Repository navigation
share: a clearer board card that names its repo - #89
Merged
Merged
Conversation
The card starts with what the board is and how to read it, calls the rating a rating, links model names that are Hub repo ids, counts error outputs in the warning note, and shows slice labels as code. The full settings stay in board.json. A collapsed section has DuckDB queries. plan_share takes board_repo; board.json records it and push_share refuses another destination. The private-id check skips the board's own whole repo id only. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016isSPowmvG84onwTGWiNoh
…le size - Size: models the registry does not know get the Hub's safetensors parameter count (hub_model_sizes, passed by the CLI). - A private-id mention is allowed only inside a whole board-id match; an overlapping mention is refused (Codex P1). - The board repo must be an ASCII Hub repo id, also in a dry run (P2). - The card uses the drawn sample size from model_coverage, else shows --max-samples as a cap (P2). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016isSPowmvG84onwTGWiNoh
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016isSPowmvG84onwTGWiNoh
- uv.lock: restore main's lock file. 69fa50b committed a lock that a local uv run had rewritten to an internal package mirror. - Card copy: ties, interval overlap, error outputs "in the sample", pages in no board, the judge criteria in order, a load_dataset hint and a token note for private boards. - Size: 10B and larger keep three significant digits; the Size cell is escaped. - CLI: a note when the board repo name contains a private repo id. - Tests: markup-shaped model names, the FAILED note, no sample cap, a slice column with markup, size formats. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016isSPowmvG84onwTGWiNoh
davanstrien
commented
Nov 9, 2026
…o_id - The card links the judge prompts (judge.py at the release tag) and no longer paraphrases the criteria: a judge reason on the v0 board repeated " running headers and page number", and the page-text check refused the card in a real dry run. - Size: the Hub's safetensors count for every public model on the Hub (read without a token), the registry's size otherwise, so sizes are measured the same way. The card says what Size counts. - Board repo: huggingface_hub's validate_repo_id plus a namespace and no surrounding whitespace. - The error-output note says the count is over the source dataset; the comparisons line says the auto-ties are included. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016isSPowmvG84onwTGWiNoh
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016isSPowmvG84onwTGWiNoh
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016isSPowmvG84onwTGWiNoh
…II board id Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016isSPowmvG84onwTGWiNoh
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016isSPowmvG84onwTGWiNoh
Merged
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Makes the board card that
ocr-bench sharewrites easier to read and to use, after a first look at the v0 government-documents board.Card
baidu/Unlimited-OCR-2 (tiled)links the id and keeps the label as text.board.json.load_datasetlines and a collapsed DuckDB block (overall board, top model per label) that name the board repo.Behaviour
plan_share(..., board_repo=)is required;board.jsonrecords it asrepo.push_sharerefuses another destination. A board repo that is the results repo or the source is refused when the plan is built, so a dry run never writes it.me/scans-boardfor sourceme/scans). An overlapping mention (public/me+ aliaspublic/me/private-source) is refused.hub_model_sizes, passed by the CLI). Not on the Hub, gated, or no safetensors: no size.model_coverage; without it,--max-samplesis shown as a cap.Reviews
Codex and Fable reviewed 3068542. Codex: board-id stripping could hide an overlapping private id (P1), board names were not fully validated (P2), the sample cap was shown as the sample size (P2). All are fixed in 69fa50b with tests. Fable's copy and test-gap notes are addressed in the commits after that. Fresh reviews on the final head will follow.
Checks
pytest(1012+ tests),ruff,ty,rumdlpass. The DuckDB queries were run against the private v0 board.🤖 Generated with Claude Code
https://claude.ai/code/session_016isSPowmvG84onwTGWiNoh