hao-ai-lab-FastVideo

nodes/hao-ai-lab-FastVideo

Fork 0

Commit Graph

Select branches

Hide Pull Requests

4090-flex-compare

SolitaryThinker-patch-1

SolitaryThinker-patch-2

add_agents.md

add_ci

alex/opt-ltx23-nvfp4

aryan/h3-spark-gb10

attn-loud-fail

b200-ssim-full-quality

backend_dev

ci/precommit-status-post

ci/slash-test-precommit

codex/performance-dashboard-cohort-selection

codex/selectable-gpu-ci

comfy_custom_wendy

comfy_pipline

comp_time

cursor-agent

deltadit

dmd

docs-build

docs/add-release-skill

docs/readme-realtime-demo

ds8/dreamversev2-dev

experiments

feat/add-model-skill-revision

feat/cosmos3-action

feat/cosmos3-audio

feat/cosmos3-i2v

feat/cosmos3-reasoning

feat/cosmos3-tier-a-port

feat/dreamverse-monorepo-main-sync

feat/dreamverse-monorepo-main-sync-plus-profile

feat/h3-dmd2

feat/h3-dmd2-vsa

feat/h3-t2av-dmd-training

feat/improve-ci

feat/kaiqin/add-mg-df

feat/modular-training-profiler-qad

feat/pan2

feat/pipeline-startup-profiling

feat/startup-loader-foundation

feat/studio-ray-multinode

fix-h3-regional-compile-recompiles

fix/ci-retry-and-ssim

fix/fa4-cutedsl-api-compat

fix/interleave-thinker-review

fix/pr1807-collective-agreement

fix/setuptools-package-discovery

fix/vae-tiling-reset

fix/workload-type-from-registration

fix_timestep_sample

flex_sta_rlsu

flex_sta_rlsu_debug

generate_syn

hangliang_sw

hao-cleanup2

hao-edit

hao-test

hunyuan-gamecraft-v2

hunyuan15-support

hy_experiments

hy_experiments2

i2v_demo

i2v_fixed

impl-kandinsky-r12-repair

infer_sta_tea

inference-compile

infra/gb10-image-sm121a

integration/h3-perf-stack

integration/h3-vsa-fullgraph-all-20260822

jinzhe/distill

jinzhe/fix/parquet_dataset

jinzhe/flux

kernel/gb10-vsa128-tile128

kernels

klin/5090-fixes

klin/comfyui-precision-fixes

klin/dv-test

klin/h3-studio-editor

klin/trackwan

klin/ui-testing-1

klin/world_model_ui

kv_cache_fix

loay/ltx2_fp8

ltx25-integration

main

maint/ci-runner-rewrite-only

maint/cosmos3-final-refresh-20260720

maint/cosmos3-video-parity-20260720

maint/cosmos3-video-parity-rebase-only

maint/cosmos3-video-parity-seed

maint/dreamverse-remove-equal-delay-race

maint/ltx2-finetune-rebase-only

maint/ltx2-nvfp4-qat-rebase-only

maint/ltx2-rmsnorm-dtype

maint/perf-comparator-statuses

maint/perf-host-profile-guard

maint/perf-lane-cpu16

maint/perf-worker-log-capture

maint/pr1030-rebase-only

maint/pr1290-rebase-only

maint/pr1290-refresh

maint/pr1302-rebase-only

maint/pr1321-rebase-only

maint/pr1339-rebase-only

maint/pr1371-rebase-only

maint/pr1382-fullstack

maint/pr1382-rebase-only

maint/pr1442-fixed

maint/pr1442-rebase-only

maint/pr1449-fixed

maint/pr1453-fixed

maint/pr1453-rebase-only

maint/pr1461-rebase-only

maint/pr1471-fixed

maint/pr1471-fixed-v2

maint/pr1471-rebase-only

maint/pr1476-rebase-only

maint/pr1488-fixed

maint/pr1488-rebase-only

maint/pr1494-rebase-only

maint/pr1496-rebase-only

maint/pr1505-fixed

maint/pr1505-rebase-only

maint/pr1509-fixed

maint/pr1518-fixed

maint/pr1538-fixed

maint/pr1538-rebase-only

maint/pr1538-wave5

maint/pr1546-fixed

maint/pr1546-fixed-v2

maint/pr1551-fixed

maint/pr1560-rebase-only

