Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• e061dd1b47 [Diffusion][Kernel] Fuse Wan FFN GELU epilogue (#36592) Xiaoyu Zhang 2026-08-27 21:06:28 +08:00
• 1af95ffded [diffusion] Fuse Cosmos3 Nano T2I attention on Hopper (#36571) Xiaoyu Zhang 2026-08-27 20:59:15 +08:00
• db4125bb56 [diffusion] Accept mesh benchmark artifacts (#36553) Xiaoyu Zhang 2026-08-27 20:53:25 +08:00
• d42fa5e10a [diffusion] align video BCG warmup frame count (#36485) Xiaoyu Zhang 2026-08-27 20:52:00 +08:00
• b647ae82f5 [NPU][Bugfix] Fix OOB gather in decode KV allocation when free pool is tight (#35727) xdtbynd 2026-08-27 19:44:45 +08:00
• 97ba99067d Publish per-scheduler load on a dedicated socket for load-aware routers (#34608) 3 people 2026-08-27 19:33:18 +08:00
• 94183a8d2b Move server args config parser under utils (#36681) Lianmin Zheng 2026-08-27 03:34:50 -07:00
• 536f570e66 docs(cookbook): fix Qwen3.8 Flash Next H200 MTP verify with BF16 SSM state (#36611) Julian Huang 2026-08-27 17:49:03 +08:00
• ad911a5ec0 [kernel] Tune LingBot-Video MoE TMA configs for H100 (#36543) Xiaoyu Zhang 2026-08-27 17:47:32 +08:00
• 08315c56df [Fix] Account resident weight memory in KV sizing (#34053) Schwinn Saereesitthipitak 2026-08-27 02:43:04 -07:00
• 56fdfc3b26 XPU: remove SGLANG_USE_SGL_XPU flag (#34492) Xia Weiwen 2026-08-27 02:38:02 -07:00
• 78d36f5f62 fix: kill_process_tree waits for the reap by default (#36589) Khoa Pham 2026-08-27 02:34:05 -07:00
• 2ded8a6aea [AMD] Enable moe_a2a_backend=mori for DeepSeek-V4 prefill context parallelism (#35611) AMD-yanfeiwang 2026-08-27 17:31:41 +08:00
• 636a6f7dba cookbook: fix GLM-5.3-Flash speculative flag, size Hopper memory, record GSM8K (#36660) zijiexia 2026-08-27 02:19:47 -07:00
• 3402265989 [Core] Refactor server argument choices (#36586) Lianmin Zheng 2026-08-27 01:56:03 -07:00
• a126a5fa31 [CI] Graceful teardown for the radix_cache server fixtures (#36605) Alison Shao 2026-08-27 01:36:04 -07:00
• c2c3320cf0 [VLM] feat: size the multimodal preprocessing pool by where preprocessing runs (#35349) MickandClaude Opus 5 2026-08-27 16:32:58 +08:00
• c608f9bf75 [CI] Fix Q8KV8 sparse prefill test fixture (#36639) Po-Han Huang (NVIDIA) 2026-08-27 16:00:22 +08:00
• b8a6adadfe [Bug][Spec] fix startup crash and reduce CUDA graph memory usage for speculative adaptive (#35275) 4 people 2026-08-27 15:46:13 +08:00
• 1a91c232ea [AMD][CI] Add targeted Mori test labels (#36636) inkcherry 2026-08-27 15:24:02 +08:00
• c967cd19b5 [AMD] Optimize Qwen3.5 MTP unified attention on gfx950 (#36330) jacky.cheng 2026-08-27 15:13:49 +08:00
• 72bf8c4d53 [ci] xpu: trigger pr-test-xpu on multimodal_gen changes (#36100) 3 people 2026-08-27 12:20:26 +05:30
• adcf73d7f7 [Weight Cache] Enhance test and support EPLB (#36198) Xun Sun 2026-08-27 14:11:53 +08:00
• b294bd4bc7 [CPU][CI]: fix a few issues that cause XEON CI failures (#36413) 4 people 2026-08-26 22:59:08 -07:00
