<$Z*v*RlVu^0z1X zo&z|{3h(F;nyMgwjKnR^WM6S=Juh2iUuNv^qNR3y%urU*fc0KnrFL`gZy7B z2j*bkt*JE`yUgP3o}QiIbM`pw_e{3cLQYXtvf*)U_3@<$G({wQE8-0pcXnn*ouD8HBvhpdv?ZZoIE_vG>c{0A zB&q%I1fd>R)=p0^i_J0kUADxT1fi6qhDa>Zq~}+Z%IyW|_8HpIlOOznI-X3)U-v#T zbjN10IH>RsL{l75_q$c;+(r{AEW?=WOogWY85$X0rh@5QAy*|=I8<(p`o%0sI;0Bp zLtjie(WBPhA;TgS8V^R5<=!5iL|KX#P9a5_Vi}Ge_ zCY2=xMeb-(D8i_?kZrvvnw&R1lGQT?K7%FO;d&DT(B9*SC(feYM)2R+@NoiEeoGIs z6UEv)t3k1`GdfrDhG~fMxN=L&x-kE3L!{lrF>{r*`jXXwocG1vujg%IZs!AZ#UPs7 zdL}zlJkmRjr)QgOD1=ZhP23rFa;hh&CG2!->zvdg=bL>IED%7EOiL?$1@jzmNyZv` zD#YOix%xXg%`~K(89xxW{q;SuHc?TvFj>2_)|npNQcpoy+@7vRa>DleAS0a>-vims(0x%G$iNciN;Y~9}x#tCiFZv`b)pq5Dp1Z2_JV|mHwNaz8A z<=(3VcmA}zzsH0_o2I@v#@yt8L&EV>39YsN%sL^~U0ePt{JMMURi`$`*ZBc@WU^VT z-8WepV9c`z+Q|v-ywWUk*(eurjc~-8+)&I+TCV&u{~BGoB5MVUsoV|9o?Af(7D-!W zwbe_xz_`_;a4|Ie9{nlAtHxP9R-IxUm>QU!+)-{`$`7J&bp7m8tu}UT8Z^kt$-y9? zF@dcjiFKODQ(%S7X_8V$H=yEt(Vahkm!? zkcka%)7m)cDG_#TC$WK+e!UG}>V%T4ErON^kR#shuWtEB#-l{!)f9rHZa#d~D6A26yP@+W2)4gIKyQKqaDfn~ zfVVPS2*A0GJR%fZ{~G&yy{TAgwLojX>@4)X1n8oM$0k-4W4nwrjFFl!f&YJr^4j2e=Y4ZC2oC0y|PpRzkyi23eIolZQ)k zDc-Z})NBBL`#!1hrVdQ#Vu=Y>o^G~K{t@*U0IdGiC2Wq*2uo{x<@f;_XX$c}7h6J7 z5;6j{aT$&DDhI=+b*#pAlN`&B7|GS%Q(z54lK_R*XM$x*&|nd5=#K|N3>qGUs1WmI zUq>FJ^L^t+v_;c4ZBqlYi7c0LwjA=DYRmF@K#ap|QCBh=9 z)X_YnPSD2+r+rycfUXRQ?0xZeDf(x~0 zS98(fW<-{PBPbdUnAF8Zoj*zEh$`UxYx1Exb?5?QK5-&W?5e{met;&JptR%RmQa*j zle@Ym48xZNuG5P@y%t%7%AyX)m~#0C6h2Upu9eP9n22*XmI8ZDwo(dpl=ok1^}VX? zO3|`IHtbsi_?vt(HrLXAfzI~l6q`}gCEf!pcUF7zJ!}4SYD{f z@{DyB-MNT$-EkU;{NpDf3&qpr8vI`sF9vtKuWMho{_>51FVKcCcI!v)PJi(oj1U$c zU}Jd>Uk<)01KJ4Da72>p*Lxksmd-N&?lSguf-^WZa5EfLy2F3Na>7ItqT}Mh=l$!* ztp8AXIExjdg@Dm7d9bK}N{p+0T8@vAK9tM*QY_LK57lWU%Z3Y5w-8g~nUDu@h^b@A z_!gr0TjA@lLA2qa$2(dFVL(Jt4%q$$)Xx*TplY!i{?IVui17aa{bHo=^nu$NA*vJ0 zi2YLH>3OH?8FR0Wi}yqRf@kBA(L=0jt>PTWq)H(d4fT*<^y(OZKz1yr7wTWQpCc}m zuY$pDTwfq6rS&b3vXb8|Pfz-ccAit1^|iT`c!4@1kW7wcuE9x6!4yQ7PGuN~2EP3H zVjs1p?$MZ^A&0A4MUDrSu$0`$8qeFJN%BSvdYC>$(Cl=CPBRE3H#U*Rf;X)S8bpFH zhP6Ni_C$!!1XPC$7sHB0JHaORZ-j_L4xknQ{@v##kt;_v9r^%)S&uedv9`nX9yTPg zHnxjLi5r3~&|n9t@`rm540u|8TC^`~-w@Y?>wsT>{s4V1cQo$3p=(V%g)`ME^^F6X z@Wgk=PW#cGvdJQeJiB9GJAiiu-R^zt?}Vpm>P+J}vv!%bDTf@sZxjv(cx`>fYkO|H zyX#pfU8(VVe@s~qb9Y@2JKpRqLdPNq>cjAQ)3>8ejL{*5L|h0HD%tQyApqiG8i(QG zcEsk=aWaT;)aatK0y0_nWdvg-0Qfg0=V7ThtlB9oN{oNYKR_=mO}(xdTO(gS!Jv=L z>_m9O{Lok1M%}paZO*vP$TV>sZ(QlQ_u7kT5^bR_J24+ip;U&+^x=pC1*OR?6B5KD_Vo?O2_3E|`;|aBP??>X5hwPn%Aw{@zbLzbpd4c$ zXIio!=rflm>^6u)f{v@!rN_Vkyg@-04pQe*$$?i{h;k{2525L}i4RG&pv^8SRDc4q z90Te!Mxt%b%YTF!Dg0Qf z8TjI5(g9E!JqA1i?MUz+hEhm#uB;;J4+!96ya4@->?WHBE}4(6ukv70J( zmPQZJ@#>I-&6xdSnPax?GVNLTu=Xu;hjdi%kU2Tv14Pc|1FY5dOn)Ct3bTN#uWozi z@6{Oj$yvgD-aO-@FX*0l(Y0r(a5LBFe)m!;NhKGmzxKRpfBjE!salVnA>0rJ)7)XRk#w7wA>@N zH!X2ij&aB!=~d@9=Qk}ec*}FVaAYdaw3lZWRzg7u!%)BW%spV3GI-b)#%-P8@CL*4 zrii+zQU%C(V!GuC12`lg>H}QLU$}^sSSf^fdOzy_IM~^5gcdULAyZP@6EOne;g-87 zLYqkd5R5DYuptRywUsN{Gg=<K zrkw^VwON9~XuOMmrxHc>k+O)VBI18{N;ff#VArTpw5qJ_RiEv$6Un#@GVyDYnuwLHLn3)OLDm0nM zpH$|M7<49vQv;nxB9N=#Ed#ASeW4q!MWV}h$ckdeXwo}Xd5q7H|B|xY;~ue^Il1Lp zFHUsf)#d)AveV)_az($A%HJ5yC2Xt}=J`68N?Uuf^M2Qo#`^?bUb*`7@6Yf00d^`K z>)Rr}x}Jcff?39H{R4htE?yhM9i*5lG#%uR}0aCF5)L28BeymH?@bE zDk5v(6Gj(}MnC}tr){Y(e&Ewn_ur3{zm|Rqo~*er{+apLNdSZQp6$tKA|U=ppYOz3V2h;-tkg&N_uaMu{Pv5=}*1u+uL6T zNo=8@&=mvW%yvht>y$BfHr+L^?vXSQJr>M9KtPHvg@T)j5VPWuu4|<;?9A;zJp==i zuvpPeDOCEZgeu)_FGh{a0n|w=(iC~-Kx_)_8`s}@6H^B<_!hn(1PdAMjVx>RE{>e+ zQhqo>5G5SMTGJl*seFbZk_Es4$+&^}PqAHonLO8-B-;^vk?&uGXDbV|=Jo;CO1vxH z|MZV6{@f4A{m@F!ArgPt4+URVra_1?RYfQhAhUQ%G#_(YW;`c>d!F!Q8X^(j$XXM$ z**WE-Giy|2ST@!W{i->@;abm6gE4PV@d1*ga+Zmao0?z8)=I$v#^cblQW#iNWlX0g zYLU-*Wrqw-e))cDFj!rG1mZyuT_d)8F?hTeY;hL++c9sotaWlx1+GZJvMWdq6{aRY zQs{}?iljuS8T=bPU+#&y#cB5!NsY%i=NC-qMT)Qxi7~?QU2HHL1%yN*7hy3YozG@z zMwsJOC4O#61&ZQyUr&v^d+L#PkO`8wYeiHPuEo9~6FI6;J<}0Ixv>bO;wXIPCK*h) zu6it zWcU@5@pt8c!EXSF{%yUzC$_1z6L$D#k2>T$&QO$Gq{uA_XEvDKGGUETzng|&tU;w^ zcluujPE@1^2p7*;j`Dr50II@pfVtgq(FZ7=)(6o?*xlwf7QaOiA#V~h8DuU@1TR-#VHYD;NxCb0-kz z+6hi)d{d2AKZ0kqqJ50RsGjl1z7;n@GT(rTnOSU$X{x(nHYUm^BPZ_9h%-&IGK{fX zH+F8|rJHzS$>A|W9%|G`$~p{Rhvz*_c1is|)YM_djSnYlbbwD2=0BaGjmHSj{vZyC z*+)2dMJuV3ucDy?4HFX#mXslJ+d(YCAD~)#XLr}hrPXGP9-Y)EZ1ZMC2vVl&ax>&q z_8M8vop)^8KN;HZ_dTZ~Ilej5{r>(r7`e}UV9lb!65r|VckqKmFk7rhor4yBPb;N^ z^wi$tDC)%~JSdP502ypCmc$GDaz@ zK!=lc{9k!sApFnQjj+>pyhKd3c+B6G4n#f6(24S9`}oy=)-^@v9r~|bilE8!5CX<| zr0%M@4i4W+DAVn}=!zX8XBMruh}V;SfQD&of$?EH&Lv(QSZrKRUUPz&tDZy zy-rn{pP#>E50t19<~CVyWU*#ob=qa5TrsCN-*6PatQJSyODPuJaPYXWG68<9Ia=%R zN1b2EVbKn(6T&;;@eHi8Y{Jh!=U);&>%9V)HccKk3-66I5m`>ukdq1-P=jFZJ8 zbh_6F3}g6wwCse32Gr*@$;dSK!(Ewl$)txygd(J7hxtRcyjdBFBQhrw?!8W=7p=>Q zxAZ_U;o{k2&?F2?dezn@!OF$``k{weW<|S(>9%8Q*GA(4B*^M>sj}VYP&Np!nB)rp z=TuUqov+hH0!DSlOO0Zg1Nc(o`~!5u^a?D(70Qba2c)x7nzi}7-L-MA&KjStnmw1R zLUBk(R?^Bm$GsPhM46Vx919Q17CRnD>{@Po*IVg#7feh}LXl*hFezCwNqgd`x2?e& z*uwA%w&h8ZWfZkA0L_v+FZb?v*K@Tjk8f}Gzu8C~0@RxIK<--K5<60JB_&>P?I2o{ z7gS|3=@ArbV5E!JY$#30e`K)oGU7pr@ieHIt*nuk4w#v>resJ<>+0JxNkQUTG83`{ z;5D{pi~@;m(=iE#5iE^MquppfK)*RCFv{>f)7N?Wal|BiOHbrPJkk@qS&c)GdvSS0 zBk0p&Z-ekx#7F}=MHEr{b)n-u&j_K0L>rtOE%PJ*7C_maOH)9cB@OYXFaS%TEb)Xz zILWe_VZ%*~X=GR-O}Q5l5mf*Bfp!}91U7cH#zYBrIp1B~s`_^u2d%>s4J#XUZoLw> zj7@ybx>&>(^9#)dX?;r1JRR)-Ro6%jVysR#2*yOELPv(7Q+@tZnPo_x1wSGIw3Nuo zG`VFW=Ij(YZ5cYkE)C$j)*q{I6@Gd0zgCDg@!*7oJ)F1%i8 zs)J2!&v}-A17kDb*K(KGhPg*SixM2g7w<~)sGGVt$=?_DW$PFg0V9kcS-YnW`| zGg<_$FWY>iL65~WYw9R%VjTSYGtd4# zHRSR3`VAKgQx@{HhCQ0g_3#69&-eiRj=G_GRba%4*MN%2I_bhEMWZg8j94OJB2l<@d_SjQIoZsqX`Q|m2cLG+_VEJy(z0b2v97npG z>Av}@YA4i78cc|7fs+9h|25lSwqc;H(9!B^Vu!H`c?eP<37{Lv6lTK)mjoVWhAtUO z39%ym?tJ~~g89c-$#TxR5E!pZ)t;Ys;3Hzoh-TBHvQ=}qIPC|$i zNy$;=Q4pkL63mcao>cM>&po;Hsa5=M-Ci2XtDd}(>z7Ih9bSGinP3V*soJ0`qpXn& zoL-zy^?cm4tc%2IXSBr`F5Pn{QbRoc$3MGe2#sz^kUO@`09SE}!*-38?!gs!#k~T* zOUXG*Q7Qw<_k4;eHy*#D8MbjWBna2*Cny%G_+S4jOuwcB^|)Z(s=mY3$Rsi-1f5B)m6J^9 z3kJ=_w4`?xGD%ajFTflvPe`fUF22;1j({K-m@X!vM> zCU=Xn+8m;}8DcJ*ZAMw9c>FJ|Xe*qeS5CwsWa;Tg4t%!Cd$l9+|JRS> zVJQ?4@+%+4C1WEPwtiU9p*5Er#SFDZ%nWD4fw#G&fk5>L4Yw5iTw=CFMu5~rak4pV z?AABAdc7q4!qMYFZO1wO0V1Uf%58!$T$;F?lLI)vo=nO{3yc6bWrW0Ql31c=4E~!XHj2n-#8MVa^WTZU~7LJOT1 z_6&Z?SGuaN|4I!e$6ZXAzpi2E-c9J!%^0loj^Zu?0hNqZ;)o4ai!$=*5v+t#*}=gn zNIHqBn~zvL=N@}~npiE^il;GEohMSP5U2qvD%D{&1!wVAFbW}Ic!U1=RtAv8d~RK* zlQh)HI=TwK@<|+YBl*z?V7ILMlae7xm2Nd0?&KLzpe5Tmh{C;3^yznq_``YS1bad@eMOUUyE&)RCi~03+kCjlRm(tz? zk`pj|cP1ECiYlqRR1+CY&tCTUd715D<8*0sR5c z$fP3RMlSoAywO{cB%w~{Kal;Ad!fu)(Gu)svw<@8zwZB(qYS|vb)h9Kj@UirwA`7S zo4areJ`wJ=jcYkc8{#KJv-ktr;}K~ncb3=#jW$RQ=Yw1ZcHx*MXOGI{t{#qOYFU#x zJ6t6tcU$|w8A|Z=K!?`BrS?Ne!*NRf5wXM5(sElJH@SWYdQZX%Nv`zAXz#Tz`|2AWusc)ocpMeBBs$E zBSVL@NxnI$6EoffxWX#Od{Ht@Vn(QwI%|Zhx=UE;Kjf+)q#y^W!{i z_#E#W&l7ZSitnc+FN-SnU%~B9hnJ?A9-qVp5Xe|uQN%>WOAh3eP4Q*AzW$R~if@76 zU`tRFUlO0fK8*xJjlc|zMaC*a7g!VH5<{aFwSZ;N@$9{;*mE`?2;FexgyCU@-FZj| z`H+{8*-K__H6Llw}{ zL=Pqf(ecgkutdbb0hkaR7(y62h{0S~dd>VrxCdn$z(YQ%5bwW|@v5msaiQc1%8v}a zv6w=+0vD0&WG6EtD)p9gSx6HYMqs(cMT&;VC^9UR#Rfy8(g7qSr8LoiIA#hjC{IKz znAOO>sj$4buE(yv)3v_O%>3$PfZ_~F7`Z@yEBNw=7HM0npkQl|LuUEa=@R(e@nFc< z#7LQ3T@ws$)ArN$tI8aAs_5`apE?(Bn5^`4XE+MQ=DEEKfX_cznD@6Fk)flYHI%p_ z%k)Yh_={DlyAZ2D4kc%XXXvJ8JO$P1dYGS)+94n8jj_pwc@(=H(?4x26NHy@Kjb2ohk!Nd`nE1M^)YNPJ;b23r1IJ5YSqj*}+W z8pMc%6de86H%5hpC`UYqo!0g<+C-GV4PSF;R8x-}Vlm-MPwZ=y!EcEA06no>Vd%mv zF4%I|)7b5kZFzFY zd)>WI_2m!M+rlW#p|z=%O(IRRh^e*d)6PxJoCj3;N6_|z!XPmrk-<$|IdXqKJkO^F z<9F#Nbn6op$j2(aIMl~&48bWJG-_g0mJvPDjl~V-X8*S`MjgdZokh%@2RBJc9+@;DA6t2egB(+nxF?ht`N`*1^ zCTc{Guhaf7_pr$PTXk6So22sv7)E_di4k0IZhJ}iOc*5@K zy+l(xYvg#CTrXy=3ZndLBUei3UHl7!UMG&1X_>aYoc;s!$RzvBfi*S%ozh$i%KdnG zy5z~vc9O7sYXyeE%5_nc@$8V7o2Iz-_pwHM7DwgU~Spsma!74x7fF|CpXv_7NC2mF{L)iq9_}`S<_% zmwCV6{>gUUQ?bq3+J(wwKgeu%nn!vSs7!zM@)p{l5f|n z!t)-}dY`XYYqYs7OY4QV(UvQ7&%1I6cxiR8Uv0HaSMzP^Rb%~lrg)Ney3P@q%}NYr z7LI|-Lu0ICBbuz-Oa5HyiPUUu)$LTBvB~|WkLfX)DVNQTJ6=BUd)d4xZL?(4ZyYq*n|O#Zpf?KZe;k}UPvq%wJ)nDOE(|Ae}}{J8NHJp1rb zH;0c0IKH}fT5@Sm%=P9fk4MHXJ<2Lp(iRd{+KZT-%~U;?PcehojExtrF<%X9P*p>XPV(_sSnz7@>EOi+*=i- z9r)bQSmH+E)Kjd%p@>`sscM-*vX^M_F*;~$h_KKya%Y^z)nV|sshd*+T*{gH@G)>4 zim%l=aZpBeNx`&p&nvTv^js%Zo?51w61DZ|iHE8aU7pOCJTb}N(%7Y2DNyNBalw;i zQ4!N-EuNNl=4tTBo6+{CgM-hgZdO}V%~Sn4S6O=;)ONj?_1<)Olx|$%(v?qFK0g=!>FR6WCb!K77o9Pjk7zFtWPv<% SS0nfpLx`6TBSFmET7K+ literal 0 HcmV?d00001 diff --git a/frontend/public/benchmarks/tts/magpie-f32.mp3 b/frontend/public/benchmarks/tts/magpie-f32.mp3 new file mode 100644 index 0000000000000000000000000000000000000000..c1e7ef3068e198677f017828fbdfeb4305588928 GIT binary patch literal 17388 zcmagl1#BBZ*Dz{_9c-ABhM5@~W@xx!b~tIn)G%Yi%*@QtFf%hVHk<~1oA=ZE{kqbx zdyizVcP-2F_;^OUXJ%!@*`Wa+L8hjvD)DcN0suhC8@ZTqu`+Qpv9hrI=gt3l0^j;} z|L?Aho!R$)JD>hNQ~&_6ZU8Ji(kFCuEL;L&Qc?;UdL|}T4sJd{L18gTSy=@oRShjI zT?1nf$in)&gR`@nr?+2VU`TjWOiV&@T4qj8eo<*fMRi?cYez>{@4)cz*yPOo($ebY z&i?+<>BZIU?ZeY6_~Xm}waw&T+gN!&cK#a_ki!4-Qmnfpg<0yqhyQ=J|2+Kw>heYT zToGZY0Ngk8rEcm_%_=<4e2z`zXrH;@$dxBY&+>gql^4D{r1U}Cajgmi&7HvSw8z`&qqZn^HD z(bQNF;~8K*lS29jQ;xO&-rb(YhYP0eMU04u$a(}a{R_|bHZn3YwV3EnLVbPjkXzu7 zm|YB$cAOe-G2;8Q!_JypOB%Yk^nHaXf_F>8URNA^>)%ho*8bk4_sb2+2ThFVcJG;x zHsLsk9Ff*b)wWUNkFz56!uO(bI@1VSC3@Vs5b+l`^e`p*V;$9x7}F1&DHVN#?67XQ~l+K>&NUb)*Y(&gU!F>x;?|)SPV5h%{~71 z&up6z-!2h97yewH;JiJlW}vA%!is{td%3!;_Hoa)Yp*KRoyd6GTZAE#)4H~NEA{Cd zDL|+XAA2iLkpjR|3Tav+-nJj@*fV2=5=u17K+jrfeSnDAC=lvDFH1uNyOb2V$FM4b zx%OI7IvN@4F^XXO|H>=JC^uNrc$tY?aE%PW50J)>vh6ZbLBlPG+o+Jh;9-=w zCBa}x5GFLKO2@?b!Np1tz4f9~8^-=MT%enUL^2|WVOyhdpH_CM>X_-J&x7st7YA9s z`3VjRgi+AIA6!L;cf_7HLd|fF4YXLH{4%f6x8hj^zeuj|dtYiK9L)~J4;l6mk)NC2 zBrKn-a-FD$x4MSA@Ky_NrIszmWbXfVq}});fWG_b1BA=m{#*?(nac@-?ll?J6#r4~ zsx|~;xwYC%tFSSm`8!p|@^!Ab$bPsILp(1HpMp$eS^6TwR#)~3`Ln29Qypew`K!&f#=f?yU{(r9ZM)9ew3V78O_iMD zQ>_7`F}af3EKBaqwwNss^{o1oj!6g>XDaQ%fpJUL%c$04TiVi&ko^?$b+}{u)vy?S zB@4*S)Vc3|V))m+r(7r*Aj>&MO3!@0rIeqe@TK!V8DfdQEZR1i>w(`m85tQhIp4Cp z{$`@P8lv9YF{xQedw%*bNQRMOuL2W|h z1QxIWRydYqNazs6zO>&4wc6EK0QFwOlKQfbN@}8My&NMm&g@7Rp+FvtW$rgDDkUMv z88Jfw1uZiI(_ZWlJgNn%y>o^{g%C!5t6Vn zcjP&Yi~emf%T^%_=ip4X9okv6C^Pd`9Q?bZRIW<0hHXly-f)v39&s8pXT_*f>GN;Q zdar1(1%nO14+jfs;l)1`%ii4u*Q@`eUhWW-w#?YyARJU9$1uz z)5h3-)v&{oI!fH@XI2Z(bDa!B(tfY<*SH%I{s7T3o6P1(&dx3fa&YLs;r~^;@M)_I zZMj<^nTPS^@wN)jMH;0>^HqYxJ5N+D8CqP_Gc9bSm2GNTAYGkj$Q3PUIozkcXCLp#CnD0vt;|mFraC0Su6&EcA?uz3>Yi`J^ zpi+%EJydE?3Zxf>1?@maXFlmA!){gu1z-k$fS8%)v73pdvw2ifqnx<8X_W?~&O@H+ zsli?gNvD-;IKP9F6K(02TYs49Ngrlcz}X#45N0j%#}`NMu40-`ucT#{YvpV*eq%|N z#UBcNr=6}8h69<7g4?=3q=L0eo^4;96&EBjWHzsrEI?|5A$9ZL}wG7*kL zi&e%WPh+(0lY13~IEH+Xu@IT2LZ+==#&44@Lm%@gSFuHnEKJKSx~&jGPHmk|d)}xX z#1WZ-MBo9v)LE(lRvnqgl>Wl&&90tZsbTg3;-(Qt|E()MS0_e9aIHXvZ3!uyD>L3a zWf2O*@Z1sB0KMiKcj9oqIaI&zl(P74Aux ztF?YC2g9qoNB`F4)HY>jnKiSvz*zCEH@&TzrY?%Sj4hTUnbL~9k{m=K&Z{FCQzc7F zZ5WnVtQ}2*ZGe^B6e7+~W(5yo8O~It(wAu`~DDS$ESC7KF!}?#yD`@4g&xZk|c?k zm*-O^Er-lcFQ}3GB#^Xnq>Gh>cdSvHA0QXH@`tk5=~(xzQ~e?jp_A%|Lz}_}Gl~r> z-ofw+%rVmdWo5NP7m67m0nr{zdPl5kP*6aOwviN3B88fY7-Mgh@br?ko)ZL=sxXo) zToDjzNH`}R3qn}}nhG5jkC*VLtLnaESpLpO&@Kg0%hVtTf z%{u-1{%5Bhk_BlIpmw>{STx>(^6Pz&n&ro4!Pgu1#vHq-~Ggt`|cQF0M?JaAV{o$W4}3QuMXE}Vk&vn7e^ zJh~;i%J0Dx0C6%`dOgf_ULS44bbe=GtTw|=2@9dBZIV59Jb$d3yo*1>%34_R>wzs) z?;uNxw3k@8nmX)G_i?Eb4+a+Juq|SJAf^=$JaG^m?Yn{j=wrlQ(1?IUX9)gW^R|9klR0%*DsxS8xO9Pp&FiaY~C&;(X4w4%Zx@jpOA zWNWZxC?gUon0frTSS={yZFg=o@e|}Nb>nqnq%v zaJF;ayiwD?34t?wZoc{GCZ@y&cNY%}<3avv5N3JxhZJ5cq5l#>)%pEhV>rE4hn!!x zf>6_ml^_m$4}N>aB|28vdfupJX2;BmE2vgO@C&3zfkYmiFc2lYI{rT(KfqA8&~2?p z%(7n?R+DTmX)TgpPHaoS4d$}7DEVtCBV>w1bo)#ihAgqao}?&8)dy&c78)|?%cz7> z08E!G({?9^!`+keH!TXQ6rCO?Wdk;PuXjyypZ(^iL7y%#clBTYiLUv}ysqwOJXzW! zj9y!m*<1?+?70IzzH6Jkv}egb%(vq{Y~L{#Pso%SfEU<0WGJLOBC;D>8$PRB*{4k8 zmy);Se@m8j^L+MyVx4hg8Mr(g6{Io!8JG^wK0=0t8$^Woro5oRE^cg1Es;&u`fD^E zoAYo`kGw$-1-dvm;?B$heh533i=7rTy8bX)l*|g^>Faq6qy8G(9|ES_q0P|I=8Vb~Mw(xjU8?P-thX6v%rnbHKWWc;>jhMY0ktx8^uV1e|OoZtz+ zv=j1-&LR^E3s{t&H$%o2Y8trGgrhT}X3fouKhQ z;fqQoF04%B;Q*1!e}`gKQJ{9~ERFnb&6P@IQIWbJYVE0j7F@{o0RnSDqr0)J&2d;- z=k;qN6P;SneNG)L%;ak^d)TIksg3m9H@{U1c3SzP^$tF$!vcqX*I{uT-T>EHu{QJm zxi;f4I+NM17;CFT^63R#xf%7+rwVF|><9`x(<}BkT@s-XJYweC2Qr~U zSH|Sy3P^wvT?SQHLEOzr-PImmb+T`o` z0Ss2EEwUiu1VC0FmVlu+bSzIf#kn(ZMi8Y>*(*bHoGeu_;{y~SbB^sE)6?*p^6V#k zg&5awH;_XE0k)<+pVKO6yq6nlXdqGk>XApq6XquzMn=fcAtKsc1cM|;WuyotPbrfQ zpv&~ZWy{Hv%q&fXW=xn(<0_Vt#~tnm0YhU6SRNyd{1D|C496kZt;4lE+1&>A{bbR{g){y{thqZQn|Lt6M?9uq)1gc!|uX*?wj1WkpYYpC1W z_xD8-I6V;|UoSBeBkWo2_bLR9fq|r)gc5K`DP+nlJd^wfh?<=IVfEcTxLpqKdJdn( zH>hJjMJ#!AfLzj9KZ;E+|LYwq#iHLa-8nmwVJ#ff3P=ZU^$zJ&)wL$VbeoNXybJs! z^ki@q`R(6O`~p5YLd@ngX-~>~&>v@lGsUkIKwf{*Y~BOsV_+gjJecCvfD#0*UT?V!JtZLHz;JrxnMpdNy1U;3BZCEmX?1 z+-Xlc-D8tg{}LnX3q?}N}NRI1@gX`ZJTn^Su2s~Y*lJ%I$uz8wejwh{_M{URg;|V z(48_D>`&ran$gD5unvq!5(d?YbQ}m29t7!$Gxv-x6_a~Vg|v>m@~89M2YWs*XprJ-*d}z+(#;j(@E;&A+MlpV&+e*3_?l_4RHDEv zL^AREVfj;B&TF$W1!tvm(^?Gu>C+OWg0RXk6-{j%Sr~uAw<;?505Ao_QFzWg(#CQ< zmn)7&{HOayHXPZ4G%WaL^|+-(aFVV=Q82dnqOR2~7<|9CU&ZK7vz929j`#KZrO#&r zuI+_+%I1SttQYU)|9o6G9>*HZYU_2$EJu)zg^Q^J~hlio5AB#r)k5P@@tBcDO;xG~}JgRN@REy`mpP{%^P4>lM#FQhYis}KP}`UJS~9A9;NmOg&vY7Ho^B|A>X=wD zeDm8WfiHx$X4&*OOgl1fwbd{{dpr|STcXF%R>V143AYNNuPA+w`$Pp~xVR2iG=|V+ z?6|DqpHTutirBtT(&@E8AOQT(YY(9zl@S^*8XH?y0?L0s@{P$*R0s{u6B8yzVkSx% z|3WzuJDtU^Hn#Ibxy%&q8~C*GdFtiwrlqy*JEO;D(mZ0@Kxb{lPoIj{r6c9a+qAvY z$adSXNxR& zx9{RfsYobTzJ6tOq}<`d6&(x-k?|lpItH`NPz235i$)K~{4V3r0wQoC2uYQ~Gxfcm ziNZN#!Dp9QgeAMVc;08h`p~TOsuQ3E#Y0S#$}W9m2!|XRWrg>=6;S`!d1csf|88_dZvT&Y-i$sv(q#&T4|m} zI24Dwr-ZrMHfA6s2ICdTVQgy6LTHj5C^zr~##9K+Q*?Bs?N38qVXmcw>JKst?dt`k&})gdQ+e&DVeA8mAzwSy+@0 zY$3Wt`{HB*<6p+GNyx~j_1;~-dg3@T4%Qstl=&?=9akOTR_AS|Z5_YXg<)IuL#Tnl z9&)l+8b#it>YfZ=rEG~o8eW-fZMvoWd@&>Sr68EfR_E3CGZDlj`*Y|1xu3B{r6~no z!M=A{;05#Uha5z-Z+Z{T{grO~k!Yw(HbNYY7>T^?~Ju z#qjy~yvxc?{E`&{s`Ej~Q~BK~oK(uRPKvCP4-kk>^|bEkt1m#KAy#4|V*p=gwoXdH zlFHBBN`^A6Aii|bpJubEZgD5hj#$Z#Fhbvu&h}#_i6FiPA?AFrlk4MG>ktm&xT+}@*8ne5HJ33Mw7z)=j zA*&d5BqJurAiIu)R?}&VgpG{*zq95xv&Xzm~+TRe$?q*_0Ylxf-zkdIsgcCd5$G=Pw#%wo<@br3bYtbLr% z0ISk0Ee)XyAar_t7gO~ZRzz!<%pJ;vT*Mr70TJ5Yh(mju+*eQ79_gW{eqX54f|tt| z^Okk?X-rRNZ*a>Y2mi~X*PG+L%g~-n)r-i)kYjCD?IFSDpSKsmB#nBVu4?d*?vR#O z8Eh$iy@?vfXJd~E9Zvh#NAsIeZ>`CRw>OioU)L7znJ?TN;_<}O|Hzs%Ygd;yuyaz* zo`g0+Cq=~o#7>7v06C-0g5YS5LYYEvNd5;1lcoV7?`YKmLZ{G27aKz4EosXu+syU* z8{6>?iXq*Yim~=_lwp8Pa~>6>kp(E_ABsPlRdt;X{hl-0oGt zJWK8r#rPQyPnhzgfm0Z{igcb_@}gS?mDsW&9xM+}=zi{x>06o(Geb`>_+MXJ`*(J+ zSQfRmp_q9g_00SXWSjRqDcHRGHgBVl1AsRW?pEP4(R8`9L2IzPaVf=LyZix7RDJZL zD3nP+D06+@XvzDC#k$GK!x-cu}6d9O*TREjeiO^WC{F=oy7eBVe6c<5^mZjOF8 z6aYTtYm=ZPjekM0E!XTqe+s!_p)Lq89pw8p1R_uCu4s9tvKuRVa7xv{n%@!ilBvRZdMSC-kQubj%1MX z532v#R&3$^@`z*Vw`iYJ0E4<^)Srsg&Y2*cQ&h1psc+|jW=j*v-|M9^RQBza}gOM!HdNxPYdF@dFe`RqoM+AQN)T^%b`G>DE8o%FV^mzTYMa_DaBd zm5|uCPecU`!82QHU{rkG)asz)Prts)FAk=j>j{bIkUeYdV`sI5PTePevhi@~?5N5L zWOm<~?&UIHl>zm@esTG{I^uM+0E}m&rJ%w_$~?*TG1dY@$2T%9wWcGrg>Q>w7VGzo z6io_A*;3Bq#^Oq3nw!