maint/pr1561-rebase-only

maint/pr1563-rebase-only

maint/pr1574-rebase-only

maint/pr1577-rebase-only

maint/pr1579-rebase-only

maint/pr1580-rebase-only

maint/pr1581-rebase-only

maint/pr1590-rebase-only

maint/pr1591-rebase-only

maint/pr1596-rebase-only

maint/pr1601-rebase-only

maint/pr1602-rebase-only

maint/pr1617-torchcodec-fix

maint/pr1622-rebase-only

maint/pr1626-rebase-only

maint/pr1627-rebase-only

maint/pr1659-fixes

maint/pr1665-rebase-only

maint/pr1691-rebase-only

maint/pr1764-rebase-only

maint/pr1796-rebase-only

maint/pr1802-rebase-only

maint/preprocessing-ci

maint/skip-longcat-oom

maint/ssim-png-reference-cleanup

maint/zimage-ssim-prep

matthew/causal

matthew/debug_training

matthew/debug_training_2

matthew/demo

matthew/moe_debug

matthew/moe_revert

matthew/peprocessing

matthew/wan2.2-new

mochi-dev

model/port-lingbot-video-ti2v

model/port-magi-2

model_config

motion-stream/causal-kernel

ms-recipe

perf/fa4-vsa-block64

perf/fused-adamw-training

perf/h3-regional-compile-main-20260822

perf/h3-vsa-fullgraph-20260822

perf/ulysses-h3-adaptive

perf/ulysses-mfu-study

perf_reg_skill

pipeline_config

pr1362-fixes

pr1703-fixes

pr1711-fixes

py/add_rl

quant

readme-wan-crush-smol-fix

reason

refactor-config

refactor-lora

refactor/auto-merge-flow

refactor_attn

revert-1172-job_runner_ui_2

revert-591-yq/update

revert-595-main

revert-955-will/release

rlsu-test

rlsu-video-cut

satyam/ci-production-performance-regression

satyam/diagg_spark

satyam/profiling

satyam/temp_per_check_H3_spark

sdpa-mask-2d-fix

sdpa-mask-2d-fix-v2

shao/eval

shao/ltx2-cfg-v2

shreejith/unified-kernel-package

sta_do_not_delete

swe-1655-orphans

sync-branch

test-hf-sync

trackwan_bidir

upgrade_torch

v2

v2_backup

vae-parallel-decode

vsa_bwd_fix

wan-vae

wangame-debug-dfsft

wei/SF

wei/SF_infer

wei/api

wei/datasets

wei/distill

wei/dit_debug

wei/hy1.5

wei/hy15-1080

wei/hy15-i2v

wei/hy15_causal

wei/hy15_cf

wei/hy15_ode

wei/hy15_preprocess

wei/i2v

wei/issues

wei/ode_preprocess

wei/videox

wei/wan2.2-distill-new

wei/wan2.2-finetune

wei/wan2.2-training

wei/wan_dit_align

wei/wan_mc

will/add-model-skill

will/api_6_test

will/benchmark

will/cosmos2

will/debug

will/debug_ode

will/debug_vae

will/debug_validation

will/demo_public

will/design

will/distill_model_load

will/dmd_i2v_debug

will/docker_demo

will/dreamverse-monorepo

will/ema_fsdp_fix

will/f_0

will/fix-rebased

will/fix2

will/fix_denoise_init

will/fix_dep

will/fix_distill

will/fix_i2v

will/fix_shape

will/fix_tests

will/fix_validation

will/fix_validation_pipeline

will/fix_wan2

will/flux-ci

will/fp8

will/gradio-demo

will/hy15_training

will/i2v_oom

will/i2v_val

will/improve_preproc

will/improve_preprocessing

will/imrpove_api

will/kv_cache_fix_rebased

will/lora_checkpoint

will/ltx-base

will/ltx2

will/ltx2_sr

will/ltx2_sr_port

will/magi

will/magi-03-dit-review-fixes

will/magi-06-activate

will/moe_ode

will/move_tests

will/new_i2v

will/ode_init

will/preprocess_ci

will/process_prompts

will/qat

will/readme2

will/rebase

will/refactor_denoising

will/remove_v1_folder

will/reorg_examples