• ea48cb04cc Pin scheduler metadata before asynchronous H2D copies (#35944) weireweireandweireweire 2026-08-27 13:21:47 +08:00
• 0f7b5b8b2a [AMD] Add GLM-5.3-Flash recipes for MI300X, MI325X, and MI355X (#36608) andyluo7 2026-08-26 22:19:57 -07:00
• f775db03aa [AMD] Fix int32 seqused_k overflow in aiter draft-extend attention (#36541) jacky.cheng 2026-08-27 13:06:39 +08:00
• 4d5d506486 Publish gated DSV4 DFLASH-family target-prefill read completion (#35947) weireweireandweireweire 2026-08-27 12:33:08 +08:00
• 7324021e6c [CI] Grant no-cooldown for UNIDY2002 (#36609) Xun Sun 2026-08-27 12:13:59 +08:00
• e6c1a3cd4c docs: sync LMSYS SGLang blog cards (#35416) sglang-botandsglang-bot 2026-08-26 20:56:54 -07:00
• 02dfd37826 [CI] Remove GLM-4.1V-9B-Thinking from nightly VLM MMMU eval (#36602) Alison Shao 2026-08-26 20:29:38 -07:00
• a2b589fdf8 [Radix Cache] Add test-only TreeCore inspector for shared backend tests (#35791) Jialin Ouyang 2026-08-26 20:17:51 -07:00
• a3ae667d67 [Feature] Add DeepEPv2 (ElasticBuffer) MoE A2A backend (#35634) 5 people 2026-08-27 10:54:33 +08:00
• cbfe54fba8 [AMD] Use fast exponentials in C4 and C128 ROCm kernels (#34296) AMD-yanfeiwang 2026-08-27 10:49:01 +08:00
• 1c8f2b38cb [AMD][CI] Add DeepSeek-V4-Flash FP8 accuracy coverage on MI30x (#36396) Michael 2026-08-26 19:30:07 -07:00
• 07570202f0 [VLM] route every multimodal processor through the worker pool's call site (#35342) MickandClaude Opus 5 2026-08-27 10:27:45 +08:00
• 5adc2880f9 [CPU] Enable ERNIE models on CPU (#35222) Zaili Wang 2026-08-27 10:13:10 +08:00
• 9d07b9e227 [Performance] Reduce idle DP work in breakable prefill CUDA graphs (#33871) Promisewjx 2026-08-27 10:09:54 +08:00
• 4f59a8dcfa [AMD][Bugfix] Skip invalid fused MoE reduction for direct top-1 output (#36309) Bingxu Chen 2026-08-27 09:52:15 +08:00
• b658d64d5b [CPU] Fix rotary_embedding_cpu fake for in-place layouts (#36443) Haotong Zou 2026-08-27 01:48:07 +00:00
• d2b3e9051d [AMD][CI] Stabilize PyTorch sampling backend test on ROCm (#36307) Bingxu Chen 2026-08-27 09:44:10 +08:00
• 058f99c561 [CPU] Fix truncated KV prefix in intel_amx spec verify (#36430) Haotong Zou 2026-08-27 01:42:36 +00:00
• 15688fea7d Fix BailingMoeV3 reading enable_dp_lm_head off live topology instead of config (#36584) Alison Shao 2026-08-26 18:41:39 -07:00
• a8b72c4c34 [Intel XPU] Fix cross-encoder rerank hang on B580 runners (#36360) ashwini rathi 2026-08-27 07:03:20 +05:30
• 862f9ed3d8 [AMD] Fall back to CPU tensor for decode retraction on ROCm (#36343) Bingxu Chen 2026-08-27 09:32:51 +08:00
• 20621aa14b [Model] Support Ling-3.0-flash (BailingMoeV3) (#33561) 9 people 2026-08-27 08:27:23 +08:00
• 8739d56a31 [Spec] Generalize hybrid SWA MTP draft pool routing (#35379) paulzhang-tm 2026-08-26 20:02:36 -04:00
• ec4bdbfa4a [Feature] Beam search support (#31626) 5 people 2026-08-26 16:56:15 -07:00
