AIPerf stage 6: qualify all TRTMC text-to-image models #1673

Request type

New capability

Problem and use case

Users need an image-generation API benchmark whose response is the real native artifact and whose timing and quality describe the same fixed work.

Proposed outcome

Qualify all supported models in this group for one text-to-image generation workload through the existing native serving entrypoint.

• Workload: One prompt produces one image at one fixed resolution and explicit step/guidance configuration.
• Client / endpoint: AIPerf image_generation; POST /v1/images/generations.
• Rollout / prerequisites: Model-group rollout after stage 5.

Public CI hardware failures — owner-approved handling for stages 5–8

Keep existing model contracts, references, numerical thresholds, test eligibility and repository merge-gate rules unchanged. If an in-scope model fails non-required public CI due to OOM or a demonstrated hardware limitation, execute the affected unchanged correctness case locally on capable hardware, retain a non-skipped passing result, and document the public failure and local verification in the PR. Merge only once the unchanged required merge checks pass on the current head and the PR otherwise satisfies the repository rules. A numerical/model-correctness failure is not a hardware exception. Do not add test exclusions or manufacture a passing public result. Deliver stages 5, 6, 7 and 8 sequentially, starting each implementation after the previous stage merges.

Existing-model scope — owner clarification

This epic integrates AIPerf with models and model specs already supported on upstream main. “Whole group” means every existing model/spec for this workload, rather than adding checkpoints that an architecture might also support.

• Reuse existing checkpoint identities, normal full-model profiles/manifests, build settings, references and correctness thresholds. Do not add model families, checkpoints, model specs/manifests, precision/parallel variants or new numerical gates to enlarge the group.
• HTTP/media adapters, AIPerf input/workload configuration and regression fixes needed for existing models are in scope. Keep model semantics family-owned and existing correctness criteria unchanged.
• Missing full-model implementations or native audio/video capabilities are recorded as unsupported/outside this integration. Do not implement a new encoder, model graph, temporal path or native Task to fill the inventory. L0/partial profiles do not require full-model onboarding in this epic.
• Refresh against current main only to include independently existing supported specs. Architectural compatibility, model cards and upstream model availability do not establish an in-scope TRTMC model.

Group-wide scope and completion (revised)

Cover every TRTMC-supported checkpoint in this model group for the workload below, not just one pilot. Refresh the inventory against github/main at implementation and before completion. Use family-owned manifests, actual build/runtime behavior, and public Task capabilities together; a model name, broad support declaration, or an inherited C++ interface alone does not establish the checkpoint's supported workload.

Use one normal full-model profile per distinct checkpoint and workload. Record all additional precision/quantization/TP/CP variants separately; do not count variants as extra models or claim untested configurations. L0/partial/random/regression smoke profiles do not establish full-model coverage. Checkpoints with only partial implementation remain explicit gaps, not silent omissions. Only independently existing supported specs on main join the refreshed inventory; do not add models or specs as part of this integration.

• Publish the exhaustive per-checkpoint inventory with checkpoint/profile/revision, native Task, supported controls, and any explicit capability gap; reconcile it with current main before closing.
• Every supported checkpoint passes its unchanged owning-family correctness criteria on the exact served bundle, followed by a real published pip-installed AIPerf run through trtmc-server for this one workload.
• Check meaningful complete output contents and native/HTTP identity for every model, with separately marked warmup, failures, client/Task timing, and workload-appropriate throughput.
• Publish a result row and reproducible command for every checkpoint. A pilot, source-only check, HTTP fixture, skipped test, unrun model, unsupported adapter, or unresolved failure cannot close this group-wide task.
• Preserve family ownership and existing quality criteria; fix real defects without introducing model behavior in shared serving code or relaxing gates.

Implementation scope:

1. Reuse the existing generation route and artifact recorder, connecting trtmc-server to the public image-generation Task.
2. Return decodable images that remain accessible to the client; honor or explicitly reject generation options.
3. Record actual dimensions, steps, guidance, seed, effective precision, and output count; reuse existing quality evaluation on these outputs.

Acceptance criteria for every model in the group:

• The actual bundle and returned artifacts satisfy the unchanged owning-family correctness/quality criteria.
• A real pip-installed AIPerf run produces valid images for the fixed prompt corpus with correlated artifacts and visible errors.
• Report image latency and images/sec separately from Task-call wall time with actual generation settings and artifact lifecycle documented.
• Reuse existing serving/qualification mechanics; keep all model semantics, orchestration, and correctness family-owned, with capability-based Task dispatch.
• Account for inputs, outputs/artifacts, failures, and warmup by request identity. Keep client end-to-end latency separate from public_task_call_wall; exclude build/load/startup/warmup from steady-state timing.
• Record tested package versions, checkpoint/profile, actual bundle, dataset, endpoint/mode, hardware, effective precision, settings, commands, and public evidence. Distinguish planned, implemented, family-validated, and AIPerf-qualified status.
• Provide one concise reproducible example using published pip-installed AIPerf and links to existing build guidance; an AIPerf source checkout is not a runtime dependency.

Model and target details

Inventory baseline: TRTMC fd79ad26ca71a11b3ec086a7daaaead03e6c933f. Candidate models/profiles for this group:

Use one text-to-image workload across every supported image generator, with each model's explicit supported resolution/steps/guidance recorded. Do not merge video generators into this group because they share an image_generation string.

Use the public native Task runtime path. Record the actual supported target GPU and software stack during implementation/qualification; no new hardware validation or benchmark is claimed by this planning issue. Keep existing numerical criteria unchanged.

Alternatives considered

Reuse apps/perf_serving protocol/media/recording mechanics and apps/aiperf_qual qualification instead of adding a third server or benchmark harness. Deliver this stage on its own short-lived branch from current github/main.

Follow-ups outside this initial stage: Other resolutions/configurations, output batching, and additional model profiles are separate experiments.

Additional context

• Parent epic: Epic: expand TRTMC AIPerf model and workload coverage #1667
• Pilot profile
• AIPerf docs/tutorials/image-generation.md
• Native server
• Existing qualification harness

Research baseline: TRTMC 0bccb5e93af0e5ad346aaf44e404396f54b2e7de and AIPerf e1daadd10faa395100384358131d784d6858c5ae; published AIPerf 0.13.0. Recheck the tested published version and native capabilities when implementing. Existing source/fixtures/reference-only runs do not establish end-to-end TRTMC qualification. Keep internal logs, artifacts, and runner details private.

Submission checks

• I searched open and closed issues and found no duplicate request.
• I removed secrets, private/internal evidence, personal paths, and restricted artifacts.

Group completion policy revised at the owner's request: all supported checkpoints in this group must be accounted for and qualified; one successful pilot is insufficient. Existing pilot results are retained as historical evidence.

Completion

Merged PR #1787 as 9a8b545f5e8c22c653c8029c95a999dc070b4f3f from e16c79cbefe3adeef561bb704d2dbf92c8ed075a. All seven existing full checkpoints passed unchanged owning-family model criteria and actual published AIPerf qualification; the PR contains every timing row, profile, settings, revisions and reproduction commands. Required exact-head internal CI passed before normal squash merge. Optional experimental Dev CI was still running at merge and is not claimed as passed. No model spec, reference, threshold, eligibility or repository merge rule changed.

The original combined FLUX.2 BF16 selector OOM remains recorded; the exact retained bundle passed the original native/reference/comparison criteria in a fresh process (one executed case, no skips), then real AIPerf. Stable Diffusion’s weak existing low-resolution/ten-step visual baseline is explicitly documented.