$|3JHmo8l}Y^eF~D+2O5v&8KY-*=Z5~UeP0;SToxIRN zRN|?`?)&y#crhlyuMu!$<2h=0AE0)sy{xKjQH4-^U$H?t1a#Cx6 zfzO)`8?~T7!iou%&lqTyKh+zcu*6F{#^}t$fP(C7G39eV((ge|J1@7sg4l zr8WKP598yp1W}pe%KPwPshG!!@)&#_8~@!`Ve`23KJmK(%U}%NDNbB=4x|V>T2TK8 z*@$w42q!%a!9r;gsV*;)2!Kv?bks1+v%io`Qh zB}@yHnvvl^)K!|$KuZ}b$rEp*1UYn+JjtHw=t&P~Jk0$o%k9@{+&`+z9MeQ2siO1U z91dW5KmEzfV$~=5pR%DZN2TnqurRU=cH`V`&5XZ5OcIj0hiz;<72KFZ{gK^Whsa1| z%1CiOTw!0JRZY@(N(yf2M}}Tm#3ZBkS0&U)F{spNq(fVOp;!&2)(j*mnkpb>!RB%B zX;KUKtdxf^)DNdweR4j;=^JyHL)j&_?{@;|U?~o_=rDPBMa=3Mn1H_VAMWOptTrO%PSJ&eTplSr9ZW1A}b&;Q!Yx? z=N!ZSwa43^mKxA$H#awXqi?w-WZIa&s0;{f0ohdjdRJewk`*&@^s7|so3wq)ar4Ds z1cgJM@AD9X06%&i0~!<2gYjhYl0}0)>7gIuCDvA_iHJz%zeb@OvDDluP@#}Xijm-f zazw>=%+VR~AwDwn$h1o5CAs#QfkkCbl93Fk1pzU;{Ja>!LaoMc2rB&i{fD(%7uWB? zh|oXAGY^kA2AmaAb3=;x(PX=UVKn#1@s+qDdTa$qjcCuBh@%pKOe-CAIWEJ?FWiuA zU;|F@56f29wG4lMuMO}<26l%i>oMVVgL#NC%w<6aiNO@ebLvC$n+=PV3ZhHk z&L!|+7uVbp&IzfVVqTBGBgXA-N|0!Dpy+S>Fs2L-9n~Sy6?|x!H3Ir51J6l8`;u#& zrS%FQvWGI!3z)$#kl{6NxTmtUH$q^&UyAv14^oe@FfZitCY zs7nx^W$=7=SjaJ~U~8hd(1MZV;r3=E`$!x#nO?m#2%qsrq*eq_!NGGC8089y$QGzU z=J%1IJ0><+FOBI`hV%{(=Y?RlM47|`TS}gZC=KRliPK$6_QjUse--xR>yBQgnhIiZ z9*!Luze8>DU>0lB)NU(SUsVCMHYM{XgSLnKJ(jD#g}qqBb+*!ct|OB3RjFQQ?ciDZ zI6Fq063e}y=Jz71DJfB@^68uk;sohi5>Mz7rxtf|Nj%+*-*9{e z@W9Tn?(~WD>?>PLis)!*t3$FdE$>}yB^AnC&UbZXZ z&F?R+(`4RfV00t#e@;F=NSHzzk)MMp6GA2RPnk0-H?uIaF_%4Ej69VgQ{WP%In;Q%JD@;zEJ}}AN6VrqA^pw8pEISGf z)shfDTcsYzkrd9AI;d434?Jp&Ftj(OHSQfRlnyr|MS_TyE{LJUNR5%CliE%?kb$jH zcc=F~kU<2jhV`zEzqQ7x2e?B`HqDp1o9qS&C3NpQFESVV#~EM@$QeSyf9 ztcJ)UV(8G&jkal=+uMWsJQONr)$vSjX~{ih6vZwx(~NE zFP=BhZynCCIY^JKAf_x46NNLPvGuIbs~7F9LEBkndU4y43+pc zP__s!H`3u%Q{&Zofg@8~&FJ(*%Zt?gg|6h66#DvLC1d_>Qo!zZlNPCbn#U_w|>O6?wCcK=dC zo#6-tWDcT5?^V=_FH7FIJ)dQwh)VmbBaok;W@)L;5j)P|Q8r5SV@U`3vN@s>`~l*k zLxQz${VEIre$q*N|CVNCQs}e2Qxqb~;%`r$_O~&u&YEI2DV!$`YJ1-ygaW>&q_1B| zjRLL5B1V@Kwig)VKd~yqksHt>o%7s=EoWJtwyF^3XGiO<*+y6Wx9c#qN`u^3>7X!y zPm+jZ45#6OT>*t&US%}lDwqy7A%owckGNJi7p<@nmj?nB*NCvAM+7g{K4Bx0#@g{a z_IMM=Zp{F|EAF-;ocjTEpJ8yYnmg zwQS67CSab&6fH0EM^Y`^?kPqn+)H-SWMs}L4@oC zr@ICzsgOY>pR9&LRfb@Jnk<++M5FAY`#5MMxgx>bC>a7q7JgS4gOm_c)a$>ASfGJ5 zvN)0Q=;$eXWC1kdKvK=_Ck{VMo#EDw3=ZR`RKs!5`0}$wN|JoadPI8N#K>TduCif~ z23=B;i8Ee-7k|#ctWRM=i9&Ww2Pb4kw9-DZHR2(0%AUs5BVB*Tyn zSci3itf~PFo=9C^ewL=RdTXTCnNi^QbVCOTfCn+3rW%H1kIgIfGQy=FOn!hc7(Ihi zzpYpRv}qcw+I{BKnFB@N!<9U|PsCN*)=9o%{C1fg{owb~w8@URE{*_I>pXS+9$dAVgZ-jvgj0y&i zG!DmPrWPVM_D?q|@gc_X`6tV^mXwrFBXc;VEqGm*U9|o!! ztozFsIx^V^BXA8O^Q2^dOMqXLznnKiZgPqmogqZ}J`tX{+lNh!qN&dTBffWM7@pLp zR=x=3N-LIDQb}H3`rHE5Pe`XCPg9n0Wv3N|B}Mg zhj3R$YKC1`RCVtI{ukm!>0K&c;&CNp@YR&%-Qi9_Q-`q8srVwkD8J@YM(%>QJ{qNrt!~692)JNiCVIRJ11;(q|m}wKLAEX=GwIt z>SYu1od}`isi~{mXPRjd4@(A5H?qbGXt~v0D<{>X{EI|Bg_VP_*`^Uf@27>LqGDv! z6^2P{8iSxhZ+G#$-fc=f-~bi?pvC7%Qv%%!8gkQZLQo^**k%ai{s3i>?24u|YYGG0 zZXh|_e1)-DS#o+r(2Yd=Ad1Kkb)f{z>U>ov`gEyj&JD$%4o`uqg^9v-O}$Y2f95Yx z5CbrIXFPWvE&HUJoR*QL(jkHWRt?U$KJq&rQTR*ZLm@$;rjuDK$;~JKphBf@{v5LE zd5b}e4wuY323&&l6?1O2>rjl=cpKMx7NF#^A7H{wnGS~?Ir=_VxMvHcJEc?FmM!+A zGNX18k=dA7!MnXn!re%`5Ctmq{Hx>gz12eZr_e_R+}>N)dAmM9oAQ|0;+z@70Pr8@ ze_hZ0doM6)QSW;Iv`?~WnwH0BJtQ{vDe&9g6!`TC4DQ-uf4w&z@)5Z2IwZ*Md$=#1 z)u7pN!NvC|NRNrL?THhfK0ULcyQm=Tz-+_vMIr$PLqh<%O||f7*l1RWkp1#7{SY;9 z)nrmsC<`8aIoW(ALm+VDPLV*77@wGHZ4x9zGrsYp{5Z?0ON@H9&89M-^vX<(+-w6C zB`I>=KXQ{+P5SXM0%PTD@r`8s$f}jo8aXerR7zk4MergNKS1`971+*QY2AR!N{cz& z;-vZHZynoG9HYg5c(y7B@i~ccxJ}**;;d18&RT}n`1G6~I%)k$mkz6xk753L@2os4 zUN(C^pLJZSC*Fe7z;99}j9)O2#w;k=6RTM)5*dt^GA`XWm-e!FXI=U3w!VWI92W)q zkV*@%Pxp{HH3N&%Se|Fa9HgxY&*Y42Q%%L_aaYQX(~>+EUSC2S$1*yTcFNvn z_GKrtwq~>i-(Zt6`NX9VSP+8B8TwKLq*W4&c-F4UIG!)CTfd;vdeJUn|}bk6+mOh<|Yr3|!YJToU9 zPRo3r|8wJurvMMdQN55dTCX)%_r;}_xDtHpb<#Kiez^O$h<+o={MBIV{+=mtY29T% zUz;<_ZR^KhO}{;muXrrEKz~(3r@C3S?cjGY>~eAeo?)(!45`v)V3#J)eco63o8U;~ zg&94OizOMp?1_C<&WzwI2?z1cD)BQgU#UM=NygXLl`9!aIJOGNcH^|p%xJenH)VBp5;W@|@EIM~oFMl5bG7oc0wHAzk5lJX`tt!}HAQ?Q=v$BUi z+$ae`_wz4Vk;1ihrxCFRgD>y}(@alEl&7E^OuhD2`VXj_X;P)jL zvG?Dh%Cvcq04Piit4M!g#OO7xbd^kYeh+=4d|NL2(VP#^588e7W?bo&7^$-GVwx;T ziYN*LLE|(KLhSnl4xW=~Fr)5cgZ>xyTFn0YwUGt=@`}niV`W8{%MQ~JjMj-pkZ{_8 z%Uyg08)ZH!d}N7%;Wfl3`(y7*{nd#>?eyz3xju8I3wi&8X2Bklfht>ZQW6*5&z%>% zQU{2Vc;-ymw?Y#GuN=*s#(-;>I_M?TTV;mnF(;?VS!c+C345`As9~7U!lLmp%emkw ztCh;p%BhO$uZM=1{TWzlx^Za?(

kf^o1Aj2LOr4IiLLn)2r&C1Ye>#Z^ZWedXk| z_rNG#av1^D@z_4eAh|fpSM|SebQNS8)-b`w6p$!vW>CM`zKF0BQDGm!(c>A)X``aM zVxqd^P|_+d$2urL@2)sRmDE?lxonJjOg4IRjDGeJ{UJ<26`lX`s)K;RC5?}WCQ$fP z*1l*^{6g@0*t1)I{8bS^7!#GwsC(0FJD<^LoZT_GP_FZG{_s{UeKw}67FQ?qxK1V- zG_>q&ulw*&A)BvaA8tD{3B3+IBQNt~=mS(qgKr#8w~~@lqg0AYz?zRt zQiEzq4-tYdE?LDI`Cf2gb2cHkXm4@7;>8Rnq_Lio^)TsWpRIS9941e*(@3vH9IpZ! zfu_3tyPa#YuQKwYcJv;>cmHjmZsN|!+AEo+>MLISIXw0}2Ahod7TOs@tToyUf|*Pg zn2m5J=8<1O^0ce=yJq4d_;s-mh{i_(3kx@vhW$UqMs?^C5=u?lnU1(lb?LaL1lS@H zVOVIMADYNb2NBNv?3{wym5d&W&z^l9Q-@v#zdk_IRQ@V{!E^nuw+8Yi>LGHwo0V4y z(R6Op6A6ME24iKoqi8ALzNpA4lsRf8>St+dE9o>2__nyMcNa&H4yzKy8Y#;lkb5}q z#m#fhrd^$2^6?0uinz~-SB$>=W`qfW`jSSygtG(8+@%2jt1swNz3Waxzz>QCUS-P_ zea*w78z7o&r*buIP1tkmZhTPhDFZNO?xe-Zzi-CHsN;ji_r7YM3*l&;)D0DRa8I2; z?8>t07HwL{hd0pU+lch^lfNML`}pjd=Ssf77Xd4s0w-%(5y<;-fyMC#Yz&HyKxM+ zvnO|qg?_z0YI~G;S~C{zCTDe<`vv@B8u@m*NwBjnDE9g|5j2|HKfQ@ydUkt1BmZb7 zI&#ZU6Ejz2BMlA+(k{74u(AIoG}sKK^|yK+E)oetY8{xp$KC#PYUY;F>dGgrdd5tn z9u)YZ7(%5X$BsvD>+Y_PJ{730MbJDp$VBx;c!kE82TF)+biI|f=c(Tx<21%)bwPj>vI zk{!JQ1!+qLVgS*Qp0%P>b}Wi@(j%p9WbN5Lmrq9LGm4WQ%T^sp%b}S3T~w_;ezK9w zfYbV262~7gP$8TagPW@KCT4r3S&xm~>1CN_J%{55#W9W`87*Ikv0x&lxgkBC!X{ZpHb~+|f~d_7X);;N8L1Rh`k)xE zldQ-0zW3>&@O`kIrtrE$w*pHvSVhcU{y2tVn$2d1K+G@ZuXt&9DZ&gT7A7Dse!M$D z723{BNi?4hqt~D?71Ao&M_`J7cyhjkE+t*r7vLTsEiVt#cOU5A!*-Fl1doWvEhaf` zg%{O%YLLROg~`}kKbaaNHaZ|tAK3dfYW4$k!dSnVx6M8qgf7V171-*L#x*6bp-&N7 z5c^OfJI3%?p1Qo#k`n&R15`^yghw~HT^d#1TFG%y$Sk%{@a@lFbE?OdhDPN{rQ?}? z!_=a~mQ3>!<1IVeA5b&b>dDA2n-%uW@#(+zWMLGsmNT@x90Qh#HmQ!6m2A9DBlTz$ zX`Yq`V_jO4bfw{o>aG9os3^HU{wBztK7K`^)$)HzsQ=anx0|98D+b*|0-}fo$ougd z<&CaI`Z-`*v{QhD<^k@RlajI@pa)jZn{a^1iU<;Jq_B}8T1db^t*ie|><(LpiLkKn z6{*}JQJ>0S0{{w_1h8OKz+xQvh5Z(5(vxNZtyPnik*2y8No7vnFF_@2wg%m-;e5*L z?v%Ar9b=c(pA9uswGqB7m66GC;DOuSb!Pkk{9H!>ev2q04OhStYvo3tUbAZ2p&O-Y zAtOUs4=p7*onQ4ugugc*L}9`@VCz;YQBKe$+-V^}Rnt9mQOw2cgOm#XCB!fX9LMwh z!{j$cg-?cqK^(y1n-Go=oALoVqT6HhK-vjnkJKQU*n zK!zlVu`+373#Oct8LE^GmS(hJo`MA?cYC8$DySvhC$U4Z^2pV#_ftE*$H-sb`Qe$2w>kWz^287V?~ zUQ2R~WT{y-t|sz}N5S4?3S|p_+G+vL-h&PmF2072omCq(MI0KCnyG7I7*2wFp>LAI zhMz*=0PTkV`qQWsrO!R$XU!3aSjq~eIv zTpPx~j-SA9XY)`R5W%2MlMe?7?;ey5%_2!KHx)4i<61~{x}v;aI{5`MDn}x z{*^-wOB0;5gRg``Yi+SS7I1dtmPEm?Pbl>a^5j2!92+|-MBXu|Wh2;|_e96^WM?nF z-fC1Kc5uHPNzL*BAGv^=XXwGt0;$-WE6Jv0(j){ZF}&*0_2q17MW0Y$JFR=F9~Zg- zvE~eXd$?qUpG+rw7d|1%FWKI(nA2a8e~wUOm8{j1mG>xb;mE*^Dk%3jAJYV2yWvu0h{91Aeb;e#U1& z$+a+&FktHtw_Q1ehB|)Ug6=tDbGQyCDm5m6MTz`3$hZQCt0#W1zQ*!Nz^lxb_ zWbe+LP}nGfKTB9hQ4YxmXn{`0AxvXNIq6!BOoN^U6-y@s@;EX;lBl&%9tU8APE>B> zFHc(^T{8R|u^lS^3$*-hUs&jbA10EzMkIovCOcB$-<-oH<49;CI4*HFc z#kD8F72gWqy_yE1nCt?7bpM+RA+%W^y0pV z;y>#Ni$(E{l`Az%RUr5P{o#`HFoit~f1M3QydGd1kt13NKbvVN^A^u&|D&TDOhEp1 z9XPRvz!em?-=uri@6s1>~tNQKzW-T07VR{WEJ2{&T0m z3jI7vrbdsdX90Iu{bD}SITPtK1HusIZ2&rES)`-T{Ar5$nMWfsDb>{>>ds~%y78u4 z@?3xI=V8ZXf=Pen33YX$htDJrr>(2s_P`NFd4;_(WI&C;v^@z6CvB;<15>udPv8dy zj@(B4K}nx~KYDtzmzJ7rZAgi7tRkhKKl@~*E6N{wT1bQd{Pa3B{xW$H(qTBpZWdH_ zeHQLI>|?X?q2CP%VUlwcq^tlMug|Ijp}=6Bx!=CA0EPsBa-b5rDV;P+iyWIS#Yct~ zsnQoy^*vy|dwUnn;$!xvtdb0Bu+v-mX`=#M&Hs`=h3|?(@H~V1p*2IJ{H}g6qbu2< zN~Im1fTVaG6NaQb%G)N4`qVruuzF$=^t#KC^QSvWc_&q%#;Hw;aZl|BUBHiEeeep( zKg!Ei*}ZDQprgJ$+TN1D?Br!pE!sYp>WumkKW&+?rHT{B2rq*y2|)xcA~YQ1;242A z6GSYQnFgZ2FcgV6dfViAD@zhyW-d=}4wYy%h3_^4D?re28sqZ3z)`j^lEUf(^u}U} zpoy}H>E&CC#bMZs3i+a{WT@6nCI;yX2}P5!?w8Z;i%UX=$_a#Eq?`??Vvk!)tr%bQ zI=p^+zi{o?Y`?F&e1AKgE%jp5B|7dMaLM+FR%=m$Fb4GO!>W}V&0*$|uxCb!x9Tl# zz09P7-#dl=&99c@(QJ*hRh328C`FF4R$uyP#Hr<1*G7wniK0G@9Cs`I%aeixbafyg zSCb@WcL_kVxnImJk8FbhC&uj{q&E!%V zdPjN65FfT%UUtKDM9fU(wU7B?A2mgofW-`6PyT>I4YIlA_v}ilEd#XpEXex4`2uMV zyk7nLTXNQyL^iE`k}$f~TxD4t!R##78!GKtssaVNQKEsQd6q5kBVi(j^~$HsAWCGZ zvCOcb)o_Dm6&R<)Trw{M18Ct8vRl<29u`tN-mW|M$Cx{=bfXWayqr z4TS@9RpLh(ba}G1NmPAHS(!OEEVoL2h@0#i*af8h>OOUJo5-v(JK<&{v&+xk9Q~4B z9r`!pC0xc-Uc^w`Qowka5$i8|N_sKPgIHJ!E!%el1@^*x;;i@i+SZ2na9+A_GSc3- zNu`-0Iqaf8pmO}Xm(shLqub8MPG#EGI}ZisWQmRQ#IF|;#m~$C0hN_m)da}y5XY@J z;!DRNc{XGzM1AG`)W;LAEKB{_UC%YMwZ-0*cL@Oq2T3BlfNL1piqBy&^8xB3-ACtn z9|?S|vIY1m0TCb|r9^Z5C`a@4aFtEt{=~t;U$m_Gn67f-6X1hH6%5 zB_edn=jUYq?5Is!GcQh@IC0YJ$iy1TZol1ryTQuLq4b37(0IUG<};%dTrImkb8d2{kcx){ ziwlUFsN^zH_y0vD7mvlpMm>{Mqm43wO9h`>hi9!?HQ8g)fddLlX02NB$fZ+@%wz|O Uq5pvrD%)hmgimfy?=h| z+iNA6$s}v?TYF~CIeSiIAe``kx1dy4Q-_q)cVKvIY;tCPX?1mD zd-ver=jp}oo16QGXXx9`|NS}1~shc|13)MJ$%i4`oC}gf4u(l3<2mr zy|TfGiNXY;vd8}Wdejs4%&?U#9+ z4zFJ5;V_yK6!I@VdVPu}M(_K#Z#+!27lh&2PxIK{cNWUtK981+lQ;V*czAel8=&ds zC7^f z-lSRP?`lyuFXcj98!o-Znj)%@1gE^qJF6b*B}ux`2>p7C_S0FA zFw3=|2cq^RPK3Y{aXUx-N2ou%_SLS!r)PsvtuUPXiqU|t7qhG0<=hD;wuZDe4L1r- z#9Q)va21#Iaf1nZhBvBTvNOt?^aZ1CqAcpTg?QyJduP}Kd^$ewaNj*X&79o8CUFYy z0PYR|nqTWCzm6NhbTYDMA)EVg=4JxgNN%m(Kp-{<${@jrG)!BPFc6u99o5S7bv3t* zUQ+JcH`{}Y^W^A+Jh3~k9R3x3+^7`3SuQYFwwgvz*mNR9wqK&MBssN=8ySsUwX&43 zQWue$(IkAYpI3YfFNivkJ|>6U5+jJ{?(mKnK?_1hPMw9$bp+ERArqEJLCwoBz!eG} zC!zQZADdv}KDs+jF10QaQpNds;agI z#SK4Kru9wN(I2B`CIJhbI*&@0tqY#EL#Q(!u#K#AY#r3lJpGC;K_C*k%Vg$R+}H$7 zgX#n277<&u>UX^W%lBw_Z1Bf7<^yd<_OOIlypoi0%b>N z-6P_4Q}~Zi+>4#uj!c|y7m4Ja>eWJvct<4jg-JPxwu|o~KKe8b6ED#;U_`iF4Xa1v z6(}oZV)$;|7Tg;1lSK0}m5g&qx$HF%Lru zJ}POwd`be1e0$F*a+%deyXe$&Az~O{xpGT|LS_cE`gyO@p>Nr3zh5XXXH&i*?L6G= z;ajGh|HG9lBdPDR>+ZrDsE!`2raD<6&Zqf(61B}lkIF0Oe0#L}XH#5OoLd?6aaenF z`Lv-$BQB#&6QxI`3XWC)U%gy1({i;pIOcX#VdS2qw70Jem#5yJ6=|aNt9BH&B)|=T%0-tvkkO!Z$MX zFGk*-@`61*)#$Y*U{_7D^{l@IT80TZvgPdu(mi}4h_-@}!lB^M)u4c`r$V97C+O>h zWHqO*dOX#ls#K>=;oZmj$X`<9@*H+atMDF4Nwo-!@{3L2|`Kh4THb(dcqGhbbsD=zoz2rd`oUVU5<>=7-B2_+knPA3P}bDHfyhyV0m_`E;Bick{7 z5V53LoSLY6=PB5 zAH^C+DZ(wzKh3Vgb4domYIl~K9^4^1#CArmsGcojelI5&?-Kk zWbmZcWbb_Kt5VfAX6EQnN!K@QcYADLdt7P%rQHpo0>wVu744b}zZ+_bi*^J}NH!O% zN33QGevTUqfdeI+KFh)-Fz|%2a2?dke^SviPdXGa3=i6Kq1%VA5 zJbCSlojp!3sqv!zHKp2Q*u%B_>`O0)Gd27w_hrVB zvFn2M)fAgADX>$MFst(P+{D$@1H~kF;S?5WUpoG4n1!Tu9Y6q#7UYonq1ZK8TvcE_ z2Yp7a{7&&O)jBiGw%|`D63g>ok>>(yu?6~>_lEdj$4pO8*xv;rlrxi7E2oCK0y8D8 zhR7Ga?&TQPf|I>_S4PiCsPbIpDaP1DVL{O3Ovd4)jA}5hIMN*z-;2QzlEvY3yuS_< zT9u{qYBl{-aA8)@h}%qWSvHHVKaloVj#LMLh( zG6SH^PG>T!uhIm}c8^_G8zR}bTgX+ItS$~qjYnw9i_9Ua*1_yZjze?oQnvL@4ACEZ7aRgdkW^~0R8m?rb7?hgx{iJ1lP@?odD}BnNn#Vi_{=k zJOy_k8kGgJ6f92ExUF4-0tEtio*f1cPo)Bb<=^$=zbNuzc%;7(nACI;jxkiMse@JT z(1-=V!btP5YHkg@f$G7N{?+mea`rbU<^3KHdXAAlT{X7KLx}+W=_>n|qGG^(q3J1F zG3=&-7anD0i}^&YmP&eNjj^Y8+0kOBht8GN#;uix<8H5ULBaU_?Y?{AyyHSFl=0d) zSc?13379^a50>Kz=KQ>Q)gi&nsk=9+Wf#!G-9s|A{jPbPi421P=xu9C)+YuPrpq)( z6|+(&kw5;wIT#x2TxML)Db^>h)h@a#T4YLJ3@Vr9(4f#Uh32MYSfJgeg2 z;d!${%d9|1_ROGUHCa>#(jG)GTx=2)o>B%5&W*R^KzZh3 zMXZuHB442|1uysWt`CIss0!-h=PqN(;8T;sGz`JUzt(%|WR^k|}oSS zI^M-or~_iTt%Tb4{Y?Dpevy4FAvTBOlr(TXu7^>_)zwwzNn(wznxo`{N89r>qDB3Z zsqLmznr0g-iune*V!lO;J71jT7KzyIpH$`fka5t+I>TqDRO}&Mz2E7oxm;Gm*}L;< z6#jYjzGZH)WkqM9nuH^_rnL|-pe5r|UmoX-Y`!flFV zK%no!!i-8O;*WLm&D?mbDSj6oQpa5I&D@?9BNe0CpqfV|R3X~(ThbnY ziVnc7EymG4(t5(*aRNqY0=W=oXHVH%q$1^rJZfsN`pn?MDLGMvBf*Gv6L!8NhkyHH z6Nt%i7J>@@YlmLAxG+8wZ^Rt~iFAG!bE&PpfEOiJuk$h+1%&*4gjb+XSM=`!>Q7g> zcV7IVP!{NA07J&lzjlIK(Xz-DYyH^xnqMu4rIL+s_8VU3`h!$r90w2Y>8d-x%NaJX z*i2rtyGLsmlhm%?XdmiL(w^0K?slG6yDQ_rcvzQwGjvZ?T0d8u)9{OJ!M)b2beArc z!`yG|d+`jj7oFRG&7^pquo4rz@ZWjfl}yAUTZ5(g)SwIJjP^OHFzsxl#1d_#cKVL_dzge}@_uN!e4ig?j zNljBe(m6dy#&E={LbENea)v^t_yiR#eZ(bdX11yMOgy6yF+ z7S|C2@oH(?Dud4Ol7B~$(SN2>?oWSh7{p5KcBVFbU(rP1Z!n?*@}dUjp+$;*UK(cU zD?$Qup_;1U9s5kcTVW~d*r>utAcv0!D_L_lNl&9UqV2FN( zh#Q;37lGeEi?U20AL;oPLSy4mdA%We{M1+*G)pEapn>rLu}zhUT`V=J%DlXa7F*+q zt2W;*skfSXIynry6zZ*6cH$Amh$+=L0Eob0A3m)?IADU64I)EAFLbOP4w}TM*37O~ zC`+-+Rv-Ruf^vnqWWRCAOJgz`d@X!eGm@d1&D1{iLCHEFYl%$KkP7uH%7J7wFi`*! z9L-Hxg^Xo*;c8MlG(v;^bMGBJu%D{lLVdBQDI{1}okg{6kqM$)u+P;gKB7&gZ3f(R=$ZF@_+Z2aGXWZs zTTHtT>|NfZbhHFlgW5jP_Ob`D-EmWvuZPAf4)UBVS*Y&TR|L zk9q#LmYyxQStq#{Pki5bP%qTpGwn^@IZ#%r4rhFwvFF(|lPxlE?e8ISog4>S_n6QF zN#4JKJi$|-7Gk!nSY~=MjOnpu>}IUr_F80oQta$rc%$*}MW%L4{p40WbN$q+hqklC zo6qaDKaXliLDJffcEv8s-hqw7QP2c&P_Vc@fkMIS>@4A%5-FI{+C?b#i~AefU2+F5 zGos(uK>^Uesq*?mV?fYPOC;Pl*T5;a9+cPMl|*6+T2v8W|R?o zHTw;e&an8H{#Q(AShy@hLG#7IeD-t8U7fLkovf-2wsIN@c|Xr%>Jk_x3q~xlT6wL7 z7!rloK;7JeBh%DY*=J7?2#=tcR-QefVaV(O-m_^)d+iMe?4$sSg+NmlesTm6(jy7?(M366S#)PLxlvITz-j%by(tb^{i+P~YWwuGjm>_0g z5M{tjCByV9)uOhpZjCN%Y?ZD;LZY26E{^WgDn@|{=%fLn=`pz$8%Qi9D3pKp#1sd5 zyn*Uz7t4)LWJFgQXYRdUwLM zRMAfNv-#-X*z6N3q1;c3?i*KcmlF@D-oL(9#7(r+P0Y?gJB|4Eq{-nr!eV-Y_YJfl z<7dN@&+PVt5R<`bgW^pY#JtCnme#zKJ{CeM3(E$yrl{P8U`X?s4rVkCqJ6qu5&8Y| zu#HI*^uK%`WrXsNP_<4_0T}ZEi~Ec2K{=R7>|_CtdbXKYBQRI{DCGXZ>4>km$+FWj#W`!M+waoR;vH& z4WGOdW@$g}C7KLxrJHjQm_8#Xq18S^>{ySc>vp%biOSe?@N5w6It~mv!EuSqvT1w+ z;nLyZ%sk5^ZY#JoN~T_jW|j+1whUA?c2&-xLjJ^00NjIB;fhDs^*-$Wru-R72J1kxIz!v|} zJaSyGYA#$a@SFBD5R*3nHlsH{-b`!CS1icpXKQKp$;!MMmZC@B&~fLQuGSM4)*DEG zb_-|V(R)!iENJWcD-R^!NZ9B1k#@+R^0tfJ+?{&+?q^2xh{=aDacv!)-ZGuui3V01 zEyy>?f8D)&1&bCsp`35qj zUjp?K52EEhG%Bd+RX#H3Jj?OE2mL>ubqzPo(dZrpxk#H4MDWDPns;2K|FOt zIYJ>r?RJPT1cN%H{h6OK2VXKc2oPIHFLvN+X=&KPXN-o9fR{q%Kxg~2vA*wl3$OV{ z*ib8;kYEevu`sEx)yv5@@1*M1WOT6w@lB2^EJU#Y9N`lAA#oB-DlX1JfqlD;H;_Mb z({%&heZF$kFhtPzvl{dXj+XOvZP{74hmcI7ak!UBQ)6z~OkeJ7E7J*$X2AKO7=a<> z?$u(_H|WC@#YfDtOe|tTPkDeSU|wr}CxBp$JaCL{VOrQR{g1WwaU$}UhL;srCKi7B z77JRWc{%6puh_{hF9B+I$uuIrO(4*}l3lbt&0Fcb9DfYt4MP-A3r>Og$)S{MEm#9& zKHpw4Gu(Cpu+*6@FCUN3MCQIcroISb-FqrP0x3qJ&^zc;!1GRt|4Nx;ZG_nyCY!7zk5XD_t7JMKFZH(Z6BLL49&F z;D`iZEP<51Z_I8}3;nIjjyy*z$@kik1VbaVN^pd!OIyC*G_Q+TLXHNnrPiyQkl06c zwj;y`38mdTN-D~u+1>dYUM8WsJet%x%~c~ysXoy-AiGC>xm>Apm)a*52!-ATybZeV ztu;#UDq|xoY+EO~T`Mv-7>PmABZ|Y6x$%*{LKM=4FaR7j%prC(t?XF1H&8uT5GP%= zWH^#XNXUh6y!e3?!NI8IVd?P0Jj%A;5%F@9%k^4b0Tv-U!-)|WF#Dm^e&;>bT$(r$ zcI*ymf3uzzhVl>NQzcHxwneyeG#V~_=~}~Y`-TX(RR^~(!CdY%t^?OrbF(gFgjG7f z7mkmVyZGTqv9LPamU71`ht2hi?WZp$uZ8}9qo8r&7PNPr6_{C;4XQ|-6f#$*cEAf!Y zq32)xdLkFGPoHjgoY&2a5F?3IO5E}C4V1!#e>pt{P{ARloboAt>VD@4-8CHTq_l|5 zzzoH9qJ^DN7=Pq5Mebtv+|q|*D-)a%bJi#z{}4&|mrrw7zE(T-Bu=8NCShJ?DMi0Tw%2FX5kt93C)${^^xupzAK`zhza||B(tUAQbn_t zF;hx&JEOD%{{FE{1+z4c#s9Nb((V~a9o;LfAh_Yqn}3RFc*M`o#@Eh$_yC47#3R*M z*W0JkTLb(&mX6eCEj^{&Djv!ab^IHZ!qJpAzjnbOCnLuztt0tFEAPMF_AQ6q!i=J? z(fdxzj+dI>Cw6+sg(LkY$}FLgq!4>7Q;41HMiH%(1}XF5D`j3atcH(<4gOgeo%83r za<8eDNy?sm#t$RhOoDd2(0-mk~pR;;{jW9`@l*z zeT`C_*z*GNg4~c&PvC{Rl1Fy{3nMe2i=4WyN|BMvNdEH@?r`*`|g0|aa2=6 zD!$yC9!V~%0eWnz8(ux`J2p<;cG{D$&=jgnEcPsBVjF&SfIA*MDCrIK!d|mb{6lU% z%v!kdoa~2zyO-62;*&GbqGNV03RZmZS{oO8{`g3VG1)Bk#V@-RufyS+O@d#xDS@#l zANYpkw!$h8R*ZeIoKiYkQK0!SmPL2beO5?Fb?;cXiEo7~YvYfCa8H5moy&wdtZPNd zYnhl5V=p$cx3w%-qr9&kJxX00wO=63>&-@x55rD@#x1HoQ8lB zlm)Izj7CO45FZSV_5M5vKZRvJ@Ll1e^4JzbGIJOvt_MB}zT?Re{4hk`qO(Q9>A%1KiMv6DU9L2ybq^ZjC zXqAlEV&1B?G6x_ap~JEuhvIaeP#^bugzk^ka`$x&R1z0~KS?6UeWb>U)098 z84huW#ZNf>G4!{*^2tj{hnfzD@&=m?(21#VG?<)h=!!Qy*+1X|a~uyK+A5-vEntqp zOsE;mxXz@rcnYy-WM_!tCqmMDsW`DnXLGcY=&E!DJskXDrtJSfP>y1$aO!Qsb87x5 z9CG+frL|b^KGA-ed_*pzWo{=`5V{qI6)go$MR-f6=8@M?-S-w<1H+Neg91in-54{& zCGy#tXZ=|70yy0A3Q!VpQ)yuA_*hhyeHS(bzbBn{d<99EY<-eAGKEwevLJiZ46Zj8 z5@x3GxhomQG8FpK&hsKLv&@oUm`5bP0nNM4^XZWv91E6E88Ub~?j-vFD2Z7R$OyA# z3w%i6f~hh#y@&V{3@Tg{+YcDH-<9m5V^j*AENh>JO@?$Q{ zM29VrP{%{fxLs&+kB9`Dg-OvplUA}^B{jy-=I^46f_c=~cqlA+9f^Q^G5PB|R@|T3k9<2G!mbiQiLyEnL0!t~a?C=>0g!&zjBkT`0@fRx}`#s*s4}b8c%2ST@@! zD_g{HiARdBYRYRLLH9$m}c?d2A_%dzZhG8a<|n(`uc$0tywb$bIq4TaU}) z^GQ&Xmq?E6oKA8s!+5Z9OdWau^S?XoKI*00#mm$MW2dPP zWKif%FS@X*BbFIrH(mh#-iT-#pPpp{v_6^Qmx@h|$w{g%KGpd?xhUbwSiu@GN_(|J zsbFzmARx<(5vwOj%@YU*b8~wV+=sbt53qIqYz!pj=-vQRZ=q@8AGoM9!rpMmN|d%y z9%WJdY9%Za!^oKzwPa4~2*sk-Hg5);IkvxnMkohN(}l<90b2_d+!?PreqFs+{2X02 zo6?*WRs5WsGP%bF0fr(glSgHJ40y`2bdog%B}TEC^Gyf}8wQ{2stS0m2+nw-By3Q2 zhDseyV!lYztzjqnaKMii1&=SJM6^q6KeZ1Kxr^oxH*D@$N6^l%BfValW#w!LTQz%O z+WT5sbxD+pE46SsVOCg?zne-10T^aCGIC@ye&OZxIp^7O0tISkn?i{J`a*72o6dn`$el`SbC#-`rsGEP$K;P=EQlLeVtv_C z!?(vnxQ}B}+f!d#ap%ZgQB+?da369qw|Jt@l@p=!J+z%2CN<~FS}HeAH2g>k_NZg3 z!x$b`A1pqIP-sAY5DOp$kEZYu=YW_Rt%nLOm=FR8p^YU9WFn*W%8J%_yxp@xRjI3C#@Skie*B6;&;XfG7U}Tu$iEl;v|kV!g3LgNUdEJvW1va zD`d4Pt1XW+x@S0P!^$rQn~Ucb7tHPb(rH7so#E6{s*BM?Gtlhr`OryM>u@yP5hP<2 zZsiu~g{4VqGK3dOG}}c%GNVCNqH=cI?AxxdN@eWV>2w$WxioI^ke3@TR&8GH(8o^O zpsZ+?NOEiAz@JV@;k`#u)MsQ#jZ|LbH;|7E6DZrzyG^-MJubM2?Q3BnnhM%aIr8tD zs~V5Sawl;8gW4R16k@j6nKQxK5{O*=h}g(*hRnhJFp`PEK+0)x7{Wc{5UIXEFh!Ih zJfi*}r9EM1HleAh6z;StsEs(d5WjGhlrsqY*gAYjL*pF>pz!^n8{sC)7Sty^B_`{MZD3I6$0G?={ zu1nsel!67wSm4R2@zi-=hYmDV^2EP^2pQn++qF5-2%@F?;b&F2ek>A*T8}rW-|D=l ztv$!3r9ryjYHxA!zku4_L4Vr9!$i^16do)9#aC<_Mn#BZT_=dURhn^DGm|2^Y%eE7 z81!7PGq>b;+?s6VR*ThD7#+a$V_P|K-}2TPvl2CMT)FI>xME7k3kmOZdiib7RVXFL z7&viI5~@qv4%W7|ce3L&$60{~U1|5Jj%>)m14XLz+~(kRigauidKJ|4>}mV1+EI}@ zCVHAU4Z&g zl_T+^utdT(u8k`EqfGT{brGcDLKCG$A)cEU_d{r~I}8Vj;0^1|W$LOWXIr)J!z

