Conflict in request_adapter.py with #2001, which adds continuation_video and
materializes it through _materialize_video. This branch replaced that helper
with materialize_clip, so continuation_video now goes through materialize_clip
as well, with the same "continuation" upload name. As a result it gets the same
policy as the other video fields: media_roots, the private-URL check, the size
cap, and base64 data:video/... URLs.
- docs/design/server_contracts/openai.md: continuation_video is listed in the
video row of the media-source table.
- docs/inference/fasth3_omniref.md: continuation_video may be a data:video URL.
- test_openai_serving_engine.py: an inline continuation video is decoded,
tracked and removed, and a data:audio one is refused.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Picks up #2059, whose pytest 9 fixes clear the fastcheck Unit lane
(test_lora_side_branch.py setup errors and the FA4 JIT-cache test) that
failed on this branch, and #2045, which fixes the docs-site Prettier
failure behind the Deploy Documentation build. No conflicts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Conflict in request_adapter.py: keep main's Open-H3-Regenerate-2K constants
and this branch's _MEDIA_SUFFIXES tables, which replace _VIDEO_SUFFIXES.
Open-H3-Regenerate-2K arrived assuming data: video references are refused.
With this branch they are decoded, so:
- test_openai_open_h3_regenerate_2k.py: the refused case is now a data URL
whose media type does not match, and an inline draft is decoded, reaches
the pipeline as video_path, and is removed after the request.
- The serving YAML comment and docs/inference/open_h3_regenerate_2k.md now
say data:video URLs are decoded up to server.media_reference_max_bytes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`POST /v1/videos` decoded `data:image/...` references but refused
`data:` video references and passed every `audio_url` to the pipeline
unchanged, so clients without a public URL could not send a reference
video or audio file (SGLang Diffusion accepts all three).
- `video_reference` / `video_path` / `video_url` accept
`data:video/...;base64,` and `audio_reference` accepts
`data:audio/...;base64,`. Each decoded file is capped by
`server.media_reference_max_bytes` (checked from the payload length
before decoding), written under `output_dir/uploads`, kept in field
order, and deleted when the job ends.
- Invalid base64, a non-base64 data URL, or a media type that does not
match the field returns HTTP 400. Percent escapes and line-wrapped
payloads decode, as in SGLang.
- Audio references are now materialized like video references: http(s)
URLs go through the media policy download (public addresses, byte
cap) and local paths must sit under `server.media_roots`. Before, the
raw string reached PyAV, which opened any local file or URL.
- `docs/design/server_contracts/openai.md` lists the accepted forms per
field.
Fixes#2018
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The unit lane fails on main in two contract tests. All three failures are
bookkeeping that drifted from the code, not runtime bugs:
- `test_env_policy.py::test_env_doc_table_matches_registry`: the generated
table in docs/contributing/env_vars.md is stale for
FASTVIDEO_MLX_QUANT_GEMM. Regenerated with the documented command
(`python fastvideo/tests/contract/test_env_policy.py`).
- `test_env_policy.py::test_env_access_follows_policy`: two new violations.
- fastvideo/tests/mlx/test_mlx_affine_dq_gemm.py wrote
FASTVIDEO_MLX_QUANT_GEMM through monkeypatch.setenv; now uses the
registered field's override() through the env_overrides fixture.
- fastvideo/tests/train/models/test_load_kandinsky5.py read
FASTVIDEO_ATTENTION_BACKEND with `not in os.environ`; now uses
envs.FASTVIDEO_ATTENTION_BACKEND.is_set() and override(). Its entry in
KNOWN_VIOLATIONS is deleted, as the shrinking-list rule requires.
- `test_schema_parity_inventory.py::..._extension_fields_are_classified`:
MiniMaxH3PipelineConfig declares video_scheduler_shift and
audio_scheduler_shift (training-time noise clocks rejected by
check_pipeline_config at inference); both are added to
pipeline_config_extensions.internal_only.
Test plan (CPU, no GPU required):
python -m pytest fastvideo/tests/contract/test_env_policy.py \
fastvideo/tests/api/test_schema_parity_inventory.py \
fastvideo/tests/mlx/test_mlx_affine_dq_gemm.py -q
# 45 passed
python -m pytest fastvideo/tests/contract/ fastvideo/tests/api/ -q -m "not gpu"
# 501 passed
MultiprocExecutor created two streaming multiprocessing queues for every
generator, but only StreamingVideoGenerator uses them. Each queue uses POSIX
named semaphores in /dev/shm. When a job epilog deleted these files, spawned
workers failed to start. The executor now creates the queues only when
FastVideoArgs.enable_streaming_ipc_queues is set. StreamingVideoGenerator sets
it on a copy of its arguments in queue mode. Standard generation creates no
streaming semaphores.
For n > 1, the image endpoints returned one preview grid instead of n images.
The endpoints now run one model batch, save each decoded sample as its own file,
and return one URL or base64 item for each image. This applies to generations
and edits.
A keyword-only return_samples field (default False) is added to OutputConfig,
SamplingParam and ForwardBatch. Batched image requests set it and do not build
preview frames. The worker keeps the sample tensor for these requests, and
non-output ranks clear the field. Defaults and positional arguments do not
change.
Adds opt-in regional torch.compile to the YAML-driven modular Wan trainer (training.model.enable_torch_compile / torch_compile_kwargs) with FA4 and Wan modulation forward/backward as opaque custom ops, compiled repeated DiT blocks after FSDP setup, and activation checkpointing applied before FSDP for trainable Wan transformers.
Merged with main's #1716 (WanModel resolves the checkpointing type in __init__; the PR's pre-FSDP wrap uses it and the post-load wrap is gone) and #1757 (pre_fsdp_model_transform runs before this PR's pre_fsdp_transform). Maintainer review (Swarm order #776): AdaLN skip names are canonicalized through checkpoint-wrapper prefixes, test_inference_regional_compile.py joins the unit lane, the Wan modulation custom-op test moved to the transformer lane, regional_compile is classified in the schema inventory, and a 1/2-GPU training smoke covers fully_shard(CheckpointWrapper) in eager and regional-compile modes. Validated on GB200: the FA4 CuTe forward/backward parity and fullgraph-backward tests and all four smoke cases pass.
Also exempts MMAudio recipes, whose student builds its pipeline config at init, from the fine-tuning recipe construction check that #1648 left failing on main.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
