nodes/hao-ai-lab-FastVideo
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
1,195 Commits 344 Branches 8 Tags
d866c2d5077409cf36d320377a17dc0440fcd583
Commit Graph
304 Commits
Author SHA1 Message Date
Aryan KumarandAryan Kumar c845a8dee2 [feat] Per-request community LoRA in the OpenAI video API (#2026) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-04 19:44:47 -08:00
tianwei-moandClaude Opus 5.5 36dc971dc6 [feat] MiniMax-H3 Ref2VA: continue an earlier video from its last frames (#2001) 
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
 2026-11-05 01:24:11 +00:00
tianwei-moandClaude Opus 5.5 229c43c8d3 [feat] MiniMax-H3 Ref2VA SFT: train each clip at its own canvas and frame count (#2003) 
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
 2026-11-05 01:12:34 +00:00
Haoyu Cheng 9e1915ceb4 [bugfix] Honor output_video_name when selecting output filenames (#2044) 2026-11-04 23:25:14 +00:00
Aryan KumarandAryan Kumar 661d18ba55 [docs] Pika H3 Flex cookbook: --weights for the NVFP4 QAD releases and FP4 attention (#2054) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-03 01:07:54 -08:00
Aryan KumarandAryan Kumar 4b4df1a94e [perf] Fused Q/K preconditioning + NVFP4 quantize for FA4 NVFP4 attention (bit-identical) (#2052) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-03 01:07:11 -08:00
Aryan KumarandAryan Kumar 7066408fb3 [feat] Pika H3 Flex NVFP4 QAD training: calibration, QAD, FP4 attention and export (#2051) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-03 01:04:26 -08:00
Aryan KumarandAryan Kumar fcec554456 [feat]: shared FastH3 NVFP4 FP4-attention QAD recipe (OmniRef + V2 T2AV) (#1989) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-03 00:54:28 -08:00
Aryan KumarandAryan Kumar 181725e3fd [docs] Fix cookbook component selection and memory reporting (#2045) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-03 00:54:18 -08:00
Aryan KumarAryan KumarAryan
 		4f80c5a3da [feat] Open-H3-Regenerate-2K: regenerate FastH3 QAD, Trim and V2 drafts at 2K (text to 2K 3.1-3.4x faster) (#2037) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
Co-authored-by: Aryan <aryan@aryan5v.dev>
 2026-11-02 03:44:28 -08:00
Aryan KumarandAryan Kumar b0047cfedc [feat] Open-H3-Regenerate-2K: NVFP4 H3 video decoder (decode 23 s to 5 s at 2K) (#2036) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-01 22:51:41 -08:00
Aryan KumarandAryan Kumar ac81198d33 [new-model] Open-H3-Regenerate-2K: 768P H3 draft to 2560x1440, served on the OpenAI video routes (#2035) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-01 22:45:09 -08:00
Aryan KumarandAryan Kumar cb02062930 [new-model] Add Pika H3 Flex inference support (#2038) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-01 22:33:10 -08:00
Aryan KumarandAryan Kumar d954fad9ec [perf] MiniMax H3: faster FP4 attention padding, uint8 VAE pixel path and cuDNN FP4 GEMMs (#2046) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-01 22:23:10 -08:00
Aryan KumarandAryan Kumar 166b9b8a07 [perf] MiniMax H3: fuse VAE decoder QK RMSNorm and partial RoPE (#2041) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-01 22:15:35 -08:00
Aryan KumarandAryan Kumar 7e1ed5e28e [perf] MiniMax H3: decode eight VAE spatial tiles per call by default (#2040) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-01 22:06:23 -08:00
Aryan KumarandAryan Kumar cb6a9a13e1 [bugfix] FA4 JIT cache vs tuning env, MXFP8 import and scale dtype, H3 VSA gate double quantization (#2049) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-01 19:11:25 -08:00