will/rewrite_validation

will/rl

will/sf_causal

will/sp_training

will/streaming

will/studio-l1-jobs

will/t2v_i2v

will/test_pr

will/training-VSA-rebase

will/training_mp

will/training_readme

will/transfusion

will/update_docker

will/update_min_python

will/v2.1

will/veo-training-data

will/videox

will/wan2.2

will/wan2.2-finetune-rebased-prof

will/wan2_2_sf

will/will_dmd_i2v

will/zmq-worker

william/pr1253-review-and-stable-audio

wip/flashinfer-sageattn3

worktree-agent-a58b1c80

worktree-agent-a67f7ec1

worktree-agent-a7083254

yq-update

yq/Wan2.2-time

yq/distill

yq/distill-i2v

yq/dmd-vis

yq/wan2.2-training

zihang-gradient-checkpoint

zihang-textencoder

zihang-training

v0.0.5

v0.1.0

v0.1.1

v0.1.2

v0.1.5

v0.1.6

v0.1.7

v0.2.0

81f3fec7fd [bugfix]: workers killed by a signal now log the reason (#1725) Kyle Hu 2026-09-24 17:06:55 -07:00
d265a454bf [bugfix]: Fall back when large pinned output allocation fails (#1759) Raghav K 2026-09-24 17:05:59 -07:00
c100c66578 [bugfix]: isolate cancelled streaming requests from GPU result readers (#1848)
Xingjian Guo Gxj230958 SolitaryThinker
2026-09-25 08:04:59 +08:00
8760eb7a06 [feat] Add FastH3 8-Step V2 MLX inference (#1863) Aryan Kumar 2026-09-23 12:42:03 -07:00
361f919c88 [feat]: add a narrow FastH3 8-step MLX ladder Aryan Kumar 2026-09-23 11:06:55 -07:00
8b5377aab2 Merge remote-tracking branch 'origin/main' into aryan/pr-1863-fasth3-8step Aryan Kumar 2026-09-23 10:52:20 -07:00
d995516da0 [bugfix] Restore page interaction after dismissing Create Job or successfully creating a job (#1869) Yixing Wangandleo 2026-09-21 12:03:54 -07:00
f47ad3f5b7 [bugfix] fastvideo-kernel: install the Python + Triton package when the HIP toolchain cannot be configured (#1871) Keith 2026-09-22 02:59:13 +08:00
10bdf5e076 [bugfix] frozen offload: no host copy for an already resident module Hexu ZhaoandClaude Opus 5 2026-09-17 22:34:26 +00:00
3e26db40b0 [perf] H3 decode: pin the frozen VAEs' host copies Hexu ZhaoandClaude Opus 5 2026-09-17 20:01:01 +00:00
c73dd0ab55 [perf] H3 decode: offload the frozen VAEs without the device-to-host copy Hexu ZhaoandClaude Opus 5 2026-09-17 20:00:30 +00:00
430e52154e [bugfix]: Restore exact Qwen3-VL vision interpolation (#1737)
primorLee William Lin SolitaryThinker
2026-09-18 20:00:05 +08:00
384eee8aef feat(mlx): add FastH3-8-Step-V2 DMD schedule and thermal-safe MLX inference vanch 2026-09-17 16:12:02 +08:00
c4824c7764 [bugfix] FP8: gate the ROCm _scaled_mm path on CDNA4 (gfx950) instead of the capability tuple (#1859) KeithandSolitaryThinker 2026-09-17 10:58:55 +08:00
0c16ec91b0 [feat]: connect Dreamverse creation settings to generation ds8/dreamversev2-dev
Davids048 Cursor Agent Aryan Kumar
2026-09-16 00:02:24 +00:00
e57543b79d [feat]: Dreamverse creation studio UI uplift (#1853)
Aryan Kumar Cursor Agent Aryan Kumar
2026-09-15 16:45:08 -07:00
9b0e57fe4b [ci] Make Dreamverse provider race test deterministic (#1729) Raghav K 2026-09-15 14:13:56 -07:00
0100218594 [feat] Support the FastH3 8-Step V2 checkpoint: checkpoint-defined shifts, explicit DMD schedule, new example (#1852) William Lin 2026-09-15 14:13:45 -07:00
