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

aryan/mlx-rotary-test-ulp

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

fasth3-spark-exact-perf

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

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

Mono Color

• 9be2cae235 [agents] Add issue 1558 investigation handoff Mac Lee 2026-07-06 05:23:40 +00:00
• c426847cb6 [bugfix]: FastWan-QAD weights moved to transformer/ on HF — try both paths SolitaryThinker 2026-07-09 13:17:18 -07:00
• 4a5f0af453 [feat]: NVFP4/FP8 attention benchmark (accuracy, TFLOPS, latency) SolitaryThinker 2026-07-09 13:07:22 -07:00
• 73b56f0a63 [feat]: FlashInfer SageAttention3 backend + attention shape logging SolitaryThinker 2026-07-09 12:48:18 -07:00
• c096fda1eb [docs] Add LTX-2.3 distilled inference run configs (t2v/i2v × 5+2/8+3 × resolutions) (#1568) alexzmsandSolitaryThinker 2026-07-09 11:53:12 -07:00
• afbacb5dfa [docs]: mark fastvideo/workflows as experimental (no stability promise) maint/pr1509-fixed SolitaryThinker 2026-07-09 10:04:27 -07:00
• 8f47e85be0 [bugfix]: retry remote image downloads in load_image (#1570) William Lin 2026-07-09 07:33:24 -07:00
• 90d3bd19eb [infra] Deliver per-job Buildkite env to Modal CI at runtime, not as image layers (#1569) William Lin 2026-07-09 06:33:37 -07:00
• afb4f7d3c5 [ci]: emit v2 performance result schema (#1551) Mac LeeandSolitaryThinker 2026-07-09 14:11:15 +08:00
• c1332cb4fb Init kevin314 2026-07-08 07:29:59 +00:00
• ee623198c9 fix baed on review maint/pr1030-rebase-only H1yori233 2026-06-13 06:21:33 +00:00
• 1e2f2bfbca fix fir transformer 5.10.2 k1kong 2026-06-10 02:44:08 +00:00
• 5e90f421d5 fix H1yori233 2026-06-07 09:49:47 +00:00
• 546ad1fb32 deterministic pipeline parity H1yori233 2026-06-03 05:05:14 +00:00
• 9e261fef21 fix H1yori233 2026-05-31 06:20:46 +00:00
• ea4b2d3195 some fix H1yori233 2026-05-31 02:01:05 +00:00
• c7767910d2 precommit H1yori233 2026-05-31 00:54:14 +00:00
• cc61fc2714 lower SSIM resolution H1yori233 2026-05-30 23:54:29 +00:00
• a0749e98fd cleanup and local test H1yori233 2026-05-30 22:50:14 +00:00
• 2d7e262b13 chore: remove unused reference image Shreejith SG 2026-02-09 00:36:56 +00:00
• 65de01f1f1 Baseline: Deterministic priors, debug logs, SSIM 0.715 Shreejith SG 2026-02-07 07:54:33 +00:00
• e20deeb5eb fix(glm-image): resolve SSIM issues by fixing T5 glyph shapes and KV cache dimensions Shreejith SG 2026-02-06 23:41:33 +00:00
• 51b9761334 Revert fix Shreejith SG 2026-02-05 23:36:56 +00:00
• e99d2d9f16 Numerical alignment fix Shreejith SG 2026-02-05 23:11:58 +00:00
• 8a9ec8621e enable sampling for quality Shreejith SG 2026-02-05 21:25:01 +00:00
• 53440d4562 CI fix - Upgrade transformers and accelerate versions Shreejith SG 2026-01-31 20:09:53 +00:00
• 3d82e5e24c Fix pre-commit errors and SSIM test ImportError Shreejith SG 2026-01-29 07:33:25 +00:00
• cc2c419c4b Fix mypy error: add __init__ to GlmImageDecodingStage Shreejith SG 2026-01-29 05:54:59 +00:00
• 81fb8f2c10 Add GLM-Image SSIM test and fix deterministic latent generation Shreejith SG 2026-01-29 05:27:46 +00:00
• 0d80a668d2 Add L40S reference image for GLM-Image Shreejith SG 2026-01-29 05:24:51 +00:00
• 9dec070d4c Fix GLM-Image seed handling and add SSIM test Shreejith SG 2026-01-29 05:19:18 +00:00
• 326287dffa WIP: add GLM-Image SSIM test and diffusers reference generator ShreejithSG 2026-01-28 07:47:45 +00:00
• 5c68193009 Refactor GLM-Image decoding stage to inherit from our DecodingStage ShreejithSG 2026-01-28 05:57:34 +00:00
• 19ef9fa56f Refactor attn mask handling to use forward context ShreejithSG 2026-01-28 04:50:49 +00:00
• 7833575da8 Fix PR comments: simplify example script and properly handle attention_mask through forward context ShreejithSG 2026-01-27 02:55:19 +00:00
• 9ea387e872 Address more PR review comments ShreejithSG 2026-01-27 02:27:20 +00:00