ic*fs}J*ap1g`b%C^ayL&jUE+ih#KmvMyDmzJ4QZ&G+U+@7J&hlOH5Kp+sHj3~M$ z#Y93@6W6bGxWj%3gH=_XOUQI5tRC`%A>-ZusGZ>%jYQ zyk_!CWl`m=&X*<(2n7zRSMS#y;T`&$rf&C~4h)SB3^Paa^99>*vMR(O>)cF`4q`Vq zf4cr;dgF@r{V$KQOt*8Zg57D8DYkz@@MjCYA=B1{YM(5PPDO+5<{jwfL2Mg_dLTx5 z^7pRBag?=KO&FBYZjSSn|GIMcY<8A)8EcuantJHX?uU^1Bv=rG_zn2pBm{ zh}1|145n7*HvTfm3{;c>F2QagNWnhBR7b+f1HC_$u9>@hz>{(p|D~7$esMEC13l&`l| zJV~^J2QYT$gOL!?KIlnGhJx9CRac5r2#G*naQu*=e<*aU*MWChs}QG9Q*lNW`Q|Uq zzk*@m*aZRu(Vt8>5m3-Uo2Cfw)yxWepemHq8e-*v{hyneQ4@fzB?+taAf|}AxaFQy zy_^B&bXFuAE08!U?}FyrDI`R@=vj0$9%d0iUhcPjUQ5{R-ZDZiuB0Vf^KS?O5I;>%viaT#`g|T|b@oACU9BV)dlLBD(i>*N zIEZ3`3y$cgbxq6VnS3ht{%IgNJe88e6au*I%T`wKqsI@^i1uP-@f~ISV#LF1ynvV6e|06xMY~S92YWCFtC_8VS zFI&7(8^4`|WT0Xxq!92efqvjBM4EerQ=V&s|Cm=%em4<48)jl8OWs1eSdFnciiMG1 zv(Iu%uKyEgaU@T_&RH?Hg1(rskUA*zRE>001&dk%q=FJUG;JT5P}_x9E^dL-LSgmu zs`e_@j7Yzof7z-WgpLvhnU*FdvUHbY+i7pXAjn9CKI?S+Why*)mL z)<sox|Dlg|%Z>d%%> z1B5fH7Aemo>RoiZW4W4W&0oii=q6OY+H>YaP#~#~QJ`}b1R}gh-_F9cqMFSNBw7o4 z3kh+qQoy4Y!@y&A(}htSV;kOMFOss{gEmve znqbArs&OHHFuOd~^!WGr-nk5QtD@ZGv zhG%#oDijlj<^Ll4PT)GG5Dorvid|Bt2{emaa^lQj=6s7 zNEHGP!zk#~P*cX(8|awnCOnXOaW1ROsy@|e86%d3fyG@QM;u7Mj-|YhCVc?OjqFK$ zE;dazU=8K^%Tn5Qrh_&^-T&TQTl5o}g*~jrn({p}oYW6~!6E+lY4Q`Rl#Yc8U+V6R zwOx0sWBBmFq&L9rM!-TiI`5(*e_T{}M^*t(S-v3GRR8bjuVd8G$2#WDJT+o(QI~8W zi{0AP=`BNJ46&egLk9Naa`siQ>a}$qgl9Po+vITeS5JnJ8>tbUuHRTzPVx7i-<<(& ziNT|O6VPIa$ZmiF)?A~hi0`MKU81oUIDUnBteR$tDE-(&ckj;kZt_;EKRLY?^4Z?8 zEl6F2WpP1JCbd7n^^;SI{v^gGuY4y;K8KblB~HnbU5SYrJi~6TfT!G45ib>s0$kwR zS4hSS;;x@tPg9QmsHUsHSQ+O%oH@Wx#Jb_AT$lPy@z3NygPL{Ll{k@Wf75lNkYGUr z^o#dP^c!f64uaB4s6CxT#PSGziF34MT2WKeu#Ax}wyZS3r>?=f({UVBoov^Nb;{bp zITT~%(nr1duxCrDl8c;}mTtm4f|@! zuFOZoc^H+IVg@lFGFfgt4Wn*rv!Mu%t^4+p3$~J#LDd*HEKMe}etRcsXM^cE9Zul2 ztHV<3=guPd%J`j9^`8zYcra)aP=Oe(Idsux`7v|1=;gr#gkDeBLJ;J6Q9WJ%yHIz| zWIDIk8wkq91ITmSKCpzcXcd-*az_$-6>`=OVIOQ-JYb`Quab$r~PGHc6~OrgbY zR?f-^SFX%`@^18{*6`40!f9dWH|b|9SgJ%B2!sCBlY2R{#YzSXE7)aI7u;4hw$g2} z>fxcoALX`^U&FG}!qcH(Ri1j(@nZ6#!O-S(>K;P2_SZHktA6$7Bp{5Z>-EqZgfe=o zyzzEuRZGLl9HKGx8_G@SDg#{_unKZj`7(j@F-Z}?wvGB{t;~Re3%MM zrZ=IO1brAz?z3Ex(tM1Q7Q^`m-c2fIiuiRx)bXK8DI8VX>W@US1ecpRdemMx@X_EV za9&)uSgR9iz?uKXzaq))$>SYG5<>)^T8L!93kFw71IIw{9s*I&tdKO`f3>UsSM&iA zVyF3YjK%&okE}8lrJ)nAjIpU_lR`mp(fx#$|NV>Yi{Q}R*K=Dz^u6Ez%`UrTVFY=I zZw`|d$==&uE~-sRdwpXh5D2q8G2>^Ha7GEdN#v z^wyMg$V=@ba*?|6GCPLn$`XWDF%H5a%X3&ckg#gxzaSmcgaAAAZu0EM`EMLN4dOrv zQgohBt2jQ%*Nj@k)G|);347WlOI^40lI*k9FNU?UCub7!o{(QO5KpPqw+vk|hqSvN zZBD%GI}e!hCiYw4Ve2eJg|k`LIuvnsPFb8S;G8siqpzDCAi1W_Q!mtGLU?6{x{AXZ z)hO2(Lmy#b9>fIX_)*Mz$kSVodOex~@!lB_c+oUI%?+&g#5n7-8*WWuN>35Piw%XC zsIGs;M$Jl_*63u*h(|sPQTB|5$ZH|9lC6HKvx zKvNui97st`iX4`)TCj-!S?J3ZHyax%A^8{7E!6ZP+J#!Xb=SyJ^D0-Cey!{ZLI=eGG?U#0^n# zWlt$8$s`KN?5XMsgf&R3Y3lDtL2`v=x9B=@%uvZqR1&IEG;;iO8T8;cke|#Fs94I^ z?Hwu-n+XsFt@SsmD^4H@a3N59^1uid5W`?}AP2yKMB=$pjkV(e=U~+@tNIGXVTe^J z5qSoTZq$iqST=pTw0o<$}+N$7#1*aUVWGxO@RPqYGI0}fSO*yVK+qYTp!g~AEn*Jr_*zjrz z&C=0G|Kp9#HM~x{LPEP&j-ReuN=kwj^<0U+j62IJ)kcku^IW=eiUV%vD*rG89hLDbO=l_tsgX&}L8_|`GZ48!ntZy6G&<1DTL_~Mev z5Tn;~KTcJz)#s75LyEfSgdj3q?poM=&qBxmM zX`R+adjnsSB~YZ!s7ITM(wSk3$L2()tO23tMSo(tUA0a4E5+a0NbkU!@{#HKP{2uE zI?=hE&4^174e))5Y!zq7%@zntC(|FQZum{uFaXYV)83$BB@Mpy9D}lmuv)=9$DwAW z*O$-sOm||r?4wj${)$Zdbu~g(gKCw(hDlnTPuZSbxP#M*dO=$G8_0&fCOPiMB35o% zczfG=LwZ_xTgt7Vk(@uB!xc1B&D4e0H-2t+#+6ae` zph@Ibua^q_^FUQC40et@K-i$u7M8c^ey0;*h?Dd;rm*dqIhBJ0@1U6S9iQ@?22fX; ze6gLa2rx43lhR4Oz|R>l%>rjRw^oOMCdMBELPL|6pUEM0Hj~S`3g>SeP!(t6a1Azh zOBtZaWm3=fe){O?!8n#+h&46Os`-J`@}=C;(lOsA)-1iVj`&AvECdUv3?PSWo~!44 z>{VJk-SH#tg}(j@b;vlE6vYJ~3h=A!8KsxughxXOw!83wbZ?*nI$VEe#6<)&B*#Dt z??hW{+#lZ7jZt&pGX*l!90!QUm(&&j^@2C;kJx=+zv5OBaU-^eH4HM}0WKOU6F4Lz zamtS6~J`e*Y3`kL^OM#}zT9h2*tR_p8-jfkwcSuFjy9rE}o8Cg7}i z>|9w~x&aTKP>jr?FJvD0FHc)!Lk)>jV-pB6gNajMysrMhur~GKHI~z4ahM;*eUje_ zi~%1_@GTUicp#g4mBQw`x#5>{bML3%+QSnzR={q%k~lDnWZym1Hm3AhNVG#BILj|5HWus z!JUvT(%9!;88VO;t*|rjK$N^~U+@P2#aOpY2g`cq{5#X~xeW5M+-07Li;ubm(}dTh z0&q6%-8-rveMdpqD|>U$Qfz9TwEi&H^{&5iKPa5M{Zk4QS`U@DfkIy^@?Op8*HQ3z zBz0+6qNR%gi46x*n1W)GKnO};a0vYc_q>cT*m?Yu0^6KgQ;;!L$w+u4w;4PsEZbl! zj3S&WU-Wdavoav)h!EFE0~b-oEr|jv`{P@NelwHUegX+A zHFHYxT2)L7b~*&_TI>)2(;EoNDTw1-zcB1jqo4gt!po4uf}>3O z&HMJTN7N34SX4{33yY*U9ifrcDqXjJ?|Y%oGc81Tl~&9^a8q8HdovB0{3nV-T^jR> zUW)B{)#_ue35n`e9l{`0=K)_YLMAK*RQ4o!Z^aJ$#~Pk<$%oc&>x*Q{g8;Zg$4pV= zQ)>)WcghEM{bPG~S|_x=H_$O76QVt60m0r0&k|We;|o5K9abq&bcdh95GF}%fP{mX zI-1Rr(gKkXmZY7gdfX6vj*Jg|(R%i{Ki{qCVEm)uv|GQQe>P>Ipn%R<#-ZIjsoLS` zu^n8e)7y=C`)60ibuAf$lPg8-eIh>NQqMQlefEiJ^Iy84M$7e`Prsl=$>m9QY2=>e zjqAT$s^=GYW+dXtDBV9ERvQnO1IqZ&rdzLyM%2lx=0qEvOOp$|D^H41J1aCCV@R84 zIH&M%Feq~*AFv}ijMM)oLuS9^VUHAFxV7#}Z)ss1NUR6Uv``UX5bUCCUaerq_WF6O#|xk0O9MI-sFQu10Gn1$s)I zx++Lm^3-`7t_`K_T(ngPAaq7H0jetE`^{sV7>@!A2P$ zJG?$la1rn#@Xka7RqI~zVYU2mdwd7d=D>`~p&577?fCmN%XclhOwplBSv?ss3AytF zd8QI6){3jLBXeh%|EAFD$=A_pxlK<8s|=ow=Yrm#Gq25Q|KFRN_ZcrQek(7TV_pdv zEe$PUc3u-c(+*A=9_ zzk1eNYOEmZfy6%Eo+4Bvn8GJ#4Yw(v%0|iWg+@`PasE|A3r*sr)7S;o(r47qI`)3Y zRr$d3$gQrSpF8Sjn-gp0vHwL-cF&9STreLR_oI!)CChb&imgsdL#^FWE92a&VwqF! zD`iLtif-e(rM=(2L0hDdz+Xz>#ktMppZqH&0!t{vutB1=nvA4!mUgQnPY!v9wDn9b zW^$`}1!;Fa`cswvwL`y|NkKrI3HfJZ@4ewj>9mU2X`d{?sdjI4GYz8*A4{==G zI-^9Q$qgPm678iUmwwBn7)9K|{+aavuD0xmpF zN(%#4glLI!bxu@Nase(j5mG7ZQ}LLnvf`4)^Yj0I{?A(ZY@)JC;3Nm2Y-p&@nOQyO zfUEC?jE!=GL$gnAbvWqK$tt93Y&?QzTJJ_w{ntPK)`DOX{f79y-cwH00fAMyCp9N8y_1782sOt|MLxY z>Hp(@&&qwabaAUHTIB04THB_%C0C%>q)w4%DMv8BEJTX*lk@YvYI^xWdg z&!4|GclM7?PtUJz@1J0=KmJlTvzNMY2)xexuR?<$`aeI)0IiPjyZ_hX|IhTr(+{BU zThvt*6NSqzL*R&RL*jjJ_TN?rot>TC`_N=qCkzJb0MYu!oSxpn{Xh@=`)Ke|4DUQV zV5uH$k{(1vSUqhog&QAV{vSTl@1F(+w@yxu)|ha4o(!hG(3ir-=JS(oGcYhz)pT#I zs~5!3#*GvhexL`U%*)Fw5HQ``>wMY%WzomNl-G5!*ZmB786xQ08OJ zTy0W+qH$(3N?H6=mv5(fl_fE4vrDitTTFF$8Mkew#!USS_LRxz5itxkNk7kCJk+Vy zw-8*@sv1(1{e$=+HOY754pUF_o7K<)nc+~=N>HX@@{qmqyTL@Qxd$t@OT$$1WdqlW~YYeEguslt4;#J!jb;&zpW@W`vE~qmnz@Q*yJ}BMkX;uSYiubL2SiueO%bq znN~MfdNyxdb_RT!Q_7CePlUgl+`jLgv3K66ERJR9I(mBB1=9L8et7%K_j*kN@!6&8 z>7IXk+X*u9{VL7kav`eC3hE(@JcQ2{#XEnEM_jeg%S-qQB4ACt??h9W%l9{wka~m5 zy^SPG+N{lao&2}zeAyfp7QDXMu=`{&v?IvM0kY5pt6--Al|(oUEZrE#tO3F^B_n&Tyr1kjm)(<|9p!Gv#*yDk@N88!(K8Y>{R>0%V{F?p=mGm5IZbPum3Pf z)_qodFB!-KYNilgiafNd!=O=Oc76OPJFi) z;<06@Yo#+q8B}!(xV546L`8JvtcwW{fxxPNenuQc91KIe3=@@8Ns*)hmPVoOg)E?9 z*k~7~pbbxYZX_D$#k3Z8Pz;?QupNIAo;`wmo6~1&(;F4CBbUDUkrQXvK`Zi{fpi$D z&5)XF*1<|9wlZin<+^k-EC?0h!TLc zP`9Wxs5sHtf-p^-z`k!n7~tUU0#)#c@dDbtlTIgIL9Fa9po;xV?5CxfrLL3@6m9|v z98Jv>OVkprle$q&lN>x%i*FjIt*Vxm`H^-?wK=O?47DO^;X@GjhFTfP>{miTP0T7p zSp_$P6M4x*go?q_beFc)b+S&!+|xP?(wKSsHaf|6YBAqg(>p3*Pt|5LzuGmE^-7+N zPb`a-R6nW`1zLY*_~Otz`gj{Tck&o0FJ&DPrA;c;2*?78VWg;%*oJ>MB^CIAQVfUB zQR;~Sq(s4yz5Cau!jokyR-qXJ>E6U&@hg0Ac$|6#ak1m$=E!dikbQit?;a#3{Usgm zzilR=Yg_|Y_r9mYx)c}}n6ksvjDaWC!56pP$zOv#<7`$%D~Mlg>D=%Y%TA+AM`Q2u zihD*Yx+hRNTFskeFYX-Ank`1`*%7z+D#|6^1E)sEZ|kM4+%J31eVIJ3Ry6E#RpfNe znR{qIe_AJ8@<|>wWeKsZKxE)b5L*~uo|D~KeO`{${@moc@JpPFtKyo;D9mv_|H{*} zwK>EFa(4E2^Qbg4zzSZZiEt2Vc=EWC3YL~W?0$XXapO+j zrQKRK#@r~km_Oy{ z1{EjpVig=zp54=4*24 zFtHd(CoZyb-9C@n{xCamepg^27W$PHkJrq&t@p!bcc3y>Le?4Gu5{Wa!*8a!Xd6V= z84yQfV+~Fitz)h4n=~{n`br$6B@qx`^BmuJG*nobM%nuJyieI+-5o5qc-T*Ec^08}f1S^MA!x#Q%*=UAmGi!_K*S>bPSv+HUeq z8}84^qV=u}yp7QZ3-6gT%@D6L&+zQ$4B*-w90I-x5I9SFaC$c^4m`+tdpvv6rROmD z+3;?}pNgHTnMrCa$<5-C)S$o1bp|hZ=Q-qAk5H4HHquo79*SG~*|h~Kx7=FL)fEa#laG9|XbtNDZ}Q3dDGb&e6+m!|nmsmurx7B6 zjb>Qbn*wf{#iG7*Z#6F6m0>Qsm3ceZ{Wh6x7>V8JXZz2u8~eBbZ|$v^xKu5Dx}*;n zlaM9b8ML7=ec+JPCvY!^LAE@dQX%>XT=PCdIoU%4yJQ41 z1u{yBXEFsgWsZzm>g4JZ0;NSr!C;yZwIDW8h-eCT8$)kzgKB%_V2LeSI`Gk!Z?ncOM3-oeiQgxgPWK6e9`xgl4$UlK+TZNv z5LR+8(j-%S09?UjgBy3~$3sznc$zvgJ=sG{T_GqvK~;=!r=4WGkZHelINjl&CCANwkl^ia>=@Dm z7;;sP(yUI?lm?T6MEoKd^3_*VOCFxG`eeli6>TxZ#T&ahlS>S_CrWDp^=@-mEA6y> zXQz)}@&z24GLlWpoPkz_Oh`RBKU6}}YfNrBZP=rD$k(L{<~ywtbQy~)jO($CrQGA~ zrXCtJXHLvt>I#FkdKXXbVDzi;aW+;3p_B5zud|8pt6o@)@JB8g-FEGlTB)uG1Y`du zv1t3JQPsgQLK(ch4TAij?e~u2hrWWw84KM@*k+~dIY@Q~--3;UnUd#m;KAt_7&C$S zQ(rOeYol|15fO~BxA!{tY7K?0;I#X8suAPg#rrs-*H8e$7nDQfdxfHWw z*l!Z)KR6z^#S)NJ(q@Zj&2cTX`%4Ci=ZnSg{1m+sTAx#XsW=Q)HN~Y-qZvjjCm*?z z#gSchA&_N!x#;V{Ao6b&9u7^{v_ER3NA3W#7p_dS+Jk#pc(`XUTMvv8n;AujML=1O zf-+Q6@XX6MEsXK!4-*oXTX&Q13uAw3bn=}*w*R$8ACj`#PJ%fA4#np_?XgC#E zH0OG631g?(ooJ*pZ)?EUG{XmRe$knXt$sh3dXWFF3aSd#7!1O* zYt}&I!!vrDf)$J=@^QJqSpJz+@Owm07simDn%mEf*QhD~G;hNyrq=dF z_#Gh(ilI&|G>lx zRq!cilBC{oktSwxS}vqS;ug!Bn5-q5!oUJFa!HV)3^WmSm|v5+Q?97H()NPB3|a8g zd6<+b^YmM~;Ri*d3ED45kWFJsV&w&=Az3+-AN%jblh{dWY1~RX+`8cINs|h19mF-Z z-??g*KG!U&ZIK2{qar5qOg5=(KgR}3FrD{hfm zD6u`IacUVuL}z}}3FqJcjv>0slJxI0W85N-%2(N7TiNm7+9Eno038x#-h=OIX(fEU zNQ?O8PAa)DETi$)eIsn$0Crd53tQ`gMFo9)=(}heE&ZUb%uDzV2S5Obsp7#XYOe4m z`>0FtC7-?TR{9^Sm3q!K&z&V|AVXTdT)c@ybcE_Zo3gL+j?^fLT|c4TV<#mF%3N&4 z!mxjtrY(l@W$Xndrw`G-^gUKY3yyn3eh=dY9F&m?0*Fq97|7bN;hS@=k>X%D!(hPF zLMsl>9Gw;)19qs1g!e1xif^mMQF2r9&-i%~KV}8}lc0EiyWW(wCQCsoor=ZM=`U}w zJ_Y7c9s8D3*NbW^IR$Gk1cbs(_wdYahWvZ&bE3<~iOEY9?m9J~Jd`jrhSfMfrK^Hm zVwXT`ckekzVDDe>cj_|*sN~tvzuKhoZ>_;UVG^LG8U-SeC3@*-a?yZPoRLU5I2XHl zHs^swyis;a;SNCt;~0oi_4h)WBMwO<6yh4rhPZcfUxN{|pe8}O655<`iX>9DdaTBM zX&44`c7v*`ZxiKH$6i6pEJx4rq@#F-v`K^ZZPWUgNrNQQZ}2~%U1L~Z+GLVAk2PF5 zD0$eqKgLgb(uekt;n(IoKfNzxjejzemD4pO&ttn;{%gF(N>w67kBPzNRp5lZQpj8- zd#w<0Ig3y}xX!4TnN{u4Kp|fvzE59#%V54>x%X?0&7gWS1#;3@x+a=0Ze9!`Bo|+i zPb&F#u_z{oZNT#w!>rOS3KE^GfLERVDmpiTRi0E5jt!Cp13^amTwd>GhuBa=LJr#uO0Y>1z!rNL zT`ND@%QK&aGH!6%0*j1GAB6!$)L#W2K%;mypD7Lxh(S_|YUwerkh}^Sus8pmE#o*x zYQ1S%(PDRa!8+(NwyuDLp|j{m!Q;{13bA9&sgxlf)0WUMw~!?=J>tz-dN(M|{n17d z+IqiM70~Txc+%KvHL)?_{ZzR%b(v!6G2(Z@F`J?Dv!XHo_BW{~rbu&=_c2J&vV8n>qu=pn_VH@P_a7+rWUxTx5&fl6<>~GY)oF0zCdS)j^e={oZ4`5iTE9tMkZ;eH?NsHajbJ%k^q};F^;D-t(q(mbFQ})6gBhb5 zaCoHcy&M`aLMz%*nA1$tL$Nkd9YpCJ@kqa?g>v5WFV zU-XMBFuBeA734*mMOaA;)^!c_PLfkMEdxo>GA+JI7peaA$*nY9f8G$dkhOX&&zGm- zA6^`yYztgc8%h}DsA>YYfo$b_Rg2V8s+cg@;1%a8r^Qp`t4%792oVzBop*;eK^&gR zL+fBLKePA(Dz!f1vYM+bPcmn(5KYYc+cugK^!|^JF#!{an~%4ac(p zyM83RtV%)=>O0Gbh>Xm@{y!B#S&Wy{y_VR^ms_9bS6@46c9l0CYQSpy=gT%`BsZzj6+*%FlO92AUvJXj7g%ysE# zSW2mN5(_g|{_yW`hP>GwT)xG~(%6bRDfxu-I9`w>DjAW&<9y)n0w=dLMZv_GW?fFL;R^*nh!HM(21BJ=WIm=aaACkRZh7*Kl!Dzq%ZPLAKvK z4mo=ME9h9Gh!7txYZ*Wn+0b+(AK~t-Zz;+Rs3G=eAi7w(bv$1GcDG^2>N6*kZJB`y zhk=ZUBvgSV3vUWC33BKZviWO=m*em!jNs!X(*9%;ts6V#W+FKl8v`LGlIE7KO$Ego zP8|_27!Vc5FA5wmS&(+qwl49s4)*wU*3u$_EA}OAD4I&hWEX--j4T+25~)~#9iJ)b zNXgS%Z8mEFQClTBNvuGYU;;nm;oOmI-aU;S?C!65lH~0>L-mOe;ju$1Q9y{n&&mn6 zh`X;KL-8i0qjwqJ)I7qY+|6Y3qD=^pYAe+W_`ir$1JS1+8b;Upz3+An<_@{UAX#cf zD&rFIw#w;W8m3hD*$|$L_N-3A?%0i0gGFZF3JoQN(|p7%z^vuIE6V&{TTZwTmEn0o z+YG(+tr8$0RuME;%R072;@de-HcAWZ%k8d({rlBKe)O;W%DWyz9K^gA!n2|JL?P*? zNQxW;SE93UK{xNv(emv3l*r+TwnD11dXBelXK^=pv|ygVHz33G*DVw;ez7;WDxK^ZJ zL1K*fgtL#v6O)3oi+$ZQrM~NsgWSmb*)`-ps6U_Opw})Et>|jk2}{5NERxY{7IoE;)e?u8Qeb zkTDDT>g1E2HG-VA(0)Ul_r6crdV*w@Y+7PWx%Kga2%#Ix%nY~eAy!AC80}Tcm_=bZ zZN}`)?8#NGdclaQ%+=>)(gX!P)}4dS8ri8Tk8I^?~YV05$&X+tz5de?` z|AV;vPy|ljfD+DYr6({$F}O|xuo{?D1^7qYVAfynf=7`UM+w1X5a$uk4MvEsFGbTC zBOdjo7_7mAoQCNZ@>NIwBSE*PUZp3le}|-$7{>*%Nm-*6c4Bns$tcwV-IZq$^uK~! zWF!eQ2{L~G$WxxKpaErN{U~lpj$d&l$lj{Pfh0oNba0&w8B)F^F!+0=IDF0i>HlvK zn#^pceuO(`;xz_OnvmLVr`pV$XcJ9liUp-^X_H0f6hhn zk*v-OneFQHF}Ai(vO7l!U4kCXS7c=fxCGh&FwIeVf%fKKvh7NQq~27Ou_lptf+CuOAPS9nRA zUA-J*FXbMUF0q=%V5%S^C6GxDxlm2v+sI6Mhk{fEq)$wa zk>>4w+11lom*3Lll*g~fXi2Kc5F!9`K>OQkZ!jPbE>gUU_#|F6_Z4Ks=t4NDWt>~_ zMhuY*j-Y$EIlEV4Za;lCNK8$DE;6e@M`%0gs;Y2DSsqq1qb7lJt;?%&)Cf zx6gDPn76c0OH;Q~g~sMG!Xz0zs4p36;`WgiGj4>IRdr1`0!Lo_E9eWezIcZ;*Sm7v zSem9qE5FDy{)QCZmJ8aHOM>in9xvL#TGr-_?j>b`LQLVHv#q_;bN*eh=Q_r3t3(Ti3y>wl;mSzy3{>`#+V1+bGNJKxQQz`niF{^7qQNuU8neJ8 zxpek+k%Y+VL#`t^B$B_1h-_VI++HN6Zukl+_a*TOc=O{7(iJtYpaf=p?@km`Je&o> z0Pf27_v*a)jb=E$Xv4Wry1zLc2PK*ffDu*g{R|1zC$2?n4t-EjFe~ z0y| z$KNBPjQDW}O%q-~1c^BYr_gAMmGJ;x`a{gZxs9WW#d;^;CwsA7)aspV<#j<#s%zXe z6}GK?T)Y)zVU!J2^$n0%j*$~;%2L8Js#@~_z2;< zH~8c=gWxT}>)mgLf5eR?AQ0v&Zr5@*xQ^K1QDR1wVA+&+5WRvXnbxFhe$T47!AFGz zBXA)jPG==bhspzF|A?B*$ie%Y=R@H>d&!NV=2_y@p-|}$e~9X`rycV@6=6}*OO8b| zYu?P$dZ>zsS_qzapD=sQ^LA7VXAu>%&V9{-8kci3Ul=>YCY(Pb`$1L_2&ZWEsIpmD;U8h!Slp_86m5mUF$lBI$uMeKDmc`N zfF!yH%PbWQxm5M2u+MSOL}O%Bh*m~$ok|d>?hhA=yek$t{wrt~jEkGHxl24fNaGZP zVI^H@P0SdMWVHXOh&BR;Z80MpDj!W4=V|smr8rew4Z9?wr@$q!db-)lc5b?i)IDS1 zp7VhJdQh$XoUEY-hz1R%GB)_orsLo2=H6K;aP(}~8nS&WE+ECHB)Bm{ZdFs9jbWg; zv033%Mpr7WP|$%xeA;#Ya)P3pyi`lf&(cZLYbqsk(r@4&A_^6G9+4PlNz_cktv(q< zfpDXp_=6|Plku_9U2G(cRV`6P#Y)lKlM~^H0j7l{acFX{pab?b%e3t|bbI+}x?h4e zr+T~1y0noA^9rGYjJI}CviMqBd3AR8qq<`@z(U&C)nuE!3mf-T?s>O0lkql%^)2azYN`&25q|Q7`Rn{aaU@ zN1etb1%5(K{H>ylSNSK1E?%lI=(r4C1P-#Ui40A-BYy>*up;m#mCPZkL^S~FW~j=H z&9Hz3+V#z5Vmj{haN@;yT#Ed6N7a1+6+Zxx-|}1`P_c!pEJE*ch-s0)-!fTA z96D@h(jWK;oM9?