39718cd54d [bugfix] fix(cosmos): make AdaLayerNorm autocast device-agnostic (#1818) li-lizhe 2026-09-15 22:44:51 +08:00
37d06a832f [feat] Add fastvideo serve configs for Wan CUDA models (#1801) IshanandSolitaryThinker 2026-09-14 18:23:35 -07:00
8839ba8d4d [bugfix] Add OpenAI-compatible image generation endpoint (#1840) sudhirpol522 2026-09-14 18:55:56 -06:00
61b91220c0 [bugfix] Validate image response format before generation (#1841) sudhirpol522andSolitaryThinker 2026-09-14 18:01:40 -06:00
316f3876c2 [bugfix]: allow MiniMax H3 frame padding at the 15-second limit Lele 2026-09-15 07:54:41 +08:00
614b59543c [bugfix]: respect serialized tensor dtype in parquet dataloader (#1843) Yaegaki1Erika 2026-09-15 07:34:25 +08:00
1c14afd559 [docs]: refresh AGENTS.md maps and add Wan SP/I2V and CI test notes (#1845) William Lin 2026-09-14 16:06:53 -07:00
9a3c45779c [bugfix]: fix Wan I2V DMD conditioning under sequence parallelism (#1844) Yaegaki1Erika 2026-09-15 06:58:13 +08:00
bfc9c01797 [feat]: convert the MiniMax H3 text encoder to NVFP4 (#1838) Kyle Hu 2026-09-12 18:48:07 -07:00
3a3ad3d209 [feat]: NVFP4 text encoder for MiniMax H3 (#1837) Kyle Hu 2026-09-12 18:31:59 -07:00
aef4e9b3b1 [kernel] VSA kernel: one sm_100a / sm_103a image per listed arch; un-gate backward on sm_103a (#1833) lpc0220andClaude Fable 5.1 2026-09-12 16:49:36 -07:00
a943220c11 [bugfix]: drop dead h3_sequential_load from Spark FastH3 presets (#1831)
Aryan Kumar Cursor Agent Aryan Kumar
2026-09-08 12:38:06 -07:00
64df158881 Add logging for FA4 backend Satyam Srivastava 2026-09-07 00:07:38 -07:00
d05b59db37 Add H3 inter node profiling Satyam Srivastava 2026-09-05 23:04:12 -07:00
8d4e05acef Remove CUDA to CPU and vice versa conversion And more network payload logging Satyam-53 2026-09-04 22:47:07 -07:00
4be9192615 Spark Diagg Inference/Serving commits Satyam Srivastava 2026-09-03 13:39:44 -07:00
556ac7088e [refactor] Simplify Wan sampling and tests (#1825) William Lin 2026-09-07 15:57:19 -07:00
e7456f1b75 Add H3 support into Dreamverse (#1800) Junda Su 2026-09-07 13:09:44 -07:00
c993d7393e [kernel] sm_100a CUDA backward for VSA block-sparse attention (blk64) (#1819) lpc0220 2026-09-06 21:00:18 -07:00
1568d3d1f7 [test]: require actual tuned Ulysses execution in native validation perf/ulysses-h3-adaptive SolitaryThinker 2026-09-07 01:43:25 +00:00
a61f89a194 [docs]: explain bounded Ulysses chunks and long training policy SolitaryThinker 2026-09-06 21:27:45 +00:00
89836e8a98 [perf]: preserve the full baseline plan for long H3 training SolitaryThinker 2026-09-06 21:04:57 +00:00
76129f5a89 [perf]: add bounded H3 windows and agreed GB200 launch plans SolitaryThinker 2026-09-06 20:41:05 +00:00
2d58c9bfde [test]: sm_100a backward parity and run-to-run stability at 65536 tokens vsa_bwd_fix SolitaryThinkerandClaude Fable 5.1 2026-09-06 19:59:19 +00:00
62f0bc1daa [misc]: pin the identity-order/chunked-launch invariant; fix set_extension docstring SolitaryThinkerandClaude Fable 5.1 2026-09-06 19:59:19 +00:00
60165c7b65 [fix]: restore tcgen05.wait::ld after the TMEM loads in the sm_100a backward SolitaryThinkerandClaude Fable 5.1 2026-09-06 19:59:19 +00:00
