Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• 453ea21dd3 fix(qwen2_5vl): replace in-place += with out-of-place + on expand view in decode path (#22634) Liu Zhenlong 2026-08-07 11:51:09 +08:00
• f9e6888b5a [Distributed] Propagate semantic group names to PyTorch process groups (#32900) 3 people 2026-08-07 11:38:12 +08:00
• c54dc4582f [Diffusion]Skipping tensor copying for non-BCG GLM-Image workflows (#33688) 3 people 2026-08-07 06:19:58 +03:00
• 163b739b34 [CI] Refresh the CPU HF cache base only on main-ref runs (#33904) Liangsheng Yin 2026-08-06 19:37:59 -07:00
• fe6a05a8e8 fix: preserve priority for batched embedding requests (#32977) Nikhil KulkarniandClaude Opus 5 2026-08-06 19:26:18 -07:00
• db8f3cdd11 fix(gdn): skip the -1 padding sentinel in the chunked extend kernel (#33810) Yuwei An 2026-08-06 18:48:47 -07:00
• 1e08b865f9 [diffusion] feat: support K/V-gather style sequence parallel (CP-like) attention (#32667) MickandClaude Fable 5 2026-08-07 09:39:28 +08:00
• 9ee658d4f6 [diffusion] CI: fix output-rank test fixture (#33878) Mick 2026-08-07 09:33:50 +08:00
• 2c3ecf32f1 [Fix] Reformat /vertex_generate successful predictions (#33446) Dustin LuongandJimmy Shong 2026-08-06 17:00:48 -07:00
• af7c62e337 Fix paged SWA retraction resume accounting (#33794) Hao Zhangandzhisbug 2026-08-06 16:32:54 -07:00
• e0af47b03e Fix sgl-deep-ep builder dependencies (#33866) Baizhou Zhang 2026-08-06 16:30:02 -07:00
• bae29f716a Fix inference mode mismatch in FlashInfer warmup (#33788) Po-Han Huang (NVIDIA) 2026-08-07 07:11:41 +08:00
• b38caebf09 [AMD] Enable gfx1250 sgl-kernel builds (#32466) Oguz UlgenandLianmin Zheng 2026-08-06 15:47:17 -07:00
• 18e6c61c21 [AMD] perf: compact Triton extend-attention for ragged prefill (AMD/HIP-only) (#29677) valechen 2026-08-06 14:46:10 -07:00
• dd7e4c91e2 Fix Mistral-Large-3 EAGLE draft skipping DeepseekV2Model.__init__ (#33785) Brayden Zhong 2026-08-06 13:59:28 -07:00
• 971932d661 [Kimi-K3] Allow DSPARK verify on cutedsl_mla (fold_sq) (#33650) Yuhao Yang 2026-08-07 04:54:25 +08:00
• 2fc557254b fix(PP): size the mamba pool per pipeline stage, not per whole model (#33666) YAMY 2026-08-06 13:10:43 -07:00
• 434e646282 [Deps] Upgrade CUDA PyTorch stack to 2.13 (#28836) Mohammad Miadh AngkadandBrayden Zhong 2026-08-07 03:08:44 +08:00
• 4ad990ba7d [ModelOpt FP4] Support online MoE weight quantization (#33115) Ziang LiandBrayden Zhong 2026-08-06 11:01:55 -07:00
• 05c7ebf64c [Disagg][StagingBuffer][2/2] Support radix cache (#30545) YAMYandShangming Cai 2026-08-06 08:59:35 -07:00
• 8a1637a479 Fix serving benchmark post-warmup cache flush race (#33663) Mohammad Miadh Angkad 2026-08-06 23:27:10 +08:00
• 7195b8e4c7 [diffusion] refactor: validate and document spectrum controls (#33851) Mick 2026-08-06 23:23:11 +08:00
• 44bde3911a [diffusion] fix: resolve IPC A2A peers from process groups (#33848) Mick 2026-08-06 23:22:41 +08:00
• c212a6938c [diffusion] chore: retire released warmup and decoder flags (#33850) Mick 2026-08-06 23:02:32 +08:00
• 591cfb0881 [diffusion] FLUX.2 bit-exact residual-gate fast path (H200 klein-4B 50-step denoise -1.2%) (#33823) Xiaoyu ZhangandClaude Fable 5 2026-08-06 22:54:40 +08:00