%uJpUeG=FuikN{l&ACdjU#&)luTeh|LP^39T4Vr^&3`BA1aO28D zx2&PD+)*^|TIr3n>RY}6m~HJ^^w*;^9@iygZ2{dDc3gxRaRd@eQ-5nKnL}NhnODWKsHfMfE4H8Djk~7?X%6-W@NQ(T@)aR&GGY?1T z9RP0cmv4+#xvs&u?*`Oqc9_Z4K3Do94SKba2w-$|@a}eCRoOxM!rtO}QjvjY=WB4@ zW%oKZ+L{oroHs>dV#%v~8$-rXL9;eKw)AE*+k6GxvHP?HNUu%t3|dZ`n%K}K<_A<3 zFblE9EJmqLX2);GHfOYn&XcO8EFQhXHA;?jh&of(gZG;K3K5GujQYb#Hl3J!wKE*X zsO&V_Zp^Q!sU`;LEci6adZZlXAOzcbnu4^pc#gUBhUDB7=EVWwxd<}4U#@=`m2;)9 z<5y{dfrrquiliM)4_Kiff*Rbd18*_hty4o<0Ap5AtK<`hF#->=xF?N{K-nlfNgaXc z(URKVQQa6^e3UXXQ4~^BDLP6PVt9UHN_LLKR}hTDti!WtZ9tGqhy;a#r>+kSWPU?O zC{AbHM~hX(&5fl1i3kx+w*PTod4x49eBhOpmVa>aeXP4vJ4?Yu6Y--64>QOBsn5|= z`c1!M*;N{ZJP{2**?v#3zVgv11F{FHf3P;Jl$bccNyL$x1_b#>3Prj6f<3Rbc;S~t zsN-N&#O7u3C+06lej>qGaMn(w!b#Fc$I`Ny=+{$JhZMo4;es zag^rX2_&JU#X`Z%+6p%@L#US!)e&TiiIm%m z-KhSk+iBIqRi5PYYj5w^8<4Wpv<*17R!&Q814C z986D|Fl~C}oMpnEoe3k&oi+Ut>htiNGle)LoLlVVsL*n-Q?PL#MD0Tj4Ff*#dU&U1 z<{3d2GeXJ7X!omNVyXXJ~5Aszeebc75&*Cdl5U{(#mJ86ZU*}x#zR!vv@aI zJ=g+!p5nWlcv$vJEzL%h%woS4!wW60L3Bs(pmu$8S&5@61*eN?X_R-#rnhKU70^h@ z;WnTn;6`8+M@syA*Z`461uj6PYRnv-at4#Z*t}ZN!n0JC)GP=CK@2AWUo=e&5ORPM zhh2_mF%85>V;sS#De2Im1pbK`9S|c)APzR((Gqaj#F%e|s!-vB@02Jh$JJHafIfsx zgPNqm%t=H_TZZ_bIiT!vIT3Q-_Wx$Qf}UlL34@T&=dGPUPWf+=He5)`l?vh|Q^9q%9TI%ab zolLJR4?ZI zvmi{!;akw3J0A&0qRg7VtkQL(D_y&((jAn|X!f6<7=~Q|>ZRi&ItwT7)44<@TGS`Ges}tfvWDf>fA^*dsKk-~zCj zROC=xeyhyTp?Cu*!c@wO_J9Fl7+wkY$ze7EGzr5qUe3%QQ8voAh~mBK`=fB#*<1#>Rf|@9be1EEaClm&o)^|i0|huMmK?N{QIE_@av+GZ z{t8VNURXQIu5dOOQt0v>DiNleN$SzO0C$Ks2W}ZB79?@ZG@;Oi=?&W>P(P& z1EI6g8*RP$Dv5hJ!8vOta0yh1gh`e^vlSXvpaO_MnXNYuGON}wzP3?Cd7qum z{C0$A2~@Va*{Eh+w%bgzoFF5C>qIqDtjqIddMt;+RtvFKm+>GpuaSi;r5K$%B2!(7 zEWCmS4xspnl#n}W1OX%aMq7rp%!%F-XvJgHCGf33li`q#&fXsSj>_M`I-nUV0!@r3 z03Ps}4(>ZyhxaO6Km2tl5rVYaD@aNp3X7XLGY^i(zWE>^l>!))837UFwo0vD7y0>? zpBC;1W(u6>;4<+!y^7rOQMH>+PbT2E8Kg#q4eLCL2u?tbrn0C+wSl0bm^CF+FuVcM zgUp5_?zj&BhhK@_WnfP&c8;4*PiwxVC_!Fr^;{5JLd+@gRBV_84_h=)A-IRXl2%TR zGNZa7u0c;*3ivs<$jlCTl9rQBVA~*9x7X2fpi3a~r|M?wSXy7m&%|TJF)eseGN!hu5_vmk7Y_&$Npir`79}_5n$5t(;w3AHn+`uxaL93F90x;|w_|>j zNEWb6JMg;}xY6{fVQO%iCZZQHtD%S@1=PQ1;G7*tXJE{H%=#x z56B-X87~!XneT|pE9B#lA}Gjt`p`)e&iy&w%bJ)bz>u1@V17L+VpI`X?f#_~;wZ0J zWcsA~&^wNc6U@b}IF$vKoZuwgLoX9bRl?hl$@=#!D(ON__F(K!cM|azKpTXeTm@Hz zyP-oxnd0g~X>9NQJ{dVm?G;49ns@^(nHj3r?JfbzH6Sf#toUO<|DOoa8WNsOR`6mL4yjRv*25B>R7WD>rD=OC#^HJ+p&Ua>Za>@ zgS)Oh^fGI9U}h4T28&=}boplLxyvhwN&3qm8NY0r3$crSsUPe)!cAWwW8q`GAo=|v zRQ79K?AKfKOCJAm98TBKyve|6Q{$bA&-J5l@V%SoH;CP6$;oP#KO=*}Bc&uMXf7{t z0Ze!-D*nhEHnK#rCm8>ohL~wcm7;)Sqsxnue2)U{CPEq0`}zxjPp+XDU2U$HAsGuq z6{!OLFzuDmgO8}6->(^~ZHAMW`>rx6mZ{}n&D4nFNjfw@Y&$v%`|{^0$ngR<|7do? zns`>%4W;|b$$o&=TC$V*J4UYvj;qH%o6$MqYBITops~U9HWEm@` z$;wWl(WcI%0ei_j+vcA{gc6t>dkZCPoJSmQX@8r|BstUHS>k)Lk>X10cTm+>z#9EP zL1?8MDM_N@6jt z5olnusZTBY$DiS+t6o7Rbo$Q*z?l#W1OHQOXKeQn9V{!o*?^vlb*Dzkze)Jo4Wi5d zc*@`<(Lqf=pEvv{)2Lzk%RfUX!GhqAX$)%=wdKpDe9Nj{4Tz|+X{OaXqs)rOFCC=f z-xU7(Y=&xkT-T7_v^+&HD{A%^Zv*;KUV-Fxdx=t|K;-Q!vCXq$Fp*aA5^#`Qf#I@PYt2bVlz{44@rckwi15Q~_Hg8hivGRCS+(G0A4w zKXr7w;}>_k1bmReD`=aUn=s~sjwoPHiz5Cm0KJb?z$8^VnV!=saq_Zz8z`hXH+ZlUf-xjnQmXRpq&39->%eH zV*B9$4C(Lk(|ph2Ycd}rxxSb-XZPtgD?L9>IJQ4}G<*!4(v0do>2JVas5kj0lp1@x z@iTRMHtV;Y^g}Iic)GqhF&cspr{?8C0watWn<8p6+d6c608$MEe2Gu4UhT#anky?n zCX0bbkk!o+K2ZAZ74)EKPw1n^6AFELmxq$iyxr;bZD8V?Ky zB4NCT7z4Pgn3^LahOKj9X>bugGK2)rr+m<>9X^e=jIr9^FS;*q7E9QzT(130oj9=o zEmihj995=08^ZcPrmcMXM-MBkPt8B|O-+CO8gM>)&Saa;h=ktXmF;HEEC14Ps$O>= z^ErjJe{gb|+xTynvtf!N>?9x80Z8#XN4uzK{F#4~5c_6lC!kXFH+`U-wN@I|SSyt3 z?-yX>b_5>{_i!lr&Nb>gUhHuw&b}C?&eso8OoAF;0Z z>$}5qckT5vr(m$}y1@YK4EoXyJiJpY7g2y`G=iEmeHct@N7;;O;f*Hy>c2aTiZxJ4 z29-$>YaCP-JMA?>f%NmZ-K4T}N!Zgb7k&JJeCU$(4O0osR8%anib*)sv{J;q&UTu^ z|8|;-_|~Ev$pLM?cu=TC8`&XC_`}eteuI7|&vngKcIxN^OQf$i)~~Uu(zfw$oADi8 zUQe~pAgV~oa@o7Bj>Qz-v-_8U0?WDd8MQIU3mM$<6q%M#(N%XuP+rmWmBy@2qT(lQ zW~BUs<|I}`1iRx5wal21^lr_v(0ezHNkwmdrmTpx1{xP9m@_9c(4Vlz$>VPLE<^T%@i{^P}&eDX%GX=j6ia_98Tk<4>>Uxy4Nqz)hFR4{#%px z*J)D{zS4{e?Cm<-!{6D;^28|yOOe34xd-iEXF8!)wM3%n(^{74iI*V+U*a593g#rZ zS}SF=yM<0!4jkV4{$(_>oOM2U`mN+xbLqyeIz+kGtc;KL8+U}oQ_wy8s~6=`N}Z4s zsFF=iI({F-L0PCLf`yMVlTS$*7$hknov}V5uRv_zgah=(zaq3=5JL?mkee*gvaH1% zSX}{-P_j;beFcrsJ3S{HFA|%l;P$aUZ&i>U`F?3e?^>HJjo{dL-uHv)4%R;pmJxor z-t9Gr`x7FQ6uAxkuqEZ`Lsa)u-(HF-Tzy^7bC}4{hva#rt=Wt-g-Yf#os6*Mf*n(_ z-Dq!rr?n>r9ZIR0-`VtORpl*g?IqC_b9&*vI3;BPi;qte7Tc%RQOp}Y!Q zc;_wK^l z*>qL9WLiXuiY4iOD*HS9OGtOUqjkTUozvYcbug5E@$)emBhm!|aZHNxzueH*~|k{YW1iCu?q#OHTqXPAE|BL$shs|UsJ z27O6^aj&2owtn0%y>o-l@wH{#PL1y^Cf-(5z~`^0)(IrU4v8i;i6bDJcl%4c8=;+M zDAq}!L2&ZCMZ*mW^y)3_`;p&DgdcCZ)k!s&-$q=Lo<1qp+xcu5^Fg*)@9i-oBL%%m z)^qfLIu~b9fxw;0Mft`D-C}tbrRpW6GlasnP~WWCQc6Xq;S}X}OC@xKK=>bj1@VCH z3-sFObzL@@;~?FJ^ds%JpOYJSB5A)oZ`Af>Radx@b)QZ0yIMiREJfh-*5>M|hCl3m z;_lsm>Ar$sf?sfBX!@43Ha;7?8bO(7R;IfsZqO z(UqiSxq?kAI4FR|?7qGDH=DHnU&?uVtH^q-OSjQrx#HKB@@W3WQXN{jg35Pc@~K`Y zPeKDJ|K*HV7=lvz2*E2(g?fzEHJ{A~C<`}>^J|E z$!=p(rV;HW1&ZgKM*2A@Mbn7IhV>rL%&@j{+zNzt5iAe>78SQzZL(Ev=_-&orD3ht z5tp=TAuM|!>g$Bc2fTK@9hbEW@m@G@2q8e;x5}TzZ~PN zwx_zOYXE{9V`9x{S^%>5v^;hw3B4F@+0mHgnira)d!~IhU_VX^4nAU_?X5Vp@04Z& zA4xd@?lnRW;LgQ@-zK>u1=FaQsDJ&rc?m`ah%c=XwmalQMnU#}d@C7EefA4PEkUyp z6KUzFYT6AZrwMH8CFuCN@6>OK30JCktKTJdH_^an31l=>oQPYh*p4YJlHkipIs5;X zCJj*eD;|7&>?WanVI+t(LeqOOw3wGm`DT=}Y?GC<{&*qs=Vh7>tcz9eB`fa88W$Mh zWUGc-WM5e}6tIOEf|>rF6T7SrzL1_UN{<%2jA4w82G8>L^DOp{)~N5u7HHNKWXKkT zW3&y}{|&5v1Ya&@{FX}}5hR*ca0e!&<%Ef;evv~Jt2b!el^LqJxn?%7X-fq zBlHMPOo;H`!d~Ixhyyg*b90*fi1=YKILK`-h-HFEhl@e>1xzfD6cV0P1-Xw9cq4{g z{*m?KvwMyc7wYFclS95b;Wq!oQI)>~*a{-!EaVi8X>)P%iLr6#wR_&U2kJKsDbX1g z^YVg%)q349B`w&)JjOg_`aX109f&2RW*Fvf%4=!PysVz~wQjVPv1W|*|Kw<*(Ho=R zB!s&1VR1GxnsDdw8xvzYNLpf`X<+iS|8+xA;h)Kkq7()b_lAT?_q~G7SzRKWPFF%8 z@H+c_#%Kg^5f~Ya2I)3Na|U8{nijR91H0qmb41k-boLX3t7rX=cP3l})%Az^dpGXg z%~g+!K6YHqkC$c$fdf;b)5h{xLkeMiD?O^t9W0`QbRPnqs_A!^)-0`_E>GA9YScKB z_g-$fDSxoB2oPLyULFPGbV<;ttFL7H%OZdNNr{1%bpXmEDJ?ROE36g>KnN?7Cj*s5 z^Vo9)yHbw6g_!fp=54mTAyCxnM%Lmt%^O7VH;ygoS2K~3HehnOc?Dgt_Bng0n6{h7si<`>B~*hy5- zuy_ykl)))0h$;anjR;>}z=3FaVqui@eS1_{bD$S9Oa)y+W7*NCW!dXz-#Bp+7dYA~ z4I6BQZMvep>xE8742FD~$4UAjEhtM~{1=G}@rSe(VkjI>p3^G`#@Xp>jkTs8mu?@A zJ+v4N{5Lv2G5>8XYrKqQ#8`qPiWo2{qaclm$l)j$cjx?Tbn;gH=z4K|W!hmrUDTvUA7kJLyeMw`E^Xk?PdEXG5s0|yJ9#5{ zfedsBFXLy{zBV=c!DjVYXF3>N$@<$GX`phk(tN47;P+oO_RFPZR;V3`b|P?9Y3wQc z*)*sS7*b*a^deUaX{g0VlsoPyKHC2MH~+JZJ;^5;I1x+P+ ziNFbzk_b#<$9WQC%(#9`228lrD{<7%>88J>LJ3rh20i|J(D(?vZCP^+zlHQS4t78G~wV)teqvq6wD-Z=r_D$YB=6 zfcO_a*njq!^CpCPG%1T6u}Jj@)V#+rL^UteTHfR*ErXsQMq>*KHO6`q%wGrLqkTtV z(wH1YAgAc6^+I`QdAV0u+`a#xt*#Ma7(J|{2JHBm)bCPC56d#|5%|Uy-!DJju-lKL zYLiFVA*~G6Z`ie+^9tDPe zIkzEU(TonB1t)>aEL>bE>|S)K6btqo^us67IHH)n?q@{Wx1;@-1{Qwak`3TtqAZJpwOELe1l(Nwt}mfvR=FFdXOKRNtqbZ1uC3-48M53M2@ zfIM_v?ewe4uv4tXAl*C$LFN*#GqO+EY|eS1m-E)OzBFZ$-aSdCf|btB&PSXgr)jlm z2rh7PQq|#PT)0tFODtfbO%PLCgp)ICtK;h5+yDPfy*BH1;pfhk_P?yAt>t1;?KRy0 zM`EJl+{~wj`&U~wE3Y=cxc+9$mc1wcmf7xA)pU86S(5+k&9DFOLqfBBnkKH0(l>nA z^SePT*|R4_pkrreQ^KjH4nu}VCuCl46kYM(elzzqi=)$TyQi)SklYe-%Htc%4MH^4l93d(blx7D*L+^ecKgq+qYoRl*$9%x9SY%{<;1E#umO?AA3z&E*+=ye7lZTeP7GC?XJ?y4j~$tUpTm4cDA`Rr19J>Df~6wutQ1USS#{ z>axv+ab@1WJ5wA|mHIZE@z~tu{FZ92U5w4!1GW8(xy{&(H4 zdINN<85MqP(>nW3>XU-4C@6;h8>xu#FkEu)kq!t<7uc|hDQ!V;l$v5{aO6@^k(vI_ zj?KDxcvC++E}ub^Wv%J{__hX58h;`R^fb9CEx|zKH$FWNC<0s_a#+rnVG_!OCDuz zT*akQcvM#Iec9U;ZXH|Rf89D`ebkLfkA2QNJ$>L68&DDt-rN4gK}7Tq!|5A$SRZ!m zD7DQp-FL{OYbt~LObsEG<-rVmBHYJR*Jz7h5$g*!_r9mT>au||gSy1E$Sf|#Yz+m! z_)S|@JV`m|c-&bn;Nry#M;ck#xK8y=Tevzhr6}9}?(PJ43GNU)xJ!T#2o~JkodCgIXW{wp`+nK8 z?|$Aor>FZ)pP9bDu6wHLR+XG2CoJGyD8L#TQg3H8002hO*u#Q{orRZ$osI3koB#6x zxeDz5-$OYE3&*zs`rDxZ00<)j;E_;4AZ$EBLK1RnYC1+%Ru1lu9|c7uB&6k)lvKf= zKIs}58=G6%+B&+pyL_kq@-u%?CkR9;o%wb{_wZCnZC`9{p0)Ke+dR*ZTY6;NUPAK>Bx{o^_jW zeVdyA0FVS1_4f}h_@K>#sKFmzW)ViKBElFWcyA)K+9#hbw5k93nyhaM1G8%jh>a}{ z2le9%i*FYT9bMGvY2>%(!hq9#3JN473W1o8)il)Zw?|{6qtvdQg-XhJ`eSfON@8PE z0AdBd81LKzl634F^voNjWvRF*P+_bx`1m=26I~#v^6@-z4~^z}$W?G38OH~gB*R!Q z7}Hz+R`#dIy=*fi)63%^cc9aQA(siC;-JRaB~vpApVZR}Vr)-L7<+8C_8rZAJEo_ z&RYHDFcN@tb5CtRSEQNw^YrvzZi3eF0l(#=g}?C<9SV#~ex4FD9NIh32aZ16Kcu5Z zfSqeY@%+CYbN*v5nI&iS^>I0${{Bk0%d0tK*Gz1+p)5V~+Yqo?!=RhX{`ckJ1b>B{ zrHW7DbEnGS8Ps-vyal@q^%$SapJTcgeNFSq3rl=8li}4Cqmz%upW$y;YK<9O3!clg zrG4E;a1qXknh_Iwzek@udxU-#5%gzrV`8Qp_EI@y{OEhUAGZ#N(R#&eV4?TL{? z*d*~=FENo|g_KO9+wyv0q$lmS-&nIbi{ajt^3VmzIlp;gnas6mij6rpCnOahVMB>d>*UjHcC+wOS==cphZrVQcOPcpy!jv z^0VOl)0-CVStexd<-UA?UYmzAnJJRgNd9-X_=zxXrRQ1GyS;9pO?u?NCC6NO+3?qN ztbeewqd9W`KOn&R%kjX>gZ5I#MSauIs^>9+1h=rNIx{`Heu>H5FCz04H@X2W8paWe zVGh^KZTCv}W9Uv3OL&E+pC)Niqjw!K`kHq3mMb__qEEV9E%bJCjY^;GBs2I4pXsU;dvDA2eVSe{5XV&=aSo<`jH}2Z?a=u#GzOC0p zdGzTe)7S_V1t_uFO8^~b1}f{1BoT#5(cML)NemPPA^^;@x$1V?Pzc|Fm^oKYv%0S) zUtl{l&E1@3bw;y3bBtUN-5K$S{R&o9m$EP@aGmmxL#HwW zo7^VRh8Ac)Ic*y%r za9lloZSTNk0H9lw3}H>?I!cSZ#fYp&mwSY^>Ph?otRrr}@%>ov>o<&<0XUR0zFJa-}n$$}fZ)&=Y z9?&UMTL6N~AV z@N?CO0*;5o)3KsGlX*+O`-AOI{hO*`8UJC-}i$VfHf$Ae!WfRrnc zA`~Fw1^`7Y1aKZdcnIuSXCVuGFC`0Arl2KbBX1<q8 z)DFF!YLsJE=#I1T0~rO~bg5xNn5+`&Q{Z4Y5YajRnM+U>MZNqU}smI$k;W>Nj@s3-v5B;-Eg}IjdcWp4BTiZ zZr)Kh^I@EA9FEOgjhr}p1-?_5r)y8+RFY>emX__dEi)aTd@mjic(OZelMAl0%2O?E z_!^}0p&#&}#upIHZ!9TVzqo08 zJVNYi0()dg#+wos))PQf-R+;$y-Y0$J&vXbO|E@*rsVE@Kcw7HYS@#1WhY?vj)a3X z&?ZOUGm$%~SiT-$oVQaZZMsgM5*)$MtN4{L{Vz&HHuY_7-11#|<`Mdn-5MaTAuNY)jr17mVA%mo(#OaZ*C z$E;*VubK%0+*`B0oYFSx8FXgTN@#zg!VBZXjMbvrPR>TzU}e1vE^(T_*x_rg&pUkZ!)wP+*fvvZ3y{X6@z~y`R=b79@ zt%L~$m59xt8J}Aqawigmk@Dqvoso&6&r9Bj|%#NsWe;^Q*?DxuqE6n$#5P{XIR-6j?bNLz3F*6_|Lx&9toM6D- z&yo=23B=}{#1HVn^PMWJMEG+zou93NcJ7wMPK7a$i0Xi}ljV(Ht&sM-1{r;HH``bu zv-7;hZV~KyYZA>=Y&u7*5a&;^-A4R4TJwf==S5fr^V7ALq_55nGUu{MmtuzZk%X9; zH9D6IWGG5g?&{7vF}d;?>qrqv#QCsf2v9ej%w2p;??9Dw&<-X0Gk;X=vQUYS(@^*Z z1Xu$<{D=yocDJ#-hU)fYelBL9aS*2na+tw1rn`;SwHb<-AK<153QMy}~9P@~GxRQA;kyoB5Kvy6~Gh!8^LkMEC0;Tt?xjKw2aGz2E+Yo|Fk=E?M|=S^v(jZJK_Tj1@1jjxZgZp4dg{{ z4bl_5{_7KLJDvKi^1%0^vyzGo;c9a>(hM17vIBu=vu zVpg*JY*41{rqhW**~ddLXKCm}%I#!U-%WiIqO6Vr-ShEL%+oCluj*YGH|2Gbx&hQ# zc#MVx`ZYY|=U&N3??4biOTWjO14)Q zlky;BfymK-vs9vHDt)1}vb{$4ZK`p2iU z7+J+`01=5(S}?HF+KgvBvC$~g9+QMlWJ6?Wk*X-ze1U(&Pymwd+U%~#*VVOItMa#k zC5aYZ-vR+_T-VO;Kzm|LxRI;F&Vu^|5lSV2zyh-0tOz&zT2K=7wx+B;j2##^A#7oZ zWpSb+D3}H)PHsKtM88LSty4x7nq5C2Q6sJkIeKNsJk#H7Y6g2P`7LhVOPvPT{;-e} z+yBTFPI%QPKj+A+Xg@mJck#E`ndijq`o~ThqGPo4Y{P{o*Y>5TYyU#u2t8FC5k|!B zK?Vbb0PvhRBAz_H{-WJS#=yNRQL~;?MiC;wUD8ngL0<1v!SAmCb%y!@dp6Aqy)qPV zS`|k_Jk(VDlA+@rs910oH%ZlpT@b;mGEgj-g-=RTLjsfxrV_&to@FoN$%zE&8O?G^ z0&4#T2U3l6hcl{uE;VPujz=A0pDyb_E^dw)o8`n1jCBX$7{UYb$<(0}Sb=7L=VlyR zmX9eXkwAFuG9b)ECI!l<(6lPr5hH=dQOeIgyt<`@H>U-yS82@xwp^z^tG&pX8SKTd z@p`Twt&$T$>A$OkQTNFQzqWtb{{(iy{_C=920JV_6<7P59`JAcE%+<2Dq~ZRG-GIs z)FQ*Kpf^^Vtat}v>SO>b7$qu6dgnlHI2D7U+LM1Pt|%@r zp(6+*`I7d4!@G)4ULGwgRSu%H&TZQIoGPU(8C8or-j9CE>G7n&NYmH0_Kq)(I~_LI zbU`ly()rs{H30a?VJT)>){((s4yH%l%uf@%sZYYpW5DoV$&|tvNiEM|s%HnBnf;l` z-Cn=xA&@)BYu624BP=~4oe7H)M|Y1T;SzaT@UB%V0h}0SX$^-?Svs9Q!!}qUbfVb( zQVHnYL^YE04wOlsf;sSPG~;14o2J>5rcl<7S*SHwk+ZU#9_e_j3MZW{1tcUq2<`r% zD!(^mES>eLlFXGwR93037qH>Usx-JP!#*7XiwetD)Jdw`^?mcy5&#^v+6`}`to;r=h}i(#D4)#xR0)M_|k4$+VYOcK3) zh5AUJB==@;i62E0sF)2qog}tC_5)R!>eI%G$?j>a0(GeNH^T z{ala`I1dirfm-M#aXW})Go9)*AN?ohXzd2k)@-2W4uu zSdiwxQa#RUq#n!Dw-cuZ=j7yiav=Rv5zCzK5Xh6uRNOqJ-nReAZKKIXct*8ZXl^`ZlP1OjsBRz$TQY#aeiuK}k}m%QEoUBoF4R;>;a{FvV-@ z+bh>|rS2;>8!Rnj(=ElAvvv1uU0=NZpyGFh4(E(nX#ln;G8S{@{5YYZtpQQyK=E!a zE0F^JMs_6U^>N%x|KVN?T3WPNo^W07ufwu!LMbNWRmO2w;@6w6XK4+m4=OrGag;ca ztK*4p8zOHmV_>4IK`-vLRUSwa)yfUUC;W3bX(_F|C)+L+ z_1@5d=&yzkff}aAo{lt)zK(>1R!6qaZj3$pJN^Dp`k2GH*1wq3D_#A z%hW|7ZdpPb)pE&p54R0wb)sMcCKl|1gutkdUIQ44Hi4^b$O}%_8vx|yRdq(`ON&lK zmd!U;zXd7#F^OUsyya1|#`-@b-t^|QeLVZT0B&X4^W1QcO`3<~G_hMimD884b@DL|YX{&T{C%WJdFo%(5n2C< zW0%PO8n5n+N$nTCMS{i4*t0Ck8tEksimK&kQ#VU!6}m^bD2`UuxZOr^O$wQC&QMlI zvLZVA;WvqofDizA4nBLII%MrwX4$Z0b$G~vDhpBMVA}E;QwuUHt0>StOA-oKyCE#h zfpSno@d0G3GyI{z2d7f_wT9pB5eIUYc|p$S(P8b>ke|zA=@z=%lqycIxZK!RTKf*P zLi>!Fjy*E`3~e>6^=Uu54fP_0UXkk;1zoy@!Vs+pwq>AP1azr(jnW8{c@@pgWDCbf z30vM-iAk@&UEk)^)`I`-9HOQE^foirfJfNg-aZxe(!Sa@OmrqPk`d}I5#43F!^1Z> zkJve!x$DF**dq?ZghO$b69$D4hbK&sYL4uwE`<4;?;3VksbuX`=CKgErPW|EN;1rK zlNohKvmIZ2fr|N*hcX^MwzaO3C)GTFo{cO8Cuas~ZtGcXBYlJ+DBvo{H?7ag) zII(f#if1OD>)X`-CN$qW_$POv7iRS;v2eru)=TO|d3kk856vPQr8j_|<_p|5l0ab1 zQcIB!KIi1}Q-v>$7Bo+{bLubk0#b&Gh)tqJc8#ClWXg^$bJpQvGHxbobE(V4 zv#n(DD4jKyth}Ks98=vIPAKG~n(=9;|2bg!Z2H*x*4DW(w-HoOF%*3m5*+ z->QRh=YQX_2AH{ySp!GK$I1j-@l4R&%z>ChWXNAba@x^h?=_R$>YYJgsxNs8??4Ws zv$%!3z9c_cB>x^6DPY3JFnFB{WN-CLWa6hKMej@jW9w@9V_h%wIpSOn-Kb_VOcCgT z8rTO@XMz$@k|1J?UFdPGDNzlS)9*8739TjA3=qzllYRsn+kZ{N&5Wb(KrD=m zd96UUCDfht$A-*8dBJ8z4?=S&@&}YP{*;|9M6pW_x)O1mBBs?Qu-$=!B?k1rA0bn(P)Z`rc^oQip5%piHDDgB~!+sNX3T_Yqt_|%cjP?188(GKpW$lp(>fSOEXUEy09XSJTh z|Cmd{s$I}Q6d%eDf(v~f9d3s~fys_U%T4m%w1`K-b%=)V0ZDNSYhbc%2mP0y&XVXis5;Cz! z>UL$7i{@Z?lYB2{1WmOfrZvNuO`S7kpF9-~DvNA`?HMW#f8ShGgjCD4X;tfqKU`Sp zVPmHCswG=ud}JwzA&afTDzP98p(K@U5qpU8iV7_}Od5tW>%z&pTlC_Yh>EQ{Ll)nN z&HrLk*A&;-DjtvkgomZI7DcV;W-6HV%Kc-!^JNX8b6b7z86Cq;#p3jZ_`fB`<_BN6 zyRBfqx7>%7Z9T-3U&L_Qvbl^10w%O5uVm-_!#LTlbv(|FKNJ8@Blw0|F5*ISV2cK0 zhAbPt1k77LrbAxWkE)+0OtTI<(_tQAqY!(pd7vBwSwQv6A-i7#wVXSm*8G{DgLC5EMID%RZgX$z$qC$y1sm z*L=i8i8`uCw}B7y4ng#oSJWQPPjQA|P}ft<{pe*QGCX|B&6T0-_-ETYkP}n3N9n)} zvT9el%N1O?^r66@s?Qyc-dpgsGk?#bUY5kAUqWcHWT{-rz@C}*_{jqTd09~@Azm8C zW)d8wn4#wzFe3_#`!?Z^6DYbe`-w(I%A!qX#1dy0Hj|B+e*})GD?2l;zN$j|S1(Xl zhgXWMv;p_DF#0n(ml=JbTF-gmc9S#$jrRu^~E z)vch_IjK>ke(d{0=}u`b+Bfi;BVH|PWU*-)Jy2l2k3izS%QA!8Ds_qWmy{zvb@A_| zx!TU%bXb-v;U*W!RnKCsU4TsmUUT6e3#c3q7L08-t&#g*!a=Vu(dPb+h^7tLpXq(v zE3}J_suUbw!(^C8;_%?RL-K;lViD{8Wk3t-L)bXgHi6w3YiS$>81F#kbc7sH(o;*a z)VXaTB=BOQ*CDW?O?{y~0<~_DoRgwc8@#_MlTM=vp+hCT^Nr8N;Iz8~$uX5TD7e{L zu}&C&F-;7R4j<5TDa~v!y}hFQUNx?PM3zPhnn@rA4$m|PPuaX6cp28?&gbUSu(|*T zC^1z&*4*N9--q_?l}>l2pJl>xhT(F6v`Cng#!VFVuIoItpTC%O4FqVuML9um1_aZ= zX6&-g-!k_9{%yd-5gIq?=77Rz%s{zpg#Bt1W)xOZ3Uh5NkB?dU4m3)$!jjN3Eyay3 zTgOolVlq3B4o}GeGx_gyU?U>!LsB={PK4F=wz{%{C52}GP99+vPui>3_G+@eE;ua0 zcS=AK*by0414cBN%1O($I3h93PXl~qGpeCO zZer3*nT^N@C?=zV4<-XcNUluaun}ORS%xSGd@RysvHaqauaG*kwW zHY+4#H%|M8_^LiNnLAWH-4NdVPjrE(bx?kUg|iepl2M7K`|m&(%<43LRnw$Qf6{qO z_hqviEXG61lcj9bN)${-(;zR@jaG!j%qeP2Aj}IvbvACf)e9~&b&Am6I-FdIvyV3A zNvs?Uic=qtPpUm?=a!fiAnk;9*`+7zC8ec2`je|yiO#VkX*NMwBu$=tUfHZ!{6{`R z4fu-Kat(HYbcsAPgbfdcNK#z$S!Ta?`nEu@i2%60f;w0z`O#@1a4aT+w%J}`7|phu zu_fX+>d@ZXiCu+Bo5_fyK&*UgQXzpmUQFKX8l>2f8{^*jO$OPr^ub5VgD=XaOSYN5ciKe!{RvkShEgC2QVY z7dna3VBQ^H%uj&|GB!cL@lF8|yczw5aHGQtYziW+F#CSDv=!4ZYdG;Bs`*Br@?D<; zhG=(w@A|#!leWri&qQTb0T27wk3=uNSBD=_*`scKJZw2Tl}2W6ZP&WE9NSqkGe=*B z7c9oi8#V$CbT%Gp6_@RW(jYDBGo2!hGas|rJ>718&Ha~`vu2saZPA-Bf;P8LUVFKc zFP!KYuhOWrA2(#UxI)g9J)BFc>omov+Rqihj5I@-_>6-M5Z(6)CZ(fKjD{6rhzApE zlI7bea+ciTB&#u#%1m#h4}C4sW?)IdksAG%v_ddcD|SNZ6POjnMUUXRw3ILN1w-eD z)DIbLc~l8w>QQvlttnFta%)ji8m&x**7FCDCG1k$dgj zzWKAJbryGdIh9e}3e|FwmF`esSe22#vlKc`3I2YPs9y;x+vaXPHotR(SsyA+l$g=3 zksx^nC6gkuRBbJUrmwqsT#0Ky^^fjJb^|#1)4Y56e`Wo6ddgGlHbqe=rFwN~T)U#O z{$!JYJURGc62=--Hr0O6+a4Ac7A1iltiTqz+b8qq2t9%@GYuL(2z~mG(ShF(3S~HL z^_BIjQJqx+$|g{P9W^lU)HE5)2EZUgu~X*&b)#6nc#hOvI(6q`k)fd&0r6q+RpDy) z`GJTt^gnH2*??