fc51550920 [perf] Probe H3 chunked all-to-all with direct output slices perf/ulysses-mfu-study SolitaryThinker 2026-09-06 01:58:09 +00:00
942e8b8404 [perf] Study H3 long-sequence window limits and bounded chunking SolitaryThinker 2026-09-06 01:50:22 +00:00
7f83164233 [refactor] Move Wan VAE into the Wan package (#1824) William Lin 2026-09-05 18:43:55 -07:00
c548e5834e [perf] Build the pinned sparse kernel for the MFU experiment SolitaryThinker 2026-09-06 01:20:21 +00:00
368f6b8891 [perf] Compare launch geometry in dense and sparse training blocks SolitaryThinker 2026-09-06 01:13:37 +00:00
e55934b62e [perf] Add reproducible Ulysses MFU bottleneck experiments SolitaryThinker 2026-09-06 01:08:01 +00:00
a06e63827d [test] Cover real CUDA capture and recovery after a cold decline fix/pr1807-collective-agreement SolitaryThinker 2026-09-06 00:34:17 +00:00
e536bb8544 [bugfix] Keep Ulysses rank agreement unconditional and reuse vote buffers SolitaryThinker 2026-09-06 00:23:28 +00:00
4e52f47d1e [feat] add MXFP8 support on H3 (#1796) Junda Su 2026-09-05 17:08:58 -07:00
e19913f6e9 [refactor] Group Wan transformer and config (#1823) William Lin 2026-09-05 17:05:32 -07:00
d2c9d61240 [kernel] sm_100a backward: drop tcgen05.wait::ld after the TMEM loads rootandClaude Fable 5.1 2026-09-05 16:25:55 -07:00
57bb232432 [kernel] sm_100a backward: drop the cached identity order array, nullptr remap = identity rootandClaude Fable 5.1 2026-09-05 12:08:40 -07:00
88319b75d5 [kernel] sm_100a backward: keep each softmax warp's bf16 overlay inside its own fp32 columns rootandClaude Fable 5.1 2026-09-05 11:29:04 -07:00
0fe2db32e2 Add support for NVFP4 on FFN in H3. maint/pr1796-rebase-only Davids048 2026-09-01 06:55:09 +00:00
dd2aed7d2a Restore FastH3 example scripts Davids048 2026-09-01 06:02:00 +00:00
10593622a4 Remove MiniMax H3 profile scripts Davids048 2026-08-31 20:20:53 +00:00
f6db9c066e Add MXFP8 H3 FFN inference Davids048 2026-08-31 07:27:19 +00:00
6fac641c00 add nvtx around H3 transformer blocks Davids048 2026-08-31 07:20:15 +00:00
540b1915b2 Add profile scripts. Davids048 2026-08-31 07:19:46 +00:00
8386845e51 Update profile script. Davids048 2026-08-31 05:53:36 +00:00
9dffaa0a89 rename launch script. Davids048 2026-08-31 00:36:10 +00:00
31cc4924d9 Refactor: simplify FastH3 example inference script. Davids048 2026-08-31 00:01:41 +00:00
2413a57651 Disable old SSIM models (#1820) William Lin 2026-09-04 21:20:48 -07:00
749a9c5d97 [kernel] sm_100a CUDA backward for VSA block-sparse attention (64-token blocks) rootandClaude Fable 5.1 2026-09-04 18:56:01 -07:00
7bb76b5ec9 [feat] Add native SM103a VSA support (#1812) SYLAR 2026-09-04 06:00:28 +08:00
0bd19a976b [docs]: add one-Spark FastH3 cookbook runtime with a device-count row (#1811) Aryan KumarandAryan Kumar 2026-09-02 10:05:59 -07:00
40b93784d2 [docs]: add cookbook link to README (#1810) William Lin 2026-09-01 12:42:19 -07:00
33d3478bad [docs] Announce local FastH3 support (#1809) Aryan Kumar 2026-09-01 12:28:04 -07:00
3eabb7b40b [bugfix] gate the fused Ulysses a2a on one host, and stop re-voting per call shaoxiongduan 2026-09-01 07:52:24 +00:00
3d8ac9d14b [feat]: collapse cookbook recipe pages into an accordion layout, add … (#1805)