Aryan KumarandAryan Kumar 0d1244a869 [perf] MiniMax H3: opt-in faster kernel tier for the DiT fused ops (#2042) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-11-01 15:22:09 -08:00
Aryan KumarandAryan Kumar 4d0a10d989 [docs] Fix the docs build, cookbook memory status, and broken commands (#2023) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-10-30 17:08:41 -07:00
Aryan Kumar 6743f4b4b7 Revert "[new-model] Add PAN2 text-to-video and image-to-video" (#2025) 2026-10-27 18:03:59 -07:00
William Linandaryan5v a0e6cf1f50 [new-model] Add PAN2 text-to-video and image-to-video (#1961) 
Co-authored-by: aryan5v <email.aryan1005@gmail.com>
 2026-10-25 20:20:00 -07:00
Loay RashidandAryan Kumar 0d05b09c57 [new-model] Add initial Qwen-Image-2.1 inference (#1918) 
Co-authored-by: Aryan Kumar <156166681+aryan5v@users.noreply.github.com>
 2026-10-25 17:28:49 +00:00
Kyle HuandAryan Kumar a92c5c1578 [docs]: add spark4 CI backend routing (#2014) 
Co-authored-by: Aryan Kumar <aryank@Aryans-Mac-Studio.local>
 2026-10-24 16:29:44 -07:00
Aryan KumarandAryan Kumar dfc5550c35 [ci] Add FA2-enabled Spark CI image and GB10 reference setup (#2017) 
Co-authored-by: Aryan Kumar <aryank@Aryans-Mac-Studio.local>
 2026-10-24 16:16:25 -07:00
posiposi443-opsandAryan Kumar 8db2d71ef2 [perf] Cache MLX H3 text refinement and speed up attention (#2002) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-10-23 12:40:47 -07:00
Aryan Kumar 9bbdd18548 [ci]: Fix env-policy and schema-parity contract tests 
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
 2026-10-23 11:08:37 -07:00
Aryan KumarandAryan Kumar 7c2b447264 [ci] Run single-GPU CI lanes on a DGX Spark runner; weekly full suite on main (#2007) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-10-22 13:44:55 -07:00
William Lin e136c2b7f8 [infra]: fix the DGX Spark (GB10) Docker image: importable flashinfer, sm_121a kernels, native-matching output (#1986) 2026-10-22 11:59:31 -07:00
Aryan KumarandAryan Kumar fa2d1324c6 [feat]: FastH3 Omni (Ref2VA) on Apple Silicon MLX (#2000) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-10-21 12:01:54 -07:00
William Lin cdda1b7535 [feat] Add H3 PDD/DMPDD training and packed Ref2VA preprocessing (#1984) 2026-10-20 11:15:57 -07:00
Aryan KumarandAryan Kumar 089a5a55a0 [feat] Cookbook serving, hardware catalog, and landing page (#1999) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-10-20 09:10:29 -07:00
Yixing Wangleoaryan5vAryan Kumar
 		d4f144581d [feat] Add capacity metadata to existing cookbook model manifests (#1983) 
Co-authored-by: leo <yixingwang@YIXINGs-MacBook-Pro.local>
Co-authored-by: aryan5v <email.aryan1005@gmail.com>
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-10-20 08:24:19 +00:00
Ishan 5f96664abf [feat] Added a single-GPU VRAM calculator for cookbook deployments (#1978) 2026-10-20 08:12:06 +00:00
Aryan KumarAryan KumarClaude Opus 5.5
 		376e445f0e [feat]: FastH3 OmniRef inference on consumer devices (RTX 5090 / PRO 6000 / DGX Spark / 4090) (#1980) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
 2026-10-19 14:57:36 -07:00
Aryan KumarAryan KumarSolitaryThinker
 		2099dd5533 [feat]: Community MiniMax-H3 LoRA loading with quantized side branch (#1973) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
Co-authored-by: SolitaryThinker <wlsaidhi@gmail.com>
 2026-10-19 14:29:23 -07:00
lpc0220rootClaude Opus 5.5root
 		3658b120da [bugfix] Fix tests that fail on main: FastH3 Trim Spark audio, test isolation (#1957) 