• dd98c9572a [diffusion] Generalize the FLUX.2 VAE decoder fast path to AutoencoderKL (Z-Image / FLUX.1) behind quality=high (#33818) Xiaoyu ZhangandClaude Fable 5 2026-08-06 22:53:15 +08:00
• 21225aba3d [Scheduler] Fix to restrict the SWA chunk-cap escape hatch to true head-of-line livelock (#32700) Peng Wu 2026-08-06 07:09:04 -07:00
• 183bd80add [diffusion] chore: centralize entrypoint API hygiene (#33845) Mick 2026-08-06 21:53:40 +08:00
• 2132cdef16 [diffusion] chore: consolidate pipeline core hygiene (#33843) Mick 2026-08-06 21:51:34 +08:00
• 45dfd80674 [diffusion] refactor: simplify disaggregation transport hygiene (#33844) Mick 2026-08-06 21:50:32 +08:00
• 1a15cf1536 Gate multimodal feature transport by model capability (#33653) Mohammad Miadh Angkad 2026-08-06 21:48:32 +08:00
• e8d0fe92e9 [diffusion] Fix GLM-Image resolution alignment (#32999) AuFlowandAuFlow 2026-08-06 20:41:45 +08:00
• 3654740347 [diffusion] ERNIE-Image bit-exact fused RMSNorm+scale/shift (H200 1024^2 e2e 15.63 -> 15.00 s, denoise -3.3%) (#33854) Xiaoyu ZhangandClaude Fable 5 2026-08-06 19:58:44 +08:00
• 295784723a [diffusion] Ideogram 4: fuse RMSNorm modulate/gate chains via the Z-Image Triton suite behind quality=high (H200 e2e -2.9%/-3.4%) (#33822) Xiaoyu ZhangandClaude Fable 5 2026-08-06 19:57:34 +08:00
• eff6a11350 [diffusion] FLUX.1 bit-exact residual-gate fast path + tanh-GELU epilogue behind quality=high (H200 e2e -1.1% lossless / -4.3% high) (#33819) Xiaoyu ZhangandClaude Fable 5 2026-08-06 19:56:38 +08:00
• f8f2870a84 Profiling Enhancements [1/3]: cuda graph profile traces (#24370) 3 people 2026-08-06 06:19:35 -04:00
• f6de147b8d Remove revoke queue after hit-then-alloc refactoring (#33613) Zhiqiang Xie 2026-08-06 03:10:10 -07:00
• 48f1b14fc7 test: fix NIXL EP Mooncake FT test (#32638) Yoray Zack 2026-08-06 13:00:18 +03:00
• ba9074035a Build and release sgl-deep-ep wheels (#33498) Baizhou Zhang 2026-08-06 02:07:36 -07:00
• 8b29c90218 [NVIDIA] Enable CuTe DSL BF16 GEMM on SM107 (#33617) YAMYandLee Nau 2026-08-06 02:06:08 -07:00
• dfe53232d7 [AMD] Move test_load_weights_from_remote_instance.py to extra CI (#33809) Bingxu Chen 2026-08-06 16:57:31 +08:00
• 4cdab7b4f4 [AMD] Add msgpack to ROCm diffusion deps (fix multimodal-gen unit test ModuleNotFoundError) (#31899) kangwangamd 2026-08-06 16:52:23 +08:00
• fe55d78b7d Fix MXFP4 scale placeholder initialization (#33500) weireweireandweireweire 2026-08-06 16:46:54 +08:00
• efc99a86ff [CI] Restore the full prefill CUDA graph capture range in test launches (#33847) Liangsheng Yin 2026-08-06 01:40:10 -07:00
• 0e584529f5 [CI] Remove profiling from nightly tests (#33832) Baizhou Zhang 2026-08-06 01:16:08 -07:00
• c11ce7c514 chore: bump sgl-kernel version to 0.4.6.post1 (#33842) sglang-botandsglang-bot 2026-08-06 01:09:48 -07:00
• 04374ba5e0 Update AOT kernels for Torch 2.13 (#33841) Mohammad Miadh Angkad 2026-08-06 16:03:26 +08:00
• b8140f36ea Fix lint error (#33830) YC Yen-Ching Tseng 2026-08-06 15:53:40 +08:00
• bfce378e5f [diffusion] feat: capture-safe pynccl all-to-all (#33775) MickandClaude Fable 5 2026-08-06 15:41:05 +08:00