+7A2iST;Oa*&4jv_0qK3hhS{#rlg!#offHZ<)0~A}g<4w8?1j4io z+4H@KxO8Rj#&A#O21GmQg^%HQix^e~v_lS_dGX$jLbcpaBuR~&O{gz42g^Gs_9ULh z_hqkYY_PwbKJ!i!s<>f^sOVi^Kp=Zh5Xf_v-_IJljt{NpFG8&@I%m5k&L7p9$C??w z4?tD5Z zPM9N|D|wwB7T4Wcln5!c0mQX^RM4uDG2VetgUte$k zLMJABn4w@wGlKJ1fwSxQLm&io%pRH^X9SZf zJ|9j{+yj61Q_%04f^S!4ksqV_6MqFfa+dg0*(9o`Fm44L4FfL<^p7>@nJG~USm~hj zsj_3L`jj0kEQ!EsRH`spn6;=UX z*Os@a8uB`#Wv&UkpHF6mi_PyPUCF0G%<#a;`8!cJ5DsfWkI_^#v&Yr)(&c?f56$2gvtD=e;bt8Rb z(BZ^<*P^Y((@Zog*rbVDHg2M1zb4B5SJqFX2=R&X=#g&9*K%{YDxLT+rX4kM!zm*3 z>r?ujpqeytO!ZgEi2Fhm3W;_+Y`C<~1S=&PSf20Jq%eY7rQBiS5Du^^DM-@)NZ z{j5|rWI0#!VhT#M+yqDji7qa1Mim3-1zX(~_VHm1!VGm4>qhQ5L+ zxvcpc+<9qwhCzjU#&j*-)R8s?HESxa5Qgx84uKK^dWUeKD7ucz_Ye;RqU4L7I4wT6 zbfnGz9q~l#?&HTLVtt9!8Ye|#6P13~&?Lh(Zi;XZ=5wHLU)V;%?#>ed{%*kzjU)v) zR7_*$M(@a_%l_YqPF9u9H&V-&eisNbU5v%dcS*hG>GX=m<0YO|4IV{jY5C+N# zMl~0&kL@?r8OF_L^xLUpwM^?9!cGHa9ER;pJ(;IL*`!TF7*!L_!(2n5cP(jZmaI@W z+`-7NvULQTY?g4FMtsX_nlsqkA$?a6g!dnA3WL}B^IX?ogW_e!r+^jdcl!jnxszOa ztKR<24{oZlmW(n3v4`%bsB`rClb{kfRQMAZ+~|+MU>FFAD_c^zfCdfm(VLtnB@oH` zT(M{x@i*)^MnJTKm-zem8tlTNYs<`9ft5QZQB(U&_tV^Pk zzJv-x@iC*AxKuS_2-EVIH`LcIj{*!1uBb_3h~c7fl5%-_3c3#y70LEHR;J~T$z_KG zusLy&1)*3_sx=mhEG+_JT@&Kpf!630F&n`6-vs?aZRF0LzbhW_d5#AZ&=KtdK%sN{ zMZ+gM!(DS*f2#OS5>1F=N+c~pn0Dm~P`(Rbyynb65ZQ$F>i%F%+Y>`^=}+~!5+<@t zf8k1+E|ArqgZ|?66AuHd4OAzrtEd%9cK@Wd~dm;bRjZ$;gS_5C=#%|d9U&0~$1qag~TOP`h7%?LW%|y$%iZfFF zb2(W#bDvy+c8)8^(~yXn^||SP8se11Ej^e029_p}Zs;&Pi}(Mja~6_+xZ7}9L_BZ$ zy4$%CRvkrO*cW{JvG`{1GSS3)0ZMFZf0~Xqh9i|aJ$4~6%fpsYk90w4bl{jF)m(72 zO|>x7sWkV*#6f$)m@y+WRej8-M@9M*U5Onjb-%HGn_Hf>gyy)S(t&a{;WWE|2Kk`6 zyLfdI4UH|kz~d+mQrKyJL+{6git@J{-aZp;BU`0On4IqQ^HEDpJ#9+2aR@e@^70Eu z!DOG5(H4mB$DJ5`LP>Sg|D?8Wx!#gK=RStS^n2)fw6UI$jyVTIzdQT%Y??V(q+a5N{m#Z2wf1;&_w6*%t}B5e%cuT8Gxy<;7_v3T&ALN4W<1B@oqId> zJl0-P_uy2oX}PidYI<8q_DA*%gSZb?8`BhERi{4d`M~N1dOnPvN8M0*n%~!y6EB8f zx5+kOyZDkmb6Op3|H^6q&#(Sv;na;M+xbKMG);w{&@#%%RXbh;va)n1(uCrf3uy#(U|CT zJhHBBt7#P}LS%DuAfOxyyVXU8$aW{+)TIX#Ll3$;{CjLvuXOppwu;NJ@b8_FD(g(< zYFon;IF_e2y^OH}x$=PbC3nTLix%v%oOl9h!f?1^F^?sD-C6jmy2649!+$|2r{HP9 zcuGm`zC1$jmjot4bIt1a=%n&8^4jxw932w?1`yf~>BcAV^66)|LwaB`TO#1YDl#@R z`VnSixD8(+)wD8@w-%g$7oYKBeZgFZASYEn55FOI!Ol~`vrFYhOK-VEYN*HNo`Av0 zP{@_^#H}o>VrU3T#W$pRJ>bODikFRv7EZ2OL4GI|xBl}mWSB()lAO<-KzF$&G5>Jf`6LjkS7twMpMUc)^ zGpH1~3byIeB9~>XHaXq76&`kA)RM zmfViUrWoss@a~Xwe4=th)GFaS04TkTZ=F808MK(j(&HxakBms#Md$LS`?QZ=M~W|T zIE_+2O4MyQhY-ZR^`Di~qQbS>9U#>&IoFd^6dSl%^dx2nC7DT@y5brEB!EB!==(D$ zVt50z?E&}7ndF*j`4mAX76q^5q=%s;*M`7@&$9Vi^_Bk?^P3P$615M-w4`s^vf;P9 zf-tJJa+EnveImT1!_2h-Sohyc$U399UD^X5(SF%L*k24yA)j^I{Wg3S@S2@V@twbI z5xpFRA-Wl&U-0sO9XK3k?4Rni?G&!}N+%Ga$zv1#d_*pWx$sp}Q@xkt^6zA}NHAfu zK__o&fXMt_uc-9fnhox=5#BkiCGllomW4-q0Sh*mCZ{YnfvLEvoi-*+#KkafT_RLn%=?}#2gzieV7KPsEX&t);G4d$LS8JSq9wi z{dzUbRnnU1YzrP%*73>Iii(t?iB-^wCk%5dF+-4m9cu!M>58VEH1_-4YUqvS8em$v zC4T@MJW{r)xw`E3y#sMDbbP4=$R*9vi=xJlLb-9wZ{eah4EJj>l$%uzQxh0h#-LJX zD%D5NoJQ6EQ^&E~&Z_8a(!Rlq;cZhqu5n75v*#o|Es<5VD3)(HFwzbT!E!IquP6L2 z@!Q2%o6qPm=|yFh%p;Ig58M) za$(n=Rccps7|Ub(Q*hf0Ns-~I>~58^Plw|1C@HGlp__X>Rk6RP zbJp@0Dd*CdyUm!%JCFjsZBOB*(ahjW)3lX&kNsC+Rsp%-IxH>@ml7TefnFQ=3jd=g zdN>27`@%H1FqI)L?Kmh*K2Vz|5}ARRL1YDn8dFRGQd54LPy$yJcHwM|A;WTXU5@X^ z9mt$hWGlbyb%Rte08q@(GUYd_r~JijK}>Ortv3`J78eQ?dV3!YDc;p-MbBjiZRh-z zI|4JBZ=SSdaaSOj6Qi=>g9caE!>5!HcO^yoZpj1;PLfKVa0?!2ow^blo0<7!VjBCe z+WyoP3TT${5-k5!Q>iP<69n}^g2%1YiyHftp5)87M%I|7SSDA+V4g+5+v?|y~HH~XC&&8Y?> zV__;M0V__V2#*rMnQYuE6C@o1)klm3ZTicj0@^CWjFPE|w{_3q+|`ai1HR} zf-@%>1`CVEBq$qc-i+#YxZF|>)qN_C5%5+Dj;B&DDpx9ekN`Mv zvK1GyQRLfYAenW`DSZJC2Wk8O5K$7qV8#Y#>c|xUQ7OqM0rG{^2TUuHuwT*Ol#KPT z*<6bzuv0gfQcdtZ7@UI;t@>rB)w?|NULRv3Wl2t%kU~K;!e4a*&LF4aq+O7^a!Bp3 zCiS=Wl^<#3pIaLBBj_ZMP-@Dp^)OAyhPobWDgH(SB4fWS>@}^jBj`|m;M~OQ`2a$L zV~#=!;_UQqJY>2*s?u03pLOa=ly3q>a3cqEej z?_PCY_RY-M?nFQ;WBJS-D2L$;T&iH30X3DwM;vB<$@B5j+zrjK*Mq5i7I{Wu*hW+K zd50slfs_rh(E|KF_KmlWAjh-~N656}OAQJEJdlnZ-L%mS1_lPU!2S%PM?KywQ}Vxf z*ed={63C}XZ?IFUq{;69DEOW6mFu-H>GWMgw=BIbPH5BpjvTMJ_PS)8h9)s46jahF zL8{y3WbO<|(4*?t6k2(?n*cKRWfKV@*kjAFautaT${{D>(4xp`@*N>>trJa0g>wgz#PIO&~+ zDHDZGI7US8^CoB;k3MT5vGRj}pDa&9NLs zkgS_%#`7Lf>gyjVWx#BU1tz>j_(>4#JJ1rX!g^rKkc{s@UY%-y-Gl}-@gP7k&ghRA zJZx~mt~i{k6P&WB7TXO4qM&)e_`I1OXd}}VT zG?$7X+8c^}os`vyDk-6_GFtM-A@xm98tdlED#{jrDhrMv-Mvn|3h$-N65=mt2N72DvLabF6;UJ3jsZ zG4veg-Z~{@Zb7k-4DnSP7S__N;rWXGp({qPxW^~~QnY~!4VT7%MhbbD`c$UqM8iqd zO>aXK)*nI`okHzAPs9Zbr1yCxtZ_0=go+BaprFX2t1t<)_D29%-47L{>(18isF(qJ zu%>+3ZbBopO}OynraY{Zis@j$|Fh(NpQ@s*8~7p1;b@fTMT$c?$>swi{frY5L&T-A z5yHA*0gGnhXcSoZL~iCzN#cbUA)>%QaYU9tQG)4DbCymM|1P&MS{tv{eXhPf-Oh|@ zVftD^7wF_;=Kv&~ljWjK2X#Cuh?uZ49T_XXKWwNRG@-XL1u34l8sPk>=7CbF}ze5;_8GNsz7#-SRyP%;9{|f^pp5119@Oz z!wdx#BXgk?pR249D$rfQ^VzTAf#HUSl zE{gGD*K33Cs*RtZ!<=w(Ija%*k zN1=?rPPXNwTC46!fF1F0MEhMsj~rFFWPo*0&g!qfUFw-s&lc7ZEaiIB9p<6O1ZXr> zB_xuX6)`<|UgzR6S0fkELd=ScORYC3HWGm zmL6!(l0+DHVYK%qssZ(3ZF~I8nQ_?t6TgESuCMrS>3Rb!22N23ee~E2TCvr=Du{y;cIbkD zulfKHMDmYpREcQsi^XywJhl#?a6$U(Wc3plJU!1O%_=RNQeO%$=F(-lFPgZ9wiUP-iL;Vi4LOZg}A#5T; zT<)!e$QT)>q@Rc-gviJqmk=$d05F;zl#*O71FP?ai6_%zz=ddFN2|%phe@zd0J@`+ z$YjBn7_udQRikyZBQuXF@_bS&qDrKwnj{&l&1AphKV5u%g1kOEo9R|isgPqgN{W|i zNk&zJ)#)>V=Ap7J>^~k92H01$xS4P8ALO;dP^wW}ht=b8ad>bzBxg&UU(^1#m;FC` brT^Pwy#qnG+;FbTypuxzPhRZ*FZcd0;ap`r literal 0 HcmV?d00001 diff --git a/frontend/public/benchmarks/tts/magpie-q8_0.mp3 b/frontend/public/benchmarks/tts/magpie-q8_0.mp3 new file mode 100644 index 0000000000000000000000000000000000000000..2ad79b34d0201d4b28e83f97709e3e4eeecec58b GIT binary patch literal 17806 zcmagl1yEa2-!S;#P9Z>$;%+wAq&s<~F{T&bwkKjEi1vNDt;|DMY2REPKXAyC6DOm+&H8o8g zeIrv-3u`+^7Z-OgU;m)c(1_@`#FUiutlWa);y2i$q_RjCU0|UcjlQVO3iz{oJ z+uM7GC+C-!*LM&9pl?6^nm5zeys`1UZTvS;!}k8KRdExKEkx7*?*9K-zIp}$47S5U zFv3DG!)$T81jB5x|NFyVyIxZ2VQ6S3AbAL` zPi$TP!ZB2=xUbO50{RjuR9IN-a`B?&0UJnYf%=+=KjI*W$!xOU+P<^mG{mbIZ7fYku&}R{Cj{w?D%}_t~oHWi@0*g zBdgq(3aqJjJF#c)6D}c)BINIr=O~dAsPCA(vz!i2@$=t>98mgAQQh9n_v-09^S2mh zbX6SwhS`lFE@u;+E?aRKIc0ksyhhGg*-aoQA#w5gRh-1jm$ckx=(ml2Hu{+XGl95! zJ_6=x!Ufn!fgSv$QvFNDztN9v_D}Nq0Ln9)Y0Q=(XZR)WQ-J*4azPLK)-vmjNq#q3 z2Js4&_6B$BdZxSY-w%0{nOPaUfQ=2^daScj9|An{uas{fT$Z)FE_BoRq6+S+5P|Pt zWkbC_pTld7HQF4`my*=g^~{3F8ZYRaYq>Rx`8)}xtsWya2Ntp4t_OY`JCY8!iC8;vN(y%z9uz+^=C5+ zwDpqr*49HPS5Ey5@!YKiVJy40^5~f2IOy%nm!qs`$;ztK>z?+^m+LW~&zg39ZP(eW z%Q(IZpM=|Ztf_uwrD$V^99Yth2R?J-F|yG6@Gxmm`~N3GA7c8OetBo+5g2n3cmdfj zR<@dHAOCc)ie(_z%o+QrW^OvKES2u$i`7KS@H5w1%O-rc%`;HUB-@oj%_B@)ka7!k z3EN2-I>M$DT-yk#-qHFEQRy^&YPPF2NjR_QN|em9uHUctlc~&x>2s{_f%V{SF){@ZsZuzN&#}$&8Gq#%-;44#{Vpywa(YsjlwT0s+or>#Blb zf)+h-zBG#l*dIl3!u+N{Tn`!;4g_{#ZV$9(gkSJr?l;hT7F^i;v+L<+L>fh%I38(414EZ(V;=QO=`b7O|5^<8qx63Hs-iN0;d+Oo?Wps#Bn_ z3CVHuI&foaKwrj3YTKVbL!r;DT-vS?)dBMxi_nsd`Hc(P>?ev&2vM0vzj`2yrhA4K z4BTA7YQ`2SfKvT}th^yZLDQ4z%X{X66vt4|ztw-VZy;J$c+4)+-Yve`l^)-B**)=o zf18S9J6pD>O$A56^(t2jowb$NRVO`O=^08Q#!; z$RK2k95~PS8`{dJnNuZ4=FG!7S;U4au}3m$n~F4XceuNJKF6zK0#mH#iOvqRykX5aas0uO_wT> ztU_m{PF9?GU3G29X_!hv)V}f$b!&WP5ht!HxM12AZS?6E5&ENe?eeih}`hd zbLnmjHtrW(W(9qkDwS63L|zcM=$Q));hLwF8W(qCUmf7^y%)NgX{s999Z# zH8b+A{Of>?pJZ!>6zaIF;nZ!LQey}~^gm}fe8X`fg1<{uS}IDK@ozH3U78{OJ!YS^ z3Uf$&gk+tqG1CYqx)vss->KW!@;JVmN?|MLq7Alx11Zq&9xVBJ$Jh!EBPSSHcK3~QXg>U>Af7m>Wz@ekb?&aC zDmNghSfF;G+=qb3DM&42S#uyokA<8VSK_3Bw3$^WdiC{N@njwM)6jOO)u3@H%cP;i z*M|K9#TFq107$uVRxn}Rq2SY(jj6w*Gf$fX8!6;(AO|Laoz9olFmq%f2NA(VfsRtT zM8HLCh2{df&Z>e44-^X3P&Y^C)5cnUDPuDKfUd^otqMJLC1_=keQY8;3J=Jf#+9Ml zrw~aW#1#=Dx@htVM8(V!a;D3U`%2JXAxzlt$@+aJfZC$mc?lD-jwlmM*@Vt12FBv; zsh_mWwmdr_ZKb!xIE`cI3QddR%ox+cDH4o^p_j3g9}AUG`Zk7o$yW@g?6n(^Cb85>H z7#C3&K_NIFLX$qr<82`X>%Ov}xA`ZK&Hx@4yunanE|&Y_!Dj2#Z7F(JmVjfLT^QVO ze3)$M{)50ZbPe>92wjs0-lfyQh3_JZ^r8SU-rSRjpa!E#Jn+or-yt@>TL}r-9(%qs z(F7e&e2o|iPi-YxhDb{3WAGPXYz95q1eaZ%|ISfV$aY7Z0zzPio-|{;EogSsgE~E4 z-$wBV7d~YuuqPij0D9$EU0wP~kPNk$n4jSdl+6H(T4Xk{3NL!4Y}M7s6e2QX5>5^P za<|{F9TwsEze$M*0Bg#frpMQKLw=nwUVQdfCuros@q`CE-ZbXQScnLv^KT! z1qsDE7Q6*4NSK}N7A&fr{e~ntBo;uOPbNyr|LAU@j{zT*?(q>4Y4P~oh1cRszh18M z2ZCeN8g54eM-k(4Tr*aUd{rnghOHZk-awrUJHWJmb7W?rAQL4HUZ96y(s9&xVY4zc z3idOKUwGV0L?@oGx;-swzsJXVGCYA^PxV7T#EHb5bvlA{=dQ_g-C|~RB3u-HSR@a% zr^D8)7dxc_U7{zW8ahKFrXnNVqsoet81=aqV91bfbGtVYrIjXYE!6ls*4UuGAKAOB zmM5WZ`afisC4YCw*^n37c_L-c9p}@dgtr4t%{3!1Wty1HxmBqS)z>3i$~7#B!ztor ziHk*xNDapH8dz%7Q&E2^CkAu4A%%^-ftKh+>yq3S#Fff_?ZGyAJAy-iHd}lW(+O>g zpPBJGvu~-N?T@Nl%VfXVV&e=IeD?$kWc7jcPhlG#)`ZZb78+cvU zr53EL7^Ho;PRMRq1dElkv0Igb43Ym;I{i2l8p$|EF?whfT3bmK!sHn%FN38xdM|B! zgGI;zyMIZG(P85Y$_?B){d)(`UySeudH_>h`5i4JDZ`Y&(GN{ts1&J1;P5~v<~g+U zRTu$-dq?=h2^-&#WKLQjKGI@Uwlx=rQ-q@?Q(3m3pcsdIU#&SZx;0MEy@-{Y<-3?J z`+zP#??Rg&)p^q`NtWE1s+W6@gLXtiMhILkQ4oSfhN@7~Xf=t6k$Uysl>-j`h058< z!jBk{5t0*$ZQz=-baPZ-5(^+ko9bZ9gRqi_^^?%w1%d0+_PGl*E)V&XKXrmbU?air ziM=5@0iG@#9ws>$V+=(#wo;IJ^Ec2xu2f(k!E6>Y3Xz|;X?Ks^-SUTackKH{V&>L&zALK+IltXu;9J=M3N}K6YOxja~Z=4<1u(m;LDXZdX}i zCVSn&k~Tz%BD;y#kiDK1Il>pUAC`YYo67#4_HvUVRpwrk@%-tqk@P>Bp59r+T~wih=%|@K`TMkL{`yCu0{iBnPO6ea8|==7IOSaYSMsT zwDe9`&L`wT#&gn%gu}U`E+*KkZ=hq58;}?BSTm`&-5m2EIe-xung|Ld@!`z)$Lwu| zfXNn58cNJRgcuoV*U{Q#-*##`OQ@xpX42bEN*y<0NSJ(~a-dOpvtZjc5GE=6Zky}) zg+b1Aj`p6;B_X?MLT*(XN7cOQFoKIyCo*@~X5AZE7-k5$F4B?$)xxLIYJ@Fwn7cxj z-Q~zrgsGYCB9*m#O2Re(I`ts$JXm2cDUP?yvMA;MbC)NTjI2eE@@pjs%$m3&puhih za5v3Se}dAvfAxzqu1m7=@yYNTsEFYV)UD^Ooh}{hoq0*;(|Bs;B2AQtmMV{JJkl01 z*RN}gfV$z3mU3NT!(Hj^QR#J(5N2;mjr;9f1-)v!nse1s?ND1W(gpp{G&VlDjKe=e zgG9sGog+Smt80YOwKMs5g^RL|#O-r}51V5wR{-?&-$&UQpc;wRJ8-P2b>27yhc5ol zt<1NEm-mv)d;|;GEIH>loSWs*KEv3JJjrah3WvQbU0Jgb&MCSE=?8ve5;Q55aK|mi`;)JN+4`O;1~UI#sNb7D+SUu*2J``|qDsI}r{Bv~TC1 zsM5oSy$8-FjZ^yolC3oo(pn=HRuKfNmS%}TB9s6agJE)X6}B}5pkUT&&ZdQ>y)cK|EfXu_B5p3Q5-IHiuavz<@XlD{?4_M32{xRieg+nv3Dx0KPMA!)sWGln0l$m~ z?b;`MQ@d4@vS>Arh}vC3IR4T}hs7jK%RR}FMlByD)8iv3G+&l~6`T`sVh7_6Q69I*UAQ&5>eP?PMNr*e{cqL&kVjuBmg z#w%sjgb#OA{OhyJG_kN;4cMX1*uXYx+WK00|8f#n8P z4qyj=u3&>=d>OFSMK@y+Vs=?i^F!Lq#Vfy3_GH{U3e`kcR2{sc8r^F9lfDWeJ5jx> zhFuHphhf3E2#KGr>4k$jt!ak8`diyd4w)I#h6@3; z&1wz)P>KhtrivlI>%o+M13@`cLD_$O$cVd}=rFl}b1(UK#-)MuoFJkUBsLpHrDty_ zG|TG~f>__|4Zpp(xK%2@Ubef6d~#lmnh2Uye~^-!k{j#@vJf?zJhet!ql%=I3u$5F zG>IOhihMz1DXR{y(2h|GIrIZ~eR1TZi^B)9u2U;5Hz~!1S7dB~DW-2NU#F2ZB!Q-@ zQ#6>}df(QXx{{3gg0X?qG*_VdamP3kg=0b%2g5XS(Tr`F>nEnQ%FqWv@K1wMnd~96 z>a_CBC~?TO*QYPO4j6BsOID_(;LFv-B%Xix9ck%6v#)jVnR+ocIA~z?))*`0X3fr* zyJ>e@y$x3C8+qBN;_=&acSP2ZikZ5B_d6?XCZ{_--!O{ zVi}x*-Q(=I{t}#2`NX8wjE!=&ce!30SR2DC`6UElE!BY9e+nWTyNdGgKQTdZmA^< zc4Uy|tPknZR{QV0;EaTPn=x+u$0&0T=oE&_7`sLE=;;fY$8e}W0vss;>mqa? z`mIGjIIrl!e_#|+PQs53C-{ohLIK&%tZbe(g4rpe)#~+dvC|mOf(t``E(BC}f#8s_ zkmQWcY5n#n5!$B&-ayZ6u^@D8T`9o3Q&yOdi0lMH%Kl*-Apt|wpciPsUsvW^Q~o5I z0#+%04-X6s&^sbBsZb>ZELSERy_wg$+;@AOd~1ZaTWJ#~5@_)T2;XK@h{0a~Aar>2 z!J*9{6IW!=$iw~B3)J0!V9Rx{`0O(e>q7TpG^*5w_E9fEQX4EfA(+yDS@LoQhym{S zBrmwoHy#I0%shJoJ*usPh+%S80J9XB^w$iLdxof!ZSG53e5QIEvSg`MHxb5Ps>I5p z>OmFlKYeVZN{r2~Q**gPih^C~kibQv^xTw!AInTBM3I;>u^JOmWrbu>>M>)`VZ@c- zElX(`fzj@o?vdV^US1KI&@bmrVj{5I6TUEG7nMKz_#&nC6UN*_<$Wf{oRoIRz3>wU z8N|?u@@CLk)oi)uhNI& zbu*)t+p-5b&rG7*RG%JmC!qHQztPK0oCePzL$df!mRWZrjviZhIhoO(KDMP!G`MoT z%nGr?Ju>idZy;Q7E~xqauoqx{k~cEgl@ycK1=8^BzZAAOz&+Hu2j#jQVkQ0Z_hvA= zG~he*y7o8pzW%lQXa1kmTb$@$iz-b*9QT*;|SZn|9iQ~<*d?qS5pQ@H;5Rd)H?&6q233qyd zGOq_4M@I%O7mD^bx8$kfWtp_`k8+m7qH7CwqhFcp&?|k6Z%%SDyyooAts~@V5lTfQ zN|h_v_h(ApKzOWvpt`?OTg1bohpote&AuNs1m?|AEVmV0+=NBT!GwOC+T&M0{Oc7j zpc)3X_}%e`Laa%lKx}YETX8ah-Szm~%lkfsw+1=qX^HYTyQTFVugB8Y7b0qPr#~cTz_Cu znX7AgnVej#n~9!p=-xS4_qnm>r(5ASz>rCRQ9?DckYfTZ;08+lHCt6e0nNxplYEsy zlq`16&im+8t|d9kNW;X(h+Qb=AsxFnI3g41<&ifMzVcNDp_kjPs~<)bRWyY?*j4zh z0x!dWkK#HG&omr|VXcRKh_{{DqepR^GP}xO z(-~l%f0G;iBFw@V0dAfj-)nsDu7g4iq0hHf&y$Tki|_7j+BN?z@H_1mCjix$jh?)P zTYC|!40-`x@M|H8Z5@8ZnsKnP6(i>`faujVm_bHs&Zoy^r$6h_?A7xUgh9&9lf1dl zWk|pIJ)i&dK1wx4qEDNw2^=|mZxB9YfC)83_eNc^YBH7+xQgYa<_yg)%e2<~&10Gl z&gWix1KBV^#`6sis2E-_q(V1lgzc0@S-K7rc-Jnp?@-a^0to zW_$^64|+=JkBO;sPoE(UKEDf2d`aqEF9x-w`Md%UsMK)fXiMTsOb%%|#-bK@aOh;C ziJeLN^mFSqBFYJrgHe!&Y+HoPh|$Bz)mDfPm1)ueh(f}XX8^Isk5q6jw#>{eO?dF= zlu~b?07g4tzX89z7BO!hEUr9dFrtX8Wsk9Ur8d6^?JeAlesf9@^2V8RBz@*bh7M6d z6MPG<4&R#eEC(3gKSKgq3oy!Tzzb(WXODp%NHGzvG+g@kuF(4chhlqL76}28g^`yi zeX8gMzIJcPRHObN*eI|NKc>1fWLXxXV`(T>D1VT)sJ>?q(9beF-956Y zdIM!K!eh>WgtKbbNG7cD+iezGb5iTX@zSEbvLdbYRLL02l|&c#6X&JaO|O)$>UNqB z3uKGqDD#c+@lgypW^SriHQPF@3cAu;eou=^S&f&4m9*MU>X#+|#04;FChu^Cg`{F! zQy$MAl_*A^%x2hf^3$PYs41p@XHaSZs>mh9Ay6tm#VkY%Ud(?xp`bEEqtGy5WB2Bw zxd8HqWG9=U3>A*#@YTx6>U_H6;4ZO8r)GljQGBV}Guw`jy)?ve|2l@js~woXP&tkE z25M#S_31*^#jxK2hvY}heZSGJTji3>!Bf+ExX|z1ueHda5nDYeIzkdD?WA6S7`3ixJ1?HYb!_A(b1`!3N8u+x(-$n~ zRAv%b)1DnPaPbs*U#iX4Yv)oZ?F^KA%Q5$oD_j)ggeJ^?bIXgPDl3cSqzBi(@RXmP zf3u0_g$EeJwwAtt$vL|FX|LEQ8bXr@kIQn2glu6+J@AZRt*lhbT&#&KBVe(0Oqkw4 z^Ykeybwa-A&;8T~Zq^B?R#IOby8@VuFHIkCQ~4ZAi{&|&NI^Y{sJ8JF+^S-gnlXs; zd6hMMINWRo>+iJA18T<1hnl>KD{T6~VH7MBj58yw0<-C$UbD1{Ww0vEN^e_e@vSaN z!nU=a|H%=E)`iu@(B*WsOW6tQ9iVk>b_hlLCb-ZnP)LNHHbhj<-a;WtYL1TbXrSvz zWdI>P+vc^wEza>a9RV_n44|!Gk+c`}$n=8!^)Us1%YDFQY2XV@W5N z$iT%tRhnHct>w*FzOVJl- zL~rKRB~oI7woP>?azo<#`R3yhKM4&p5iYav*xc=5cMEb%clmpHz-3n-(GnM76CS^b z6wmkI0V)o!0}GiU}u2n8virxY$F`!JUy;-(VzRhErCU39ODo02uIZpnXQ` z{=D$%xfTF{ZmB?6NSJ6WE>;W&yc`ynoicU5+4$gD!$B@o%zoIpl-pr}AL5C%YQDlm zf;DYYGE7@mw69G31%6iYg7-QPXW!??c_^&YFOAS#C%?H3as0mT{9+nxqtUd0 zeIX>=;sZ@o+%m=2;kt{XXEI3-b*7JoFO`dqPoA6EHu1P&Mq}cP{}YnvK*TLs7k++^ zfeRsH5trLGz7Tjy5jyiALZxc@e9`L+0_Ic+5V6}}&RzkST!$|64FqLFdx?afN3aKQ z25ObCVI@c(m@=F!3E3ki+UUpjwoIXwGa2#v?yqKv>XYbww>8K*FR1mwXO)bY_~m&Q z1Qp=U3jJvjz+-gT-@K;1BIShHs%%=t0_=Q zG9Z+Zd)3sK^0hFs9$CwZzgHuGgnJ4Y^fG?kMMcyA0~Ah@cG-}BK>o^b5bja1T^mX$ z;r|46^kb|3Aw~-f#mdAvzGNo&B?B7+w{3(#B19gxK!KubJ%#xOx@B>FiELX#K= z-b`B|Lbkx1XULc`)>q>@%aOQOchp`UFo7)ES%eYBw^e_-e`!Ch*4DZEAdg=M|Bs-; zg&qg)A#@#En42=;SCxp33r=QzM%*{U_sg#>xyOeTPe%uJ4_unoKD~)(m_WeMCBUlk ztCLkx*6U&2Jb3sRfVKYufUPx^+t-7R6gtI3yFLEM;iI&*3l9-Rk)ePX zdZ>2)uByw4N%_vbW#i?08+qTR8(j#X)yc`kYJbOb;l%>Mj1;C&0;v2N49Y$OnD3R# z!DotKh4A^Xkt5oy#B$<}z>qbg^KpYjd4a}-Z=eIloU15nn0?4uC z_jCRk?5KZFem@V>k8YZA382o$aa#nNJu@#~?-JltWQO4VG8(^@@rYHnR?4OJSEY`z zBIc{qr=eHr+CIPD2rcw0408NK&cb1)4>oa_5kSpX$hk+lCrh&|@c?KMIK%%Ng=Nc8pbIzP+qa8Btl z^zT+{zYPypgH~$Un3fwIHD9~&as%ugx*S3@4lk53jZs}g%S81IMjtV>R4W&87EYN{ zmueGUJmUhO?K9I~*kn|n9m#$iH2g7`VY>B+d%ELf6B_-aRbytP#IX@q;7HN&ErKv9 z&fr|Cp+E%f>n$V=VLjkn#*F31UZFhtHE}8tqE0u55d1IX*)pL+K(8LMXt2>N3p(TW zJN1e&t&TkFD-ZAUzkuKCa96k0-#~Rzu^>0SIS%5GRcghsgTNH{2xIZS6-4xp zgdOBrquHjh$%R!gVehFZV3`?aeahF!x21dFv{3A6Fc^@d!4kA?BletdaQ;G3*@KA8 zHpOhlSYxyn*%1JGxXO~KNWNIU;B66r!9Wzuq@9c|ATe%Ynk|@|I&rfXM=Fh%2_WHBbq8I4IQ zT?PlCpGb_61Un36!i>nN+)A*{uG_Bm!)o7Es%Z{2pIL`r+sT~OXwTr$qzWi{VqN}y zrA(GzsUH)H_|Il}vC<$p8S6jmu|Nb#!R-(9F-dW^YBnDs)lle5z9G}1Hz$GTb=vo3 zR`tt5dISwFL+D>W3yX;1&rpa?VEfzrKYc2+wqh;g@U6!brCy0{ojh9TG2$npswyrO zu9teqK4Dh#Wq)w&YB-_-Ef+1SXaM2YmFh4*0|+&t>ul44-%%2Ekd&>-dkN=2tByG7 zi^aw@!Sc;NK3`*WzIVOwzP!