• 615da36296 Fix precommit error ShreejithSG 2026-01-25 03:51:05 +00:00
• c54e0737d0 Address PR comments: GLM-Image Port to FastVideo- Replaced diffusers FeedForward with FastVideo MLP in DiT.- Moved SDPAMetadata import to top of attention layer for cleaner code.- Removed unnecessary __call__ override from pipeline to use standard composition.- Consolidated CFG into a single batched pass with concatenated embeddings for performance.- Fixed attention mask compatibility issues with optimized FLASH_ATTN backend.- Robustly patched flash_attn unpad utility for variable version return values. ShreejithSG 2026-01-16 03:46:14 +00:00
• 02e1143f22 [feat] Add Kandinsky-5 T2V/I2V pipeline support (#1471) 4 people 2026-07-07 14:43:30 -07:00
• d445d53c9f [bugfix]: keep sdpa metadata masks 2D; lift to 4D only inside the torch-sdpa impl maint/pr1471-rebase-only SolitaryThinker 2026-07-06 17:29:05 -07:00
• 264a2524ae [bugfix]: normalize flash attention metadata masks Aryan Kumar 2026-07-06 17:15:41 -07:00
• 6da4b7dc47 [misc]: scope the kandinsky ssim FA4 pin per-test and name the min-length validator Will LinandClaude Fable 5 2026-07-06 22:53:24 +00:00
• 977184984e [bugfix]: probe gated-repo access with auth_check and keep transient hub failures loud Will LinandClaude Fable 5 2026-07-06 22:53:24 +00:00
• 98df094cbc [bugfix]: coerce integer sdpa masks and keep torch 2D mask semantics Will LinandClaude Fable 5 2026-07-06 22:53:09 +00:00
• 1b9952c32b [bugfix]: stabilize fastcheck lanes for Kandinsky PR Aryan Kumar 2026-07-06 11:42:26 -07:00
• c28fb3f8a1 [test]: add reduced-size Kandinsky-5 T2V SSIM test SolitaryThinker 2026-07-05 16:12:20 -07:00
• 00f7be9be3 [misc]: drop the bundled Modal checkout rewrite (separate PR material) SolitaryThinker 2026-07-05 16:09:06 -07:00
• 57a4c2396d [bugfix]: keep text-encoder CPU offload on by default SolitaryThinker 2026-07-05 15:25:06 -07:00
• 1d08d28eca [bugfix]: retry the non-pr ci checkout too Will LinandClaude Fable 5 2026-07-05 20:53:35 +00:00
• 44dc2acf14 [misc]: document why i2v keeps diffusers cond-channel semantics Will LinandClaude Fable 5 2026-07-05 20:33:01 +00:00
• f8edbec923 [bugfix]: raise on unknown rope types in custom qwen2.5-vl encoder Will LinandClaude Fable 5 2026-07-05 04:56:53 +00:00
• 67852c1407 [bugfix]: restore pipeline-class override short-circuit and dedup kandinsky5 detectors Will LinandClaude Fable 5 2026-07-05 04:56:53 +00:00
• 1d2f329a54 [bugfix]: kandinsky5 stages: official rng order, attn metadata, tensor resize, decode dedup Will LinandClaude Fable 5 2026-07-05 04:56:53 +00:00
• 4bc87f1934 [feat]: register NABLA sparse attention as an attention backend Will LinandClaude Fable 5 2026-07-05 04:56:36 +00:00
• 9862aa05d8 [misc]: default text_encoder_cpu_offload to False Will LinandClaude Fable 5 2026-07-05 03:37:08 +00:00
• b3c526a4b8 [bugfix]: swap cpu-offloaded text encoders to GPU for the forward Will LinandClaude Fable 5 2026-07-05 03:17:11 +00:00
• 3d7ff09e01 [bugfix]: kandinsky5 stages: nabla validation, seeded i2v latent, trajectory support Will LinandClaude Fable 5 2026-07-04 00:17:01 +00:00
• 7bd6263187 [bugfix]: surface kandinsky5 flex-attention failures and preserve rotary dtype Will LinandClaude Fable 5 2026-07-04 00:16:50 +00:00
• 1caed1f789 [bugfix]: preserve tokenizer_kwargs customizations across update_model_arch Will LinandClaude Fable 5 2026-07-04 00:16:50 +00:00
• 2cfcb7b85f [bugfix]: fix kandinsky5 registry detectors and distilled repo name Will LinandClaude Fable 5 2026-07-04 00:16:39 +00:00
• 532b09fb73 [bugfix]: fix bash syntax error in modal ci pr checkout Will LinandClaude Fable 5 2026-07-03 23:49:44 +00:00
• 539884b9ba [bugfix]: keep kandinsky5 example text encoders on GPU Will LinandClaude Fable 5 2026-07-03 23:49:44 +00:00