• 31c1e5943f Facade DSA index-cache: MTP topk-reuse state + index-K storage (#28609) Xinyuan Tong 2026-08-06 15:34:31 +08:00
• 735995e7bd Implement random tie breadking for cache_aware sglang router policy (#33138) William ChangandPatrick Toulme 2026-08-06 00:31:40 -07:00
• cf7923615c [AMD] Gate DFLASH non-greedy verify on the target-only kernel being registered (#33694) YC Yen-Ching TsengandHaiShaw 2026-08-06 15:10:22 +08:00
• 5d1a0c7129 Revert "Warn on risky serving-time Triton work" (#33826) Baizhou Zhang 2026-08-05 23:36:10 -07:00
• 32e5d788bd [mm] rust-server: native multimodal processing for Qwen VL (integrate sglang-mm, e2e) (#32365) 3 people 2026-08-05 23:35:50 -07:00
• dea07b348b [AMD] Update amd k3 cookbook for fp8 kv cache (#33825) Thomas Wang 2026-08-06 14:35:38 +08:00
• 8e11feb68e [HiCache] Support packed and sidecar draft caches for MTP/EAGLE/DSpark (#30393) 4 people 2026-08-06 14:31:11 +08:00
• c84ddc0e76 docker: pin nightly image source to workflow commit (#33738) ishandhanani 2026-08-05 23:26:18 -07:00
• b6876fc652 [diffusion] ERNIE-Image bit-exact residual-gate fast path (H200 1024^2 e2e 16.17 -> 15.75 s) (#33734) Xiaoyu ZhangandClaude Fable 5 2026-08-06 13:43:09 +08:00
• 604d3561b0 [diffusion] feat: data-parallel serving (--dp-size) (#33725) MickandClaude Fable 5 2026-08-06 11:50:13 +08:00
• 407a65d3cb [AMD] [CI] Track the MI355X disagg nightly in the AMD CI job monitor (#33753) Michael 2026-08-05 20:02:22 -07:00
• 9bd1461757 [CI] Bound the CUDA graph capture range in test launches and lift the spec fixture's admission cap (#33776) Liangsheng Yin 2026-08-05 20:00:30 -07:00
• f33f6a522f Relax GDN ReplaySSM fold test for Triton 3.7 (#33780) Mohammad Miadh Angkad 2026-08-06 10:44:14 +08:00
• 4ea227fa91 config: the draft runner carries its own attention backend Cheng Wan 2026-08-05 19:32:24 -07:00
• 64eeb153df config: resolve the draft worker's config per runner, not on a copy Cheng Wan 2026-08-05 19:31:23 -07:00
• 99cfc90658 config: retire ServerArgs.override in favour of derive() Cheng Wan 2026-08-05 19:30:53 -07:00
• d33ab39ebc config: template-detected parsers go to the engine's control-plane overlay Cheng Wan 2026-08-05 19:29:51 -07:00
• bebebb8f6c config: retire the alias-form process-global config reads Cheng Wan 2026-08-05 19:29:20 -07:00
• 1d47952c7c config: pass the Ray placement group as a launch argument Cheng Wan 2026-08-05 19:28:39 -07:00
• f9b954ddb1 [CI] Temporarily disable prefill cuda graph for qwen3.5 nightly test (#33772) Baizhou Zhang 2026-08-05 18:52:57 -07:00
• d90ef69802 [AMD] ci: run vetted nested multimodal_gen unit tests on AMD (#31483) Michael 2026-08-05 18:39:33 -07:00
• d9b1cba255 Fix DeepSeek-OCR batching crash on variable local-crop counts (#33214) Pavan Sivaram Girijala 2026-08-06 07:09:19 +05:30
• f01f706960 [RL] Skip rotary cache tensors in weight checker (#27692) Zilin Zhu 2026-08-06 08:22:12 +07:00
• ceaeca0b9e [Fix] Vocab out of bounds in DSpark for Inkling-Small (#33748) Eric Zhang 2026-08-05 21:20:02 -04:00
• fc74c35546 kernels: scalar scale A support for fp8_gemm (#33469) DevashishLal-CBandDevashish Lal 2026-08-05 18:11:25 -07:00