}eR~MCtwN2n)Hh;KJW-|TS`e>qgcoN~u^Y3hm&Gwx# zQ;Y3ZEU4Uv&~CN_)qOWUEQbcMHZID#IFKM@$jFGB-;|RdJABlKS}+@AOGBU~f>Ui$ z>8@4Fq02swh9)dB7j5%H5`)v+Y4R;VDReQ9%-C8&Fh5)Lke1S!#7RbL)L|vyT_j;C z8jS*FifLhC0+r=}#5q)yCD|E)v97>@dAgC1Y}k_}bZu0tz8K?7NGytCP(K49`Egx# z-k9_>DT#*K>z=&kEH5GYU)N8H3k$y-OmW%1WIO%BP1X8F;rA~QBXD#K4lPI?*E;EM zM;!x!wRSdK<|ttp8Uh>+8)cakLdNNZ%D~M|Vc!748e1Q%&=tASwpj!%3}=J#tSANo z37me^ zjm5Kt3P+hzI<5RMhJC=~lFIy$ff3oCIh{|O?B`7>*7zmS6(K99+Ed-pQYK4%N>1Cl zaJGL@SA><4~GFYDdcuB+!0m6$rcE*m{6-WC`yn&EPMOr8k}@42gsS3CBENWJ;Bg&bMhv9mWw7=mVTJn^A?gokOJy44IKJiS)98bDA29)+cUm&)tS zDN%>Hxqv8(#0n8g!c0B2*JSO`35Id$;Q=^W9qQiHFZ=Dctg7ee+dqDo+JrPx!zl;N zZ&jcu29V*s8ZLzAtZI6pk=dT(B+5o|g5JXjB9i15H*pOQY>Ob52umderB*Cce~*M2 zqb}25*k@kw2#YvV1~xJj^s-9?`l7r>30Yc^R7SY|5xu=69q+^@H8aVRO|2->T_>$2 zQ5xMAX#3Stkl#QY4AqdxciL7k%KJj4^85{z*|9E(xIncu(!eBysJW>7@{jUriMOPc z#h=lE5@Jyu;;{6w>Es67I&AvtO%8u|`D^FU8`87jIz48=jLlsEd- zWfI?l}^O=RetKK&&w;(OA$>W34bsu*IK>vW_4V z;S}nuG4i4av#0_HOc>Rl7MLH=!K)#CeSxs#yppgm7I27&xPgK8{2>7VQ*x6Pgucid zcodu_W2;i|8%T+FfJJCNuMi*v?kqkAydUfk((7A>qjtkW=6ok51lAjZ zn;oij8vmI{!p75e5Nw#~1z6dVEhxq96HJOd!RsPGK~|o9{t<-i6^i9za3(J{l#hn` zI;yhkdD6+zN{`8NeyMh8*gjlGT}#150{1+YBxw((Ga3)Auzc|#m{jHa{h%O5zH4We zpTr}tym##J7wl>0^FT)Nn@scHZ|F^PWeDCq<*9vssfTt`2X#8+EWSV=p>wNh9hz(u0~aLlLK;RaPe!5*%*C;FznTW{ z5wPg{&V+WCb>;g3T@Busn#=Pngmn&0J*Nkb(SJ{P?S{1(qr2G0qpOV>d zCUPrCj^98`ENY-a1Fu!$!_RwHbrg}JaXAH;V5jU*j^9@uu!zX^^spEvcE4Dck|y6sNz3JNd*D@{ISHJAKDZV>gjquWNa?a( z-5EUAzJa(H+(31D^oCA6|8$o)HgvuSPv7cWPuZZ`yj*kEKjh3%!8QR8^oOcP;}<#& zy=u%1n~J)}56l>%QD!5=SurK8eVFN$&9dJg<{SXk^l(N@fVsh6rONvYpZT$CkZ$ly zu>8_1i;1Wtz30MGV{5{&Lh`2kz#Dzv)hFX!?pN~w>>ZBwD~ z=%Js_{C8sGbQNUv3A-_D?D4OEoYo!Y2L~tdFzr929w0{cwNhNy+BmAKqn zO631D87qZ}Dk|*E zxQ~2W4aYy*-Lc^%Kr?9}T~W@9nqJDt$mOq;lOSf;FcPXk`|D#I9P`qpOIj{)Lk9$M zq7KKp(S7RB&#rigF&UHeRYgk1pI_+fUmCKuh3d9d9DYL)bD+mAp**ZF82{3@4c9gB zd0rgfA_QTq`k8%opk=)XFv23@=}e6egjerxK_)_I2riW?QQ2qj7u_lAx9*NVNYN{@Gs7xr3*NR^pC8F;DwQR z&azNJ?R4(eM;+D0A%m32gHkz0Wz%g8g0M4A+@aTplEw=|?~Q2TkD$-J`?r-S9${tFoc{0Bd!fT|MzGh4QnDy z=Y+RC;c%^HJOV8_i2SfhWcnEs`x+$}p8+f5>evqRiugUeq~GVi3kRDDcN`ClFY3ps zKU(%AQG*L{Auu|47N|!-7Htqse?I?@ULUxqX$eK>abeXuKmss@M>EglVgYBGFME;U z0Ii+jU}BlD(h?n;B~7ftz=kdjYNYqhN2n_&4#x6GaV}bAGYiGW2iD6*RT$J$?H08% z+G{+_y6xfZiIKv0H@S3C!d^Gna1F7&+A;sBw!e|{&Be9N14A$2o;zS3i#b02Sp+>K z6h{-@6gY^BC4NGGNiG|jDY%)VzAx|>ZS)oU4$PkpNdj9Dq6`%7X@!Ox0D~0HwQ(*^NbnYu%m8sQp1e8Q7h4WU-NYeB}QR47=Hd2X-fH! zLE$LeB&lC=OZH8H1a-)yiSbJoTGAktf58sA1s<4*52K2i3aJ2av491%cVV1jmB;V~ z0`S{0T9{(nVqcO??_cyUTmA?dxn`N`1L@u(1mzlNgJ|@VkyVQz%hDotAZ>i0v`kO! zgQn@2(pQo@J;b@ORg*$KWO?%thvNlctktWNlu%m5qTK5Koq4Ze;CG+TseG``G% z7g#|tP$o*BOr^DwdB_^Z)C8(C$GYCzu{Sg4N9h}tlEku%z0&%KR8snp`I{3FdgNY! ze|^@cYPaDUOOe%dN^oPh+i*qRzYilVNsc`UGQI=rYlBH#m3xetmf@WXRCOW-lxlUQ z*G2BcpLXfopv_(SQ9t(^)AI05eeJ*PcjLc-&e(9S8ta4$)DJ(%fPb0{f~owz^(>pI z8@GBm!&iD4M^L(84MI4%H9vngCgiN~?M_m~EF#&h(ZwCSu`OK!W&L)P6}1)^Xsl2= z!2g83>ZyZ1)`U!I;8^|>=j~N? zfKt3_4X;AoQuT6LfwwEai%!e=tslCZ=;#~hlvVYrZqGPnKRty-MWL}&kxo!g#jMjF z2*xQ!XBjx0+R#2@Cc{&qQ$mtH%i>PLlKGB6O*BWn{Lqe$uUTv@fD9Bc05o zU3$9yQ#$w1@!MXph#_DMo5cW#7r7eE#I%lrjSU|il@tzt%wn`nP#QX&jz^7x!#vF8 z=1NAx<-vyQ+Lu>j^>XAYBQYj3{ku9WwS1FSvQ|t6J^I*S(vXOU>6q-B3eSN5Xfq}= zpQ$`4DHmz#RPDCZUP165A^WTmF9zwk^*7KTmahB?fN{#$+UhsFi7)xPln?a@RJA*L z*Il;y_VtzVnFcL2W`5(;mNM}sGQP(enicsrcUJebEo&_1?pbv|JZ>h>eQdRJG2~S1o zS&qeupR;tN(`jS`!^+;oIvi1Re~ao%eFHs$uOAbC`6BrNLdmUGg6uvz7Ur>obRlcv zhU(GH=1MZ>_8ylF+9Qt#c-A|XSRx`(M4aUfHjJdn{AX%A7of$%WGRc$2T zc%5|T1YGYnm4>$2wRwG9N3MXH#adtc`sXj~E%Syzy8tMZv}V$y+58=CR;u-lQB~1m z!;gavvFODXM~MMugLK)dAi1P6t~T}m&-(elRr3E{27+s+Hs)9x{{v!k&9^Jw#L8S# zEoX0oO{kb0WG-C={(zz7ifO_lj)6aM5MNhV#_b3SWjw^>HV#kbwi(`9nXz`7wNYAS zZWnUix1H|O@76yz{?wctWfwp7a^Z0Af6l68m(L|UP0P8!e*T9RMAcc=qJGy&T*1g{ z4sA>xgTA=#FK;nnC#Gq%7tUh;QQn+m5Nkcqv{sr6T(Dpt++(i{#sm%v4G?0PN=TSA zgfg=e;s$5_LX8e7*C?~d#6O^dLsVjlz($lHf&(Oq^(7O4S^m==nxtm3h+P_sVdq4p z$CR7E1#0ytsZAQePhAnfi)OwKxrpF_f0+g{!((%BCN81{M-lvyDs%`eoA^6HV5;z0 zpGkG7@v{&ttP_qYHtMHSn2yG*Gd;zVM?qv~E|NB_ysXYKBr0t`Zww$haPHt!AWp)s zK;pyNtoMXkx9@ak6L%vJ;0aQp56P8P0Y55r3g})gpNO@v1gSlhXBZHX5R^i)7#m>% zk3Pf_xWXWz#Nmt52&g!zpU2*)3sF#3ZCIoP3OnHta?ocleEQFL7gPW(SgZVWf*3iza(m5BoH5C}~? z0~|UB6dG`HFzN9W@Jv+$T^5~4WaK0z=Vp|bjN)qCW^)OXDmjF%loI$G6`tTd;QgP) z4+UMpT%92@Ya+FG8^#o(Vba%w);oset>W}n5M$7Yv2kZUPZhk5KCao^ zZ$Y8ZhdnjVrcP{RPw`JF0t#a<-(G&w2uT7nVep9G!4pU#s_3`Pf9eze{{f+(TYDnB zr=7M_HTs}LksloGoJ*)mdlm@@ML8Z#vw(} zC<~UPJQgG7Ht16gMMqxGO!4%{7cr8>`f7Ej^?U2zKe5}k-&t(ux3>j21ow7cV>oTB zF?@8|33wR7Rmzg%N!h;*rhJa7|$Sc6tK6< zY66AFV6%pCaU9_ydSD*C%6o?7HxQIf5Tu8=Hem8jM@a46!zY_1sO*+m7fL3>NH`}3xilT(j{ zW4OHD(|q3-jzwFtr`3ktN!_T2l;L&$rvyU&n*hBN6W&`Q`-FM}i{nZhm6+=C?*x`= zjCq*fKrftTh*E$}g=bH5b|UG!;x*mG>3P*plP6pnR5C02!*&e~Zn`pQC7?2gcKm~+ zInQ(=#0@DMJ#2C_7%??`#PSrmAyq>RCLPG~N)Oz@b?bJM{$St-z)CG1S5Zs?RqkEvAx0*R5$oe6~1B@!Xi}ID1AEi=EMm=A0|Z zOr=sQ7*-rHxN|bmQ2psg9($hPzyAb|PP(^B^_9|%O_@t49Cey2vD^Qm#3|*N4?YP2 zJ!TUYv!-!(o`NK~C1M;Lhum{G`MOR?L^Nfu?vQkQ{8lr-OpBkZL+S3y33-MloB8K$ znj^EKm3u*$V0@56?^fQ89P7)DV$QgvMdqcwM%~tYs5H&ssd%Cl+*;;+?riomQGmG?W@ePX3~kpE^F?X z^WB^ub>*SD(B{A*18VMT!bt-d4E!zrVUl zmb%K3H9cXc)D{)FoxE*r|7Y#onOT)h5uRy_id-ia=}dn7@P@a?)0}g&I#nm0x_8y+ wq|KiH{s!KhIwA)a8SZ|%+Sfq}GvmNA5MCM-L%$5Rh<*^N@_B<- Date: Mon, 24 Aug 2026 13:18:28 +0100 Subject: [PATCH 22/34] Draw the real-time rule once, or not at all The band marking the two-times cutoff rendered three times in one table, each copy claiming everything below it was slower than real time. Two of the three were lies, and the rows between them were on both sides of the line. boundaryAfter tested each adjacent pair and drew a rule wherever the cutoff was crossed, which is correct only if the ordering separates the two groups. The comment above BOUNDARY_SORTS says exactly that -- "the only two where a line drawn at the cutoff means anything ... under any other sort the fast and slow engines interleave and the boundary is not a place" -- but the code took the sort column as proof of it rather than checking. Sorting by RTF separates them by construction. Sorting by compute per phrase did too, for as long as the two agreed on order, and Magpie ended that: RTF is compute per second of audio, so builds differing in how long they speak can order one way per phrase and the other way per second. Under the default sort the cutoff is now crossed three times, between q8_0 at 1.83x and q4_k at 2.01x, back under at q5_k at 1.98x, and over again at f16 at 2.13x. So contiguity is counted rather than assumed: exactly one crossing means the groups are contiguous and the rule is a place, more than one means they interleave and there is nothing to draw. Verified by sort -- none under compute per phrase, one under RTF, none under word error, where it never belonged. The article promised the line under both sorts, so it now promises it under RTF, says compute per phrase used to draw it, and says why it stopped. That is the same fact the cutoff paragraph already had to admit two commits ago: near the line, per-phrase and per-second compute no longer agree. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- .../content/benchmarks/open-weight-tts.md | 7 +++- .../src/components/charts/TtsResultsTable.tsx | 37 ++++++++++++++++--- frontend/src/data/benchmarks-posts.json | 2 +- 3 files changed, 38 insertions(+), 8 deletions(-) diff --git a/frontend/content/benchmarks/open-weight-tts.md b/frontend/content/benchmarks/open-weight-tts.md index af03da21..3c126abe 100644 --- a/frontend/content/benchmarks/open-weight-tts.md +++ b/frontend/content/benchmarks/open-weight-tts.md @@ -76,8 +76,11 @@ face value. that cannot keep up with their own speech, because "how far off is it" is a real question and a table that quietly omits the answer cannot be checked. The **Real time only** control above the table hides anything above 2x for readers who only care about what can be spoken -live, and sorting by RTF or compute per phrase draws a line at the cutoff so the two groups -are visible at once. +live, and sorting by RTF draws a line at the cutoff so the two groups are visible at once. +Compute per phrase used to draw it too, back when the two orderings agreed. Magpie's +quantisations differ in how long they speak, so per-phrase compute and per-second compute +now disagree near the line, and the rule is drawn only where the two groups are genuinely +contiguous rather than at every crossing. The line falls at 2x exactly, and two Magpie quantisations now sit either side of it: q5_k keeps up at 1.98x, q4_k does not at 2.01x. Three hundredths of a real-time factor separate diff --git a/frontend/src/components/charts/TtsResultsTable.tsx b/frontend/src/components/charts/TtsResultsTable.tsx index 4f9bbd58..a2036807 100644 --- a/frontend/src/components/charts/TtsResultsTable.tsx +++ b/frontend/src/components/charts/TtsResultsTable.tsx @@ -3,7 +3,7 @@ "use client"; -// REVISION: tts-results-table-v7-min-three +// REVISION: tts-results-table-v8-single-boundary // The open-weight TTS comparison: every configuration that keeps pace with real // time, and every row playable so the reader can hear the engine that produced // the numbers. @@ -21,7 +21,7 @@ import { useSamplePlayer } from "./useSamplePlayer"; import { useClassFilter } from "./useClassFilter"; import { useTtsScores } from "./useTtsScores"; -const MODULE_REVISION = "tts-results-table-v7-min-three"; +const MODULE_REVISION = "tts-results-table-v8-single-boundary"; if (typeof window !== "undefined") { console.log(`[tts-results-table] REVISION: ${MODULE_REVISION} loaded at ${new Date().toISOString()}`); } @@ -321,13 +321,40 @@ export function TtsResultsTable() { * not, or null where no rule belongs. Only shown when everything is listed - * with the filter on there is nothing on the far side to divide from - and * only under a sort where the two groups are actually contiguous. */ - const boundaryAfter = (i: number): string | null => { + /** The single index the rule belongs after, or null when the ordering does not + * actually separate the two groups. + * + * Contiguity used to be safe to assume from the sort column alone. Sorting by + * RTF separates them by construction, and sorting by compute per phrase did + * too, because the two agreed on order. Magpie's quantisations ended that: + * RTF is compute per second of audio, so builds that differ in how long they + * speak can order one way per phrase and the other way per second. Under the + * default sort the cutoff is now crossed three times, and a rule drawn at + * every crossing rendered the same band three times in one table, each one + * claiming everything below it was slower. + * + * So it is counted rather than assumed, which is what the comment above + * BOUNDARY_SORTS always said the rule was. Exactly one crossing means the + * groups are contiguous and the line is a place; more than one means they + * interleave and there is no line to draw. */ + const boundaryIndex = React.useMemo(() => { if (realtimeOnly || !sort || !BOUNDARY_SORTS.has(COLUMNS[sort.col])) return null; + const keepsUp = (r: (typeof rows)[number]) => + Number(r.cells[RTF_COL_RENDERED].sort) <= RTF_CUTOFF; + let only: number | null = null; + for (let i = 0; i + 1 < rows.length; i++) { + if (keepsUp(rows[i]) === keepsUp(rows[i + 1])) continue; + if (only !== null) return null; + only = i; + } + return only; + }, [realtimeOnly, sort, rows]); + + const boundaryAfter = (i: number): string | null => { + if (boundaryIndex === null || i !== boundaryIndex) return null; const next = rows[i + 1]; if (!next) return null; const here = Number(rows[i].cells[RTF_COL_RENDERED].sort) <= RTF_CUTOFF; - const there = Number(next.cells[RTF_COL_RENDERED].sort) <= RTF_CUTOFF; - if (here === there) return null; // Which side is which depends on the sort direction, so say it rather than // leaving the reader to infer it from a bare line. // The subject stays put and the direction flips. Naming whichever group diff --git a/frontend/src/data/benchmarks-posts.json b/frontend/src/data/benchmarks-posts.json index 3a7f421c..00a3951f 100644 --- a/frontend/src/data/benchmarks-posts.json +++ b/frontend/src/data/benchmarks-posts.json @@ -31,7 +31,7 @@ "slug": "limitations" } ], - "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nEvery row can be played, so you can hear what a word error rate actually sounds like.\n\n**You can run this yourself.** The harness, the corpus and the scoring are open, and\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this\npage from nothing. A single engine takes minutes\n(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it\nspent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template\ndoes the whole thing in one terminal and shows the report in the built-in browser.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for thirty-one of the forty-six\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones, so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the nineteen\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2x for readers who only care about what can be spoken\nlive, and sorting by RTF or compute per phrase draws a line at the cutoff so the two groups\nare visible at once.\n\nThe line falls at 2x exactly, and two Magpie quantisations now sit either side of it: q5_k\nkeeps up at 1.98x, q4_k does not at 2.01x. Three hundredths of a real-time factor separate\ntwo builds of one checkpoint that this table reports as otherwise indistinguishable, so near\nthe boundary the **Real time only** control is a rough divide rather than a verdict. It used\nto be a clean one, with OmniVoice at 1.51x on one side and Chatterbox Q4 at 2.03x on the\nother, and it is worth saying that it no longer is.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes and feed-forward over one,\nexcept the six Magpie quantisations, which are averaged over three. Six builds of a single\ncheckpoint differing only in precision need more samples than one run to say whether they\ndiffer at all -- and at three runs they still do not, spanning 6.5-7.4% mean word error\nwhile individual variants swing by up to 2.4% between runs. Differences under roughly two\npoints are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" + "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nEvery row can be played, so you can hear what a word error rate actually sounds like.