Ishan Vaish, Ishan Claude Sonnet 5
2026-09-01 01:51:51 -07:00
aaef49bfc6 [feat]: run FastH3 across two DGX Sparks with Ray sequence parallel (#1803)
Aryan Kumar Kyle Claude Opus 5 Satyam Srivastava Aryan Kumar
2026-09-01 01:37:23 -07:00
cf6a00b9be [feat]: add opt-in CUDA TAEH3 preview decode for FastH3 (#1795) Aryan KumarandAryan Kumar 2026-08-31 23:34:56 -07:00
56ef6241d4 [feat] Add FastMetal 14B and 5B MLX serving, harden test coverage maint/pr1802-rebase-only Ishan Vaish 2026-08-31 20:10:40 -07:00
247972271c [feat] Serve FastMetal (Wan2.1) 1.3B through native MLX Ishan Vaish 2026-08-31 19:37:09 -07:00
d74475ce06 [feat] Let the OpenAI server plug in a non-CUDA generator Ishan Vaish 2026-08-31 19:37:09 -07:00
1ae39562dd [bugfix] Write generated documentation as UTF-8 (#1797) Shahrad ZomorrodiandShahrad Zomorrodi 2026-08-31 22:45:03 -07:00
26064193e2 [feat] Add an H3 server cookbook and prompt playground (#1798) Aryan KumarandAryan Kumar 2026-08-31 22:14:33 -07:00
8446fc003e [docs]: add FastH3 Preview v1 news links (#1804) William Lin 2026-08-31 22:12:02 -07:00
c0dcd9efa4 Test commits satyam/temp_per_check_H3_spark Satyam Srivastava 2026-08-31 19:54:00 -07:00
72c57b4ae8 Merge remote-tracking branch 'origin/main' into push_pr_1761 Satyam Srivastava 2026-08-31 13:19:10 -07:00
a28f2bab4b [feat] Add an optional MLX TAEH3 preview decoder (#1794) Aryan KumarandAryan Kumar 2026-08-31 05:08:06 -07:00
f82d8be4bf [perf] Sequential MiniMax H3 start with GPU-direct DiT load (#1793) Aryan KumarandAryan Kumar 2026-08-31 04:20:21 -07:00
8e1775183e [perf] Speed up exact MiniMax H3 MLX inference (#1792) Aryan KumarandAryan Kumar 2026-08-31 04:14:55 -07:00
cc9bfd069d [perf]: Load H3 DiT weights onto CUDA, not CPU Aryan Kumar 2026-08-30 20:56:07 -07:00
620bc36dc4 [docs]: Cookbook catalog improvements (#1790) Aryan KumarandAryan Kumar 2026-08-30 17:50:57 -07:00
29ff16ec96 [feat] Add MiniMax H3 MLX spatial fast mode (#1789) Suhaan Khurana 2026-08-30 17:47:02 -07:00
8f9d76a80d [perf]: dispatch wide-M affine H3 MLX linears through dequant plus dense GEMM (#1788) Aryan KumarandAryan Kumar 2026-08-30 14:42:01 -07:00
a4d9a75e2c [perf] Add MiniMax H3 MLX VSA and SIMD attention (#1776)
Aryan Kumar Aryan Kumar coderabbitai[bot]
2026-08-30 13:44:54 -07:00
b2db0c0a13 [ci]: seed stable GB10 grad-norm references (#1756) KyleNeverGivesUp 2026-08-30 03:09:18 -07:00
6aa7d8a278 [misc] FastVideo Studio UI Additions (H3 Ref2V support) (#1783) Kevin Lin 2026-08-30 05:06:47 -05:00
ccc9014430 [docs] Add model-family inference cookbook (#1787) Aryan KumarandAryan Kumar 2026-08-30 02:26:10 -07:00
a3d9725d15 [bugfix]: share compiled VSA graphs across H3 layers Satyam Srivastava 2026-08-29 16:37:51 -07:00
0875b78bef [bugfix]: keep unused LoRA setup from loading lazy DiTs Satyam Srivastava 2026-08-28 22:30:57 -07:00
669bd111eb [bugfix]: preserve compile across lazy module reloads Satyam Srivastava 2026-08-27 02:53:11 -07:00
df849eea34 [perf]: test the real load and stage-construction paths KyleandClaude Opus 5 2026-08-26 03:00:18 -07:00
f247a307a8 [perf]: make deferred loading opt-in per pipeline and clean up on failure KyleandClaude Opus 5 2026-08-26 02:56:26 -07:00

1 2 3 4 5 ...