• 269d51ed4b [Fix] Inkling works with gs:// runai_streamer paths (#33750) Eric Zhang 2026-08-05 21:09:17 -04:00
• 28848bfe7c Fix Nightly NV CI (#33564) 3 people 2026-08-05 18:06:47 -07:00
• e675c7226a [AMD]Stage MI355X nightly by node count (#33774) YC Yen-Ching TsengandChen 2026-08-06 09:04:29 +08:00
• beabc5949b Enable MoE deferred finalize by default and drop its expert_weights dtype workaround (#33618) Khoa PhamandClaude Opus 5 2026-08-05 17:56:47 -07:00
• c9506d023f [Quantization] Route per-tensor FP8 checkpoints to FlashInfer on SM90 (#33148) Aditya KamatandBrayden Zhong 2026-08-05 17:54:10 -07:00
• 4c0a8940fa [Kernel] Unify BaseFusedOp and MultiPlatformOp dispatch (#33205) Xiaoyu ZhangandClaude Fable 5 2026-08-06 08:52:09 +08:00
• ba12a16a62 [diffusion] Prefer cuDNN SDPA over FA4 for dense attention on sm_100 (B200) (#33655) Xiaoyu ZhangandMohammad Miadh Angkad 2026-08-06 08:49:10 +08:00
• c952ee5ac1 chore: bump sgl-kernel version to 0.4.6 (#33678) sglang-botandsglang-bot 2026-08-05 17:46:54 -07:00
• 65d5a0ec25 Support ModelOpt MXFP8 checkpoints (#32538) Mohammad Miadh Angkad 2026-08-06 08:44:32 +08:00
• ae5f8c94b7 ci(xpu): harden nightly + PR XPU CI (HF login / tag fetch / docker push retries) (#32390) ashwini rathi 2026-08-06 05:55:53 +05:30
• 3869fe556f [CI] Collapse the EAGLE launch matrix and the scoring engine boots on the per-commit runners (#33756) Liangsheng Yin 2026-08-05 16:43:19 -07:00
• 5424d2039c [CI] Fix GLM-Image usage unit tests (#33731) Mohammad Miadh Angkad 2026-08-06 07:20:08 +08:00
• 070fde72bf [CI] Remove some unneeded CP tests (#33763) Baizhou Zhang 2026-08-05 16:07:14 -07:00
• a1cc286062 [Intel GPU] DeepSeek V4 4/N: use sgl-kernel implementation of fused_q_norm_rope on XPU (#27790) Polisetty V R K Jyothendra Varma 2026-08-06 04:34:35 +05:30
• c0ef548eef [misc] Unify MLA scaling init and remove dead buffer / scaling code (#33363) Liangsheng Yin 2026-08-05 15:59:33 -07:00
• 990a446773 Fix padded positions in breakable CUDA Graph attention (#33253) YAMY 2026-08-05 15:53:28 -07:00
• 7bc90ab394 Select DeepGEMM standard layouts by memory budget (#33474) YAMYandChunan Zeng 2026-08-05 15:50:43 -07:00
• 25035bff8d Use main branch in Kimi Dockerfiles (#33760) Baizhou Zhang 2026-08-05 15:33:14 -07:00
• 02cd44c59a feat(dgx-spark): add inkling-small MoE support for sm_121 (#33108) yvbbrjdrandhnyls2002 2026-08-05 15:06:59 -07:00
• b9d572ee02 [test] Re-enable a pruned Inkling LoRA unit-test set (68 -> 9 cases) (#33752) Yanbin Jiang 2026-08-05 15:06:28 -07:00
• 2d27133fcf [CI] Skip apt-get when the required packages are already installed (#33757) Liangsheng Yin 2026-08-05 15:06:01 -07:00
• a3a1ebc7b7 Warn on risky serving-time Triton work (#33120) cctry 2026-08-05 14:55:03 -07:00
• 988c6e6aeb Pin online NVFP4 4over6 quantization settings (#33621) Ziang Li 2026-08-05 14:29:32 -07:00
• a14c870886 Fix broken Nemotron DP attention (#33123) Brayden ZhongandBrayden Zhong 2026-08-05 14:27:05 -07:00
• 9436de717f [Spec][PD] Enable fused TopK for GLM-5.2 MTP IndexShare (#31477) Hank Han 2026-08-06 05:17:42 +08:00
• 106bcc1293 Observability enhancement for HiCache (#32388) Zhiqiang Xie 2026-08-05 14:13:06 -07:00