\n\n**You can run this yourself.** The harness, the corpus and the scoring are open, and\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this\npage from nothing. A single engine takes minutes\n(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it\nspent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template\ndoes the whole thing in one terminal and shows the report in the built-in browser.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for thirty-one of the forty-six\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones, so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the nineteen\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2x for readers who only care about what can be spoken\nlive, and sorting by RTF draws a line at the cutoff so the two groups are visible at once.\nCompute per phrase used to draw it too, back when the two orderings agreed. Magpie's\nquantisations differ in how long they speak, so per-phrase compute and per-second compute\nnow disagree near the line, and the rule is drawn only where the two groups are genuinely\ncontiguous rather than at every crossing.\n\nThe line falls at 2x exactly, and two Magpie quantisations now sit either side of it: q5_k\nkeeps up at 1.98x, q4_k does not at 2.01x. Three hundredths of a real-time factor separate\ntwo builds of one checkpoint that this table reports as otherwise indistinguishable, so near\nthe boundary the **Real time only** control is a rough divide rather than a verdict. It used\nto be a clean one, with OmniVoice at 1.51x on one side and Chatterbox Q4 at 2.03x on the\nother, and it is worth saying that it no longer is.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes and feed-forward over one,\nexcept the six Magpie quantisations, which are averaged over three. Six builds of a single\ncheckpoint differing only in precision need more samples than one run to say whether they\ndiffer at all -- and at three runs they still do not, spanning 6.5-7.4% mean word error\nwhile individual variants swing by up to 2.4% between runs. Differences under roughly two\npoints are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" }, { "slug": "agent-skills", From 99a6d11a9b81f134c0114acb3e2fcdf78e5f5e23 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Mon, 24 Aug 2026 13:26:22 +0100 Subject: [PATCH 23/34] Move the real-time line to 2.05x, which puts the rule back on the default sort Two hundredths, and the whole reason is what lives inside them. Magpie q4_k at 2.01x, q6_k at 2.02x and Chatterbox Q4 at 2.03x sat just above a round 2, and the table's own notes say the Magpie ladder is not separable - 6.5-7.4% mean word error across six builds while individual variants swing by up to 2.4% between runs. Splitting three engines from measurements the table reports as indistinguishable is a distinction the data cannot carry. It also split them inconsistently. RTF is compute per second of audio, so builds that differ in how long they speak order one way per phrase and another per second: at 2 the cutoff was crossed three times under the default sort, and the rule marking it had nowhere honest to go, so the contiguity check suppressed it on the view most readers land on. At 2.05 the boundary is a place again. Chatterbox Q4 at 2.03x is the slowest that keeps up, Tada 3B at 2.13x the fastest that does not, and the gap is 0.093x where it had narrowed to 0.029x. One crossing under both boundary sorts, so the rule returns to compute per phrase without the contiguity check being loosened - it is still counted, it now simply passes. The cutoff lives in the run data rather than the component, so the filter's own label follows it: "Real time only (under 2.05x)". Thirty configurations keep up now rather than twenty-seven. The article says why the line is not round, rather than leaving 2.05 looking arbitrary. An unexplained odd number reads as a threshold picked to flatter something; the reason here is the opposite, and worth the two sentences. Verified by sort: one rule under compute per phrase, one under RTF, none under word error. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- .../content/benchmarks/open-weight-tts.md | 29 ++++++++++--------- frontend/scripts/import-tts-export.mjs | 16 +++++++++- frontend/src/data/benchmarks-posts.json | 2 +- .../benchmarks/open-weight-tts/2026-08.json | 4 +-- 4 files changed, 33 insertions(+), 18 deletions(-) diff --git a/frontend/content/benchmarks/open-weight-tts.md b/frontend/content/benchmarks/open-weight-tts.md index 3c126abe..cbf1dfbb 100644 --- a/frontend/content/benchmarks/open-weight-tts.md +++ b/frontend/content/benchmarks/open-weight-tts.md @@ -72,22 +72,23 @@ face value. **Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family. -**Nothing is filtered out.** Every configuration compared is listed, including the nineteen +**Nothing is filtered out.** Every configuration compared is listed, including the sixteen that cannot keep up with their own speech, because "how far off is it" is a real question and a table that quietly omits the answer cannot be checked. The **Real time only** control -above the table hides anything above 2x for readers who only care about what can be spoken -live, and sorting by RTF draws a line at the cutoff so the two groups are visible at once. -Compute per phrase used to draw it too, back when the two orderings agreed. Magpie's -quantisations differ in how long they speak, so per-phrase compute and per-second compute -now disagree near the line, and the rule is drawn only where the two groups are genuinely -contiguous rather than at every crossing. - -The line falls at 2x exactly, and two Magpie quantisations now sit either side of it: q5_k -keeps up at 1.98x, q4_k does not at 2.01x. Three hundredths of a real-time factor separate -two builds of one checkpoint that this table reports as otherwise indistinguishable, so near -the boundary the **Real time only** control is a rough divide rather than a verdict. It used -to be a clean one, with OmniVoice at 1.51x on one side and Chatterbox Q4 at 2.03x on the -other, and it is worth saying that it no longer is. +above the table hides anything above 2.05x for readers who only care about what can be +spoken live, and sorting by RTF or compute per phrase draws a line at the cutoff so the two +groups are visible at once. The rule is drawn only where the two orderings agree on which +side each engine falls; where they interleave it is left out rather than drawn at every +crossing. + +The line falls at 2.05x rather than a round 2x, and the two hundredths are the whole reason. +Magpie ships six quantisations of one checkpoint, and three engines land inside that gap: +Magpie q4_k at 2.01x, q6_k at 2.02x and Chatterbox Q4 at 2.03x. Drawn at 2x, those three +split from measurements the table itself reports as indistinguishable, and they split in a +different order depending on whether you sort by compute per phrase or per second of audio, +because the Magpie builds differ in how long they speak. Drawn at 2.05x the boundary is a +place again: Chatterbox Q4 at 2.03x is the slowest that keeps up, Tada 3B at 2.13x the +fastest that does not, and nothing sits inside a tenth of the line. **Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and long-form categories. diff --git a/frontend/scripts/import-tts-export.mjs b/frontend/scripts/import-tts-export.mjs index bffcf414..68cc0fee 100644 --- a/frontend/scripts/import-tts-export.mjs +++ b/frontend/scripts/import-tts-export.mjs @@ -503,7 +503,21 @@ const rows = rawRows.map(([, rowAttrs, inner]) => { const rtfCol = columns.indexOf("RTF"); const rtfOf = (r) => Number(r.cells[rtfCol].sort); -const RTF_CUTOFF = 2; +/** Where "keeps up with its own speech" is drawn. + * + * 2.05 rather than a round 2. Magpie ships six quantisations of one checkpoint + * and three of them - q4_k at 2.01x, q6_k at 2.02x - land in the two hundredths + * above a round two, along with Chatterbox Q4 at 2.03x. At 2 the table sorted by + * compute per phrase crossed the line three times, because RTF is compute per + * second of audio and those builds differ in how long they speak, so the rule + * marking the boundary had nowhere honest to go and was suppressed entirely. + * + * Moving the line two hundredths puts those three on the side their measurements + * already put them within run-to-run noise, leaves exactly one crossing under + * both boundary sorts, and brings the rule back to the sort readers land on. + * The engines it moves are the ones the table itself reports as indistinguishable + * from each other, so nothing is being reclassified that was ever separable. */ +const RTF_CUTOFF = 2.05; const run = { benchmark: "open-weight-tts", diff --git a/frontend/src/data/benchmarks-posts.json b/frontend/src/data/benchmarks-posts.json index 00a3951f..8cef1b33 100644 --- a/frontend/src/data/benchmarks-posts.json +++ b/frontend/src/data/benchmarks-posts.json @@ -31,7 +31,7 @@ "slug": "limitations" } ], - "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nEvery row can be played, so you can hear what a word error rate actually sounds like.\n\n**You can run this yourself.** The harness, the corpus and the scoring are open, and\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this\npage from nothing. A single engine takes minutes\n(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it\nspent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template\ndoes the whole thing in one terminal and shows the report in the built-in browser.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for thirty-one of the forty-six\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones, so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the nineteen\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2x for readers who only care about what can be spoken\nlive, and sorting by RTF draws a line at the cutoff so the two groups are visible at once.\nCompute per phrase used to draw it too, back when the two orderings agreed. Magpie's\nquantisations differ in how long they speak, so per-phrase compute and per-second compute\nnow disagree near the line, and the rule is drawn only where the two groups are genuinely\ncontiguous rather than at every crossing.\n\nThe line falls at 2x exactly, and two Magpie quantisations now sit either side of it: q5_k\nkeeps up at 1.98x, q4_k does not at 2.01x. Three hundredths of a real-time factor separate\ntwo builds of one checkpoint that this table reports as otherwise indistinguishable, so near\nthe boundary the **Real time only** control is a rough divide rather than a verdict. It used\nto be a clean one, with OmniVoice at 1.51x on one side and Chatterbox Q4 at 2.03x on the\nother, and it is worth saying that it no longer is.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes and feed-forward over one,\nexcept the six Magpie quantisations, which are averaged over three. Six builds of a single\ncheckpoint differing only in precision need more samples than one run to say whether they\ndiffer at all -- and at three runs they still do not, spanning 6.5-7.4% mean word error\nwhile individual variants swing by up to 2.4% between runs. Differences under roughly two\npoints are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" + "content": "\nWe tested all the open weight TTS models against the same corpus, using Whisper medium to transcribe back to text.\n\nEvery row can be played, so you can hear what a word error rate actually sounds like.\n\n**You can run this yourself.** The harness, the corpus and the scoring are open, and\n[talkito-benchmark](https://github.com/robdmac/talkito-benchmark) regenerates every number on this\npage from nothing. A single engine takes minutes\n(`python3 tools/benchmark.py --only piper`); the full sweep is around eleven hours, most of it\nspent downloading weights on the first run. In Orcabot, the **TTS Benchmark Runner** template\ndoes the whole thing in one terminal and shows the report in the built-in browser.\n\n```chart\ntts-vendors\n```\n\n## Results\n\nSorted by compute per phrase, fastest first. Click any column to re-sort.\n\n```chart\ntts-results\n```\n\n```chart\ntts-error-cost\n```\n\n```chart\ntts-preference\n```\n\n## How to read each column\n\nThe measurements are not interchangeable, and two of them actively mislead if taken at\nface value.\n\n- **RTF** is compute seconds per second of audio, the conventional measure. It **flatters\n any engine that emits excess silence**, because padding inflates the denominator. Prefer\n **x̄ synth**, which is compute per phrase: every engine speaks the same corpus, so it\n compares directly and cannot be gamed by padding.\n- **WER** is the round-trip word error rate, scored by Whisper `base.en`. The stronger\n `medium.en` was run too, but it did not finish for thirty-one of the forty-six\n configurations, so `base.en` is the one recogniser that covers the whole table. Being the\n weaker model it hallucinates words onto trailing silence, which understates good engines\n more than bad ones, so the true spread between engines is **wider** than this column\n shows, not narrower.\n- **PESQ** is a no-reference perceptual quality estimate from torchaudio's SQUIM, scored on\n the sample in the first column. It is a second axis word error cannot see, because word\n error saturates once speech is merely intelligible. It measures signal quality, not\n naturalness: 2E and Qwen3-TTS sit within 0.01 of each other while sounding clearly\n different.\n- **Class** describes the architecture. `det-ff` is a single forward pass with a\n deterministic duration predictor, so timing is identical every run and it cannot\n hallucinate. `st-ff` is the same shape with sampling inside, so output length varies.\n `ar-lm` samples audio tokens one at a time: length is emergent, cloning and emotion\n become possible, and real-time factor is floored by sequential decoding regardless of\n quantization.\n- **RSS** is peak resident memory, and is dominated by the runtime rather than the model. Engines served by the C++\n binary carry no interpreter; those running in Python carry torch, transformers and their\n dependency trees.\n\n## Methodology\n\n**Machine.