• b0dd449e5d [bugfix]: fix kandinsky5 t2v frame-0 denoising and restore trained prompt template Will LinandClaude Fable 5 2026-07-03 23:49:28 +00:00
• ead828a726 [bugfix]: make custom qwen2.5-vl encoder compatible with transformers 5.x Will LinandClaude Fable 5 2026-07-03 23:49:18 +00:00
• 33ce65c118 [bugfix]: harden modal ci pr checkout Aryan Kumar 2026-07-02 08:15:32 -07:00
• 11ffa83978 [bugfix]: address kandinsky5 i2v review feedback Aryan Kumar 2026-06-29 07:23:01 -07:00
• 36ea5747fe [misc]: polish kandinsky5 sparse attention updates Aryan Kumar 2026-06-24 19:48:26 -07:00
• 4b0206d82a [feat] add support for Kandinsky 5 Pro, Lite I2V leffff 2026-06-27 09:59:44 +00:00
• fb9b9b2b1b [misc] I2V debug leffff 2026-06-27 03:22:55 +03:00
• ed1fbee9b0 [feat] add support for Kandinsky5 I2V leffff 2026-06-26 14:20:03 +03:00
• 6de9d6ee87 [misc] add skeleton for Kandinsky5 I2V pipeline leffff 2026-06-26 09:45:52 +00:00
• 3dbd6d65a0 fix nabla attention leffff 2026-06-25 02:21:21 +00:00
• 55693a498d add nabla leffff 2026-06-25 00:40:05 +00:00
• 5784d8ad40 add exmaple for kandinsky5 T2V Lite 5s leffff 2026-06-24 12:46:04 +00:00
• eaf0f6a5b3 fix: apply CodeRabbit auto-fixes coderabbitai[bot]andCodeRabbit 2026-06-18 22:50:38 +00:00
• 4c42a73098 [feat]: add kandinsky5 pipeline support Aryan Kumar 2026-06-18 15:31:36 -07:00
• 23bcf88026 [bugfix]: keep _rms_norm_dispatch output in the input dtype under autocast maint/ltx2-rmsnorm-dtype SolitaryThinker 2026-07-05 13:11:42 -07:00
• e2f4d1a7b5 [feat]: add SwanLab tracker (#1461) KaredandSolitaryThinker 2026-07-07 16:35:38 +08:00
• f037351146 [feat] Add Clean-history Teacher Forcing and Causal Consistency Distillation (#1505) Kaiqin KongandSolitaryThinker 2026-07-07 01:13:11 -07:00
• 1ee11e08dc [ci]: add performance fingerprint cohorts (#1546) Mac LeeandSolitaryThinker 2026-07-07 15:24:00 +08:00
• 20581fe9c6 [misc]: double CI lane CPU/memory reservations (8->16 vCPU, 32->64GB) maint/perf-lane-cpu16 SolitaryThinker 2026-07-07 00:22:08 -07:00
• 595f0ea60e [feat] Add DreamX-World 5B Cam and AR pipelines (#1538) 3 people 2026-07-07 14:16:18 +08:00
• 87737d10fd [bugfix]: keep _rms_norm_dispatch output in the input dtype under autocast maint/pr1488-rebase-only SolitaryThinker 2026-07-05 13:11:42 -07:00
• 91f2acbbc9 [bugfix] nvfp4: reject non-float inputs, extract dtype coercion + test Raghav 2026-06-25 16:54:11 -07:00
• 2e8eb75c11 [bugfix] nvfp4: cast fp32 inputs to bf16 instead of asserting Raghav 2026-06-24 13:43:44 -07:00
• bb3b3d0657 [ci]: collect the tracker unit tests in the unit lane maint/pr1461-rebase-only SolitaryThinker 2026-07-05 15:46:38 -07:00
• 67a8f7ac85 [fix]: complete SwanLab tracker integration SolitaryThinker 2026-06-28 00:51:43 -07:00
• d133aef55d [feat]: add swanlab tracker kared 2026-06-15 15:34:05 +08:00
• 2892c1255d [bugfix]: bound WanCausalModel num_frames_per_block override to <= 3 maint/pr1505-rebase-only SolitaryThinker 2026-07-05 15:01:25 -07:00
• d2d76c5571 [bugfix]: refuse teacher forcing with a local attention window SolitaryThinker 2026-07-05 15:01:09 -07:00
• 1fb98cf6d2 [bugfix]: set causal-CD timesteps before the teacher CFG forwards SolitaryThinker 2026-07-05 15:00:47 -07:00
• 7e39111591 [bugfix]: checkpoint causal-CD EMA consistency target on save/resume SolitaryThinker 2026-07-05 15:00:27 -07:00
• 7bb12e6b4b fix H1yori233 2026-06-28 11:19:06 +00:00
• aaccd4a9ed fix H1yori233 2026-06-28 10:39:42 +00:00
• 51e4f3ee16 fix scheduler H1yori233 2026-06-28 10:18:00 +00:00
• 42a2c26af0 update scheduler and ema H1yori233 2026-06-27 21:19:49 +00:00
• 0f4edd0e12 cleanup H1yori233 2026-06-27 11:03:46 +00:00
• 1fc165a95a cleanup H1yori233 2026-06-27 10:02:48 +00:00
• 935ed797f2 add cf H1yori233 2026-06-27 08:02:19 +00:00
• 110c07033e [misc]: log GPU clock/power telemetry during the perf lane maint/pr1538-rebase-only SolitaryThinker 2026-07-06 18:36:59 -07:00