Co-authored-by: root <root@ptyche0124.ptyche.clusters.nvidia.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: root <root@ptyche0040.ptyche.clusters.nvidia.com>
 2026-10-18 16:07:59 -07:00
lpc0220androot d4ed61cc7c [refactor] VSA: rename the sm100a block-sparse kernel to ptx_blackwell (#1956) 
Co-authored-by: root <root@ptyche0124.ptyche.clusters.nvidia.com>
 2026-10-17 21:10:19 -07:00
William Lin 377c79fa7e [docs] Add William Lin to the contact page (#1977) 2026-10-17 20:52:55 -07:00
Aryan KumarandAryan Kumar 25940a7e86 [docs] Unify cookbook model editors and launch commands (#1970) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-10-16 17:26:51 -07:00
Aryan Kumarsy3535Aryan Kumar
 		3f45f2da21 [bugfix] Repair DreamVerse and failing CI gates (#1958) 
Co-authored-by: sy3535 <sy3535@maxwell.poly.edu>
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-10-16 17:15:38 -07:00
Aryan KumarandAryan Kumar 6212e7262e [docs] Migrate documentation and cookbook to Blume (#1969) 
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-10-16 16:09:04 -07:00
William Lin ba371e33ed [ci] Stop automatically marking issues and PRs stale (#1964) 2026-10-15 01:05:48 -07:00
William Lin 06c8c5f4b8 [docs] Add public contact page (#1959) 2026-10-15 00:26:44 -07:00
William Lin c535797e3d [skill] Remove the add-model and decompose-pipeline-pr agent skills (#1963) 2026-10-15 00:10:49 -07:00
William Lin 59fddb00f2 [ci] Remove automated community comments (#1962) 2026-10-14 23:40:45 -07:00
William Lin 9f59519959 [misc] Remove DreamX World, HunyuanGameCraft, GLM-Image, LongCat, TurboDiffusion, and Z-Image (#1926) 2026-10-14 23:16:32 -07:00
Yixing WangleoAryan Kumar
 		d5253fd710 [feat] Add serving cookbook recipes and configuration API (#1941) 
Co-authored-by: leo <yixingwang@YIXINGs-MacBook-Pro.local>
Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>
 2026-10-09 15:49:08 -07:00
Davide Locatelli e1e2559359 [perf] FastH3 OmniRef: lossless speed-ups behind default-off switches (#1930) 
These speed-ups for the MiniMax-H3 pipelines keep the output bit-identical. Each
one has a switch that is off by default:

- FASTVIDEO_MINIMAX_H3_EXACT_KERNELS: Triton RoPE, AdaLN modulation, gated
  residual and SwiGLU kernels that round to bf16 at the same points as the
  eager ops.
- FASTVIDEO_H3_VSA_HEADS_FIRST_TILE: VSA-H3 tiles are written directly in the
  heads-first layout of the kernel.
- FASTVIDEO_H3_VAE_TILE_PARALLEL: VAE spatial tiles are split across the
  sequence-parallel ranks.
- FASTVIDEO_H3_REF2VA_MEMO_ENTRIES: repeat reference encodes are reused.

The example basic_fasth3_omniref_pdd.py --lossless-accel turns them all on.

On GB200 with 4 GPUs, MiniMax-H3 text to video with audio (124 frames,
1344x768) gives bit-identical frames and audio with the first three switches
on and off.
 2026-10-07 15:35:37 -07:00
Wenbo Ji　嵇文博 8d2e1cf328 [bugfix] Skip streaming IPC queues for standard VideoGenerator (#1902) 
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.
 2026-10-07 15:33:11 -07:00
Wenbo Ji　嵇文博 4ebf11f7d9 [bugfix] Fix test env defaults and server dependencies (#1901) 
The distributed_setup test fixture now sets MASTER_ADDR and MASTER_PORT when
they are not set, with a new port for each test, and restores them after the
test. pyproject.toml now declares orjson and python-multipart, which the
OpenAI-compatible server needs. Two documents about the SSIM reference
bootstrap are corrected.
 2026-10-07 15:32:22 -07:00