** Apple M2, macOS. One machine, one English corpus, one recogniser family.\n\n**Nothing is filtered out.** Every configuration compared is listed, including the sixteen\nthat cannot keep up with their own speech, because \"how far off is it\" is a real question\nand a table that quietly omits the answer cannot be checked. The **Real time only** control\nabove the table hides anything above 2.05x for readers who only care about what can be\nspoken live, and sorting by RTF or compute per phrase draws a line at the cutoff so the two\ngroups are visible at once. The rule is drawn only where the two orderings agree on which\nside each engine falls; where they interleave it is left out rather than drawn at every\ncrossing.\n\nThe line falls at 2.05x rather than a round 2x, and the two hundredths are the whole reason.\nMagpie ships six quantisations of one checkpoint, and three engines land inside that gap:\nMagpie q4_k at 2.01x, q6_k at 2.02x and Chatterbox Q4 at 2.03x. Drawn at 2x, those three\nsplit from measurements the table itself reports as indistinguishable, and they split in a\ndifferent order depending on whether you sort by compute per phrase or per second of audio,\nbecause the Magpie builds differ in how long they speak. Drawn at 2.05x the boundary is a\nplace again: Chatterbox Q4 at 2.03x is the slowest that keeps up, Tada 3B at 2.13x the\nfastest that does not, and nothing sits inside a tenth of the line.\n\n**Corpus.** 84 phrases, spoken identically by every configuration, spanning core, edge and\nlong-form categories.\n\n**Scoring.** Round-trip word error rate, capped at 1.0 per phrase so a single runaway\ncannot swamp the mean. Leading silence is worth knowing about here: Whisper transcribes it\nas a word (\"You\", \"Thank you.\") rather than returning nothing, so an engine that pads the\nstart of a clip manufactures insertion errors and scores worse than it sounds. PESQ is\nscored on **trimmed** audio: scoring untrimmed rewards models that pad with silence, by up\nto 0.78, because silence pulls the estimate toward the ceiling and only mediocre audio has\nroom to rise.\n\n**Passes.** Autoregressive engines are averaged over two passes and feed-forward over one,\nexcept the six Magpie quantisations, which are averaged over three. Six builds of a single\ncheckpoint differing only in precision need more samples than one run to say whether they\ndiffer at all -- and at three runs they still do not, spanning 6.5-7.4% mean word error\nwhile individual variants swing by up to 2.4% between runs. Differences under roughly two\npoints are not resolvable at this sample size.\n\n### Limitations\n\nWord error rate measures intelligibility and nothing else. It is blind to naturalness,\nexpressiveness and speaker similarity, which is precisely what the LM-backed engines exist\nto provide, so this benchmark understates them by construction. Treat the table as a\nshortlist for \"will this be understood, and what will it cost me\", and the play buttons as\nthe part that answers \"does it sound any good\".\n\nOne machine also means the device-level findings generalise no further than Apple silicon.\n" }, { "slug": "agent-skills", diff --git a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json index 348da1d4..906ec9b9 100644 --- a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json +++ b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json @@ -4,8 +4,8 @@ "label": "2026-08-24", "machine": "Apple M2, macOS", "corpus": "84 phrases", - "rtfCutoff": 2, - "caption": "Sorted by compute per phrase, fastest first. Every configuration measured is listed, including the ones far too slow to keep up with their own speech; the filter above hides anything above 2x real time.", + "rtfCutoff": 2.05, + "caption": "Sorted by compute per phrase, fastest first. Every configuration measured is listed, including the ones far too slow to keep up with their own speech; the filter above hides anything above 2.05x real time.", "columns": [ "Sample · Configuration", "Class", From 76078f3844df9aeee3513ab33367ace7976a6bff Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Mon, 24 Aug 2026 13:55:18 +0100 Subject: [PATCH 24/34] Take the corrected Magpie class: autoregressive, not feed-forward All six were det-ff and are ar-lm. The class came from the export rather than from anything here - the importer only ever renames stoch-ff to st-ff - so this is upstream's correction, and it is the right one. NVIDIA's own card describes a causal transformer decoder that "autoregressively predicts discrete audio codec tokens", refined over stacked frames and decoded by NanoCodec. The table was already arguing against itself, which is the part worth keeping. This page defines det-ff as "a single forward pass with a deterministic duration predictor, so timing is identical every run and it cannot hallucinate" - and the Magpie rows are averaged over three free-seed runs precisely because they do not behave that way, spanning 6.5-7.4% mean word error while a single variant swings by up to 2.4%. A deterministic model has nothing to average. The corrected note now says so where a reader can check it: autoregressive despite the size, the GGUF declares a causal decoder with a codebook refinement head, and syntheses vary run to run. That is evidence rather than a label. Class counts move from det-ff 14 / st-ff 5 / ar-lm 27 to 8 / 5 / 33, and the six change colour in the preference chart, where they were the only large cluster painted as feed-forward. Nothing else moved: forty-six configurations, forty-six clips, the 2.05x rule still drawn once under the default sort. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- .../benchmarks/open-weight-tts/2026-08.json | 26 +++++++++---------- 1 file changed, 13 insertions(+), 13 deletions(-) diff --git a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json index 906ec9b9..a8973077 100644 --- a/frontend/src/data/benchmarks/open-weight-tts/2026-08.json +++ b/frontend/src/data/benchmarks/open-weight-tts/2026-08.json @@ -1451,8 +1451,8 @@ "href": "https://huggingface.co/nvidia/magpie_tts_multilingual_357m" }, { - "v": "det-ff", - "sort": "det-ff", + "v": "ar-lm", + "sort": "ar-lm", "tone": "", "align": "left", "restricted": false @@ -1552,8 +1552,8 @@ "href": "https://huggingface.co/nvidia/magpie_tts_multilingual_357m" }, { - "v": "det-ff", - "sort": "det-ff", + "v": "ar-lm", + "sort": "ar-lm", "tone": "", "align": "left", "restricted": false @@ -1653,8 +1653,8 @@ "href": "https://huggingface.co/nvidia/magpie_tts_multilingual_357m" }, { - "v": "det-ff", - "sort": "det-ff", + "v": "ar-lm", + "sort": "ar-lm", "tone": "", "align": "left", "restricted": false @@ -1754,8 +1754,8 @@ "href": "https://huggingface.co/nvidia/magpie_tts_multilingual_357m" }, { - "v": "det-ff", - "sort": "det-ff", + "v": "ar-lm", + "sort": "ar-lm", "tone": "", "align": "left", "restricted": false @@ -1956,8 +1956,8 @@ "href": "https://huggingface.co/nvidia/magpie_tts_multilingual_357m" }, { - "v": "det-ff", - "sort": "det-ff", + "v": "ar-lm", + "sort": "ar-lm", "tone": "", "align": "left", "restricted": false @@ -2057,8 +2057,8 @@ "href": "https://huggingface.co/nvidia/magpie_tts_multilingual_357m" }, { - "v": "det-ff", - "sort": "det-ff", + "v": "ar-lm", + "sort": "ar-lm", "tone": "", "align": "left", "restricted": false @@ -2135,7 +2135,7 @@ "restricted": false }, { - "v": "Smallest of six published quantisations at 516 MB. Across three free-seed runs the six span 6.5-7.4% mean WER while single variants swing up to 2.4% between runs, so the ladder is not separable: precision buys memory here, not accuracy.", + "v": "Smallest of six published quantisations at 516 MB. Autoregressive despite the size: the GGUF declares a causal decoder with a codebook refinement head, and syntheses vary run to run. Across three free-seed runs the six variants span 6.5-7.4% mean WER while a single variant swings up to 2.4%, so the ladder is not separable - precision buys memory, not accuracy.", "sort": "Smallest of six published quantisations ", "tone": "", "align": "", From 74a10794dcbc2c65c102b9ce3fe4a48a6719d081 Mon Sep 17 00:00:00 2001 From: Rob Macrae Date: Mon, 24 Aug 2026 14:09:15 +0100 Subject: [PATCH 25/34] Add a memory budget slider beside the class filters Peak RSS is the column most likely to decide whether an engine is usable at all, and until now it could only be sorted, not filtered. "What fits on the machine I have" was a question the table could answer only by reading down a column. The stops are spaced by ratio rather than evenly. RSS runs 246 MB to 8.6 GB, but twenty of the forty-six engines fit inside a gigabyte and only three sit above four, so an even track would spend most of its length separating the handful of largest models and almost none of it where the choosing happens. Twelve stops from 256 MB to no limit, and it starts at no limit so nothing is hidden until someone asks for less. A slider rather than a number field because the question is a budget being felt out against the count beside it, not a figure known in advance - the useful motion is dragging until the table stops containing what you wanted. The value is rendered as a size and repeated on aria-valuetext, since a range input otherwise announces "8", which says nothing; and the label has a minimum width so the filter row does not shuffle sideways while the handle moves. Verified against the distribution: 1 configuration under 256 MB, 5 under 512, 20 under 1 GB, 43 under 4 GB, 46 unfiltered. The cutoff rule behaves correctly underneath it too - it disappears once the filter leaves every remaining engine on one side of the real-time line, which is the contiguity check doing its job for a filter it knows nothing about. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_012WebP1vdtbLxbLjfWY9YGj --- .../src/components/charts/TtsResultsTable.tsx | 59 +++++++++++++++++-- 1 file changed, 54 insertions(+), 5 deletions(-) diff --git a/frontend/src/components/charts/TtsResultsTable.tsx b/frontend/src/components/charts/TtsResultsTable.tsx index a2036807..c49d6f55 100644 --- a/frontend/src/components/charts/TtsResultsTable.tsx +++ b/frontend/src/components/charts/TtsResultsTable.tsx @@ -3,7 +3,7 @@ "use client"; -// REVISION: tts-results-table-v8-single-boundary +// REVISION: tts-results-table-v9-memory-filter // The open-weight TTS comparison: every configuration that keeps pace with real // time, and every row playable so the reader can hear the engine that produced // the numbers. @@ -21,7 +21,7 @@ import { useSamplePlayer } from "./useSamplePlayer"; import { useClassFilter } from "./useClassFilter"; import { useTtsScores } from "./useTtsScores"; -const MODULE_REVISION = "tts-results-table-v8-single-boundary"; +const MODULE_REVISION = "tts-results-table-v9-memory-filter"; if (typeof window !== "undefined") { console.log(`[tts-results-table] REVISION: ${MODULE_REVISION} loaded at ${new Date().toISOString()}`); } @@ -42,6 +42,25 @@ if (RTF_COL < 0) throw new Error(`[tts] no "RTF" column in run ${run.run}: have const RTF_CUTOFF = run.rtfCutoff; const rtfOf = (r: Row) => Number(r.cells[RTF_COL].sort); +const RSS_COL = run.columns.indexOf("RSS"); +if (RSS_COL < 0) throw new Error(`[tts] no "RSS" column in run ${run.run}: have ${run.columns.join(", ")}`); +const rssOf = (r: Row) => Number(r.cells[RSS_COL].sort); + +/** Stops for the memory filter, in MB, with no limit as the last one. + * + * Spaced by ratio rather than evenly. Peak RSS runs 246 MB to 8.6 GB, but + * twenty of the forty-six engines fit inside a gigabyte and only three sit above + * four, so an even track would spend most of its length separating the handful + * of largest models and almost none of it where the reader is actually choosing. + * + * A slider rather than a box to type in: the question this answers is "what fits + * on the machine I have", which is a budget being felt out against the count, not + * a number known in advance. It starts at no limit so the table is unfiltered + * until someone asks for less. */ +const RSS_STEPS = [256, 384, 512, 768, 1024, 1536, 2048, 3072, 4096, 6144, 8192, Infinity]; +const fmtMB = (mb: number) => + mb === Infinity ? "any" : mb >= 1024 ? `${Math.round((mb / 1024) * 10) / 10} GB` : `${mb} MB`; + /** The Human column is spliced in here, right after the engine name, so the * live number is visible without scrolling the table sideways. */ const HUMAN_COL = 1; @@ -162,6 +181,8 @@ export function TtsResultsTable() { const [sort, setSort] = React.useState<{ col: number; dir: "asc" | "desc" } | null>(DEFAULT_SORT); const [realtimeOnly, setRealtimeOnly] = React.useState(false); + const [rssStep, setRssStep] = React.useState(RSS_STEPS.length - 1); + const rssLimit = RSS_STEPS[rssStep]; // Shared with the charts: one class selection governs the whole page, so // filtering to ar-lm in the table narrows the plot below it too. const { active: classFilter, toggle: toggleClass, clear: clearClasses, shows: classShows } = useClassFilter(); @@ -212,6 +233,7 @@ export function TtsResultsTable() { const visible = ROWS.filter( (r) => (!realtimeOnly || rtfOf(r) <= RTF_CUTOFF) && + rssOf(r) <= rssLimit && classShows(classOf(r)) ); const withHuman = visible.map((r) => { @@ -237,7 +259,7 @@ export function TtsResultsTable() { typeof ka === "number" && typeof kb === "number" ? ka - kb : String(ka).localeCompare(String(kb)); return sort.dir === "asc" ? cmp : -cmp; }); - }, [classShows, ratings, realtimeOnly, sort]); + }, [classShows, ratings, realtimeOnly, rssLimit, sort]); // asc -> desc -> back to the default, never to "unsorted". Unsorted rendered // the export's own row order, which is neither the default the reader arrived @@ -458,15 +480,42 @@ export function TtsResultsTable() { ); })} + {/* Memory budget. The value is shown as a size rather than a step index, + and carried on aria-valuetext for the same reason: "8" tells a reader + nothing, and the count beside it only says how many survived, not what + they survived. minWidth holds the label steady so the row does not + shuffle sideways while the handle is being dragged. */} + {rows.length === ROWS.length ? `${ROWS.length} configurations` : `${rows.length} of ${ROWS.length} configurations`} - {(realtimeOnly || classFilter.size > 0) && ( + {(realtimeOnly || classFilter.size > 0 || rssLimit !== Infinity) && (