Isotr0py be8a72f0a6 Stack torch.compile with CUDA graph decode instead of choosing one ...

Regionally compile the DiT blocks even when enable_cuda_graph_decode is set
and let graph capture record the compiled (fused) kernels; inductor's own
cudagraphs stay off so the two graph layers never nest. Measured on GB200
(1024x1024, 50 steps, seed 42, BF16): combo 3.04s vs compile-only 3.2s,
graph-only 4.0s, eager 4.3s per image; 44.5-45.6 dB PSNR vs eager.

2026-09-22 08:58:32 +00:00

..

attention

Merge remote-tracking branch 'origin/main' into qwen-image-2.1

2026-09-21 07:32:04 +00:00

cache

[Model] MammothModa2: enable Cache-DiT acceleration for the DiT stage (#7291)

2026-09-21 18:34:12 -07:00

diffusion_kv

[Feat] Add native Mooncake KV transfer from AR to DiT (#7166)

2026-09-20 14:44:55 +00:00

distributed

Merge remote-tracking branch 'origin/main' into qwen-image-2.1

2026-09-22 01:23:45 +00:00

executor

Merge remote-tracking branch 'origin/main' into qwen-image-2.1

2026-09-21 07:32:04 +00:00

hooks

[Perf][Diffusion] Skip the Ulysses length all-gather when shards are padded equal (#5717)

2026-09-06 16:23:30 +08:00

interaction

[Core][Diffusion] Camera interaction for diffusion streaming generation (LingBot World 2 as example) (#7198)

2026-09-20 16:19:41 +00:00

layers

[Bugfix][MiniMax-H3] Fix Hopper keyframe and modulation precision (#7913)

2026-09-21 16:26:09 +08:00

lora

Merge remote-tracking branch 'origin/main' into qwen-image-2.1

2026-09-21 07:32:04 +00:00

model_loader

[Bugfix] Isolate weekly CPU tests from inherited attention env and hf_api mocks (#7565)

2026-09-21 15:55:00 +00:00

models

Stack torch.compile with CUDA graph decode instead of choosing one

2026-09-22 08:58:32 +00:00

offloader

Merge remote-tracking branch 'origin/main' into qwen-image-2.1

2026-09-21 07:32:04 +00:00

postprocess

[Core][Diffusion] Add a typed pre-D2H video media contract (#6615)

2026-09-10 09:50:54 +00:00

profiler

[Bugfix] Report stepwise Qwen-Image profiler metrics (#4517)

2026-07-03 04:32:22 +00:00

quantization

[Bugfix] Fix HSDP compatibility with the new online FP8 linear method (#5677)

2026-08-14 15:45:55 +08:00

sched

Merge remote-tracking branch 'origin/main' into qwen-image-2.1

2026-09-22 01:23:45 +00:00

utils

[5/N] Schedule chunk transfers as their own stage (#7406)

2026-09-18 09:27:56 +08:00

worker

Stack torch.compile with CUDA graph decode instead of choosing one

2026-09-22 08:58:32 +00:00

__init__.py

[Diffusion] Qwen image support (#82)

2025-11-28 14:22:11 +08:00

compile.py

fix: keep DLO hooks outside regional compilation (#6073)

2026-08-22 21:44:03 +08:00

config.py

[Diffusion][Attention] Support per-role attention backend via CLI (#2681)

2026-05-10 11:59:05 +08:00

data.py

Stack torch.compile with CUDA graph decode instead of choosing one

2026-09-22 08:58:32 +00:00

diffusion_engine.py

[Core][Diffusion] Add pause_generation(mode="keep") for diffusion stages (#7685)

2026-09-22 01:31:22 +08:00

envs.py

[Kernel][Boogu-Image] Fuse Q/K RMSNorm + interleaved RoPE via fused_qk_norm_rope (#6982)

2026-09-13 09:39:27 +08:00

forward_context.py

[Feat] Add native Mooncake KV transfer from AR to DiT (#7166)

2026-09-20 14:44:55 +00:00

inline_stage_diffusion_client.py

[Bugfix] Keep async video jobs queued until inference and abort on delete (#6759)

2026-09-15 20:43:30 -07:00

io_support.py

[Feature] Support text encoder Disaggregation for H3 (#5885)

2026-08-28 14:27:05 +08:00

ipc.py

[Core][Diffusion] Add a typed pre-D2H video media contract (#6615)

2026-09-10 09:50:54 +00:00

media.py

[Core][Diffusion] Add a typed pre-D2H video media contract (#6615)

2026-09-10 09:50:54 +00:00

model_metadata.py

Merge remote-tracking branch 'origin/main' into qwen-image-2.1

2026-09-21 07:32:04 +00:00

output_formatter.py

[Bugfix] Keep async video jobs queued until inference and abort on delete (#6759)

2026-09-15 20:43:30 -07:00

registry.py

Merge remote-tracking branch 'origin/main' into qwen-image-2.1

2026-09-21 07:32:04 +00:00

request.py

Merge remote-tracking branch 'origin/main' into qwen-image-2.1

2026-09-21 07:32:04 +00:00

stage_diffusion_client.py

[Diffusion] Implement paged AR↔DiT KV connector ("v1", reusing vllm's native mooncake) (#6310)

2026-09-06 05:41:54 +08:00

stage_diffusion_proc.py

[Bugfix] Keep async video jobs queued until inference and abort on delete (#6759)

2026-09-15 20:43:30 -07:00

vllm_config.py

[Feat] Add native Mooncake KV transfer from AR to DiT (#7166)

2026-09-20 14:44:55 +00:00