• e5a1c5a423 Fix _is_compiling dynamo tracing: import torch instead of sys.modules lookup (#36573) Cheng Wan 2026-08-26 16:17:02 -07:00
• 7f27bf4708 [Kernel] Declare the PyTorch ABI dependency in sglang-kernel wheels (#36465) kuma_Gu2006 2026-08-27 06:24:26 +08:00
• 1eb629ab4f [NVIDIA] Tune custom all reduce v2 for sm_107 (#36397) Trevor Morris 2026-08-26 15:19:31 -07:00
• 3ce243da3f [NVIDIA] Add CUDA 13.4 container for initial Rubin support (#36233) Trevor Morris 2026-08-26 15:02:16 -07:00
• 06694071c6 [Spec] Avoid tensor scalar reads in spec decode allocation (#35377) paulzhang-tm 2026-08-26 17:51:31 -04:00
• 45c85c198b [CI] Lower the AWQ Marlin MMLU threshold to 0.80 (#36570) Liangsheng Yin 2026-08-26 14:43:20 -07:00
• a8d716ce8d dsa: widen the fp8 k-cache quant kernel's token_id to int64 (#30859) Xinyuan Tong 2026-08-27 05:22:24 +08:00
• 2935bb8e79 Fix OOB read in mxfp4 MoE weight scales on Hopper (#36456) Mohammad Miadh AngkadandMohammad Angkad 2026-08-27 04:36:49 +08:00
• e7e7894016 [Weight Cache] Make daemon socket/ready paths configurable via env (#36299) siyu 2026-08-27 03:38:50 +08:00
• 5263568bcb [HiCache] Keep auxiliary load-back out of Full KV pending ownership (#36317) weireweireandweireweire 2026-08-27 01:30:38 +08:00
• e27a7fac77 GLM-5.3-Flash cookbook: default Blackwell recipes to FP8 KV + TRT-LLM DSA (#36519) Xinyuan Tong 2026-08-27 00:51:05 +08:00
• f8cc1f9525 GLM-5.3-Flash cookbook: FP8 KV + TRT-LLM DSA benchmark card (#36513) Xinyuan Tong 2026-08-26 22:39:27 +08:00
• c8b56b1f44 chore: bump mooncake version to 0.3.13 (#36493) Shangming Cai 2026-08-26 22:21:15 +08:00
• 924aeee59c [diffusion] feat: support batching for cosmos3 action generation (#36301) XuFuandMick 2026-08-26 22:15:03 +08:00
• dfc40e0efe Add GLM-5.3-Flash cookbook (#36440) Xinyuan Tong 2026-08-26 22:00:16 +08:00
• 8eaffdf382 docs: point the Qwen3.8-Flash-Next cookbook at model support PR #36497 (#36499) Yuhao Yang 2026-08-26 20:53:55 +08:00
• c7b5e76fa9 Add Qwen3.8-Flash-Next cookbook (#36496) 3 people 2026-08-26 20:36:59 +08:00
• 413df1f8db config: ServerArgs holds the raw input (#36255) Cheng Wan 2026-08-26 05:14:05 -07:00
• 27c36368b6 fix(moe): guard FP8 delegate activation params (#36275) 4 people 2026-08-26 20:12:21 +08:00
• 937af8538b config: the runtime readers take the published bags (#36254) Cheng Wan 2026-08-26 05:08:25 -07:00
• 5b7fc61306 config: resolution reads the declarations, not the fields (#36253) Cheng Wan 2026-08-26 05:05:28 -07:00
• ae5feb4b9c config: stop handing the record to code that does not read it (#36252) Cheng Wan 2026-08-26 05:02:17 -07:00
• d7b144f64e config: publishing is the process entry's job (#36251) Cheng Wan 2026-08-26 04:58:41 -07:00
• 702de26310 [diffusion] make benchmark caches seedable and cover missing native families (#36463) Xiaoyu Zhang 2026-08-26 19:04:46 +08:00
• 170da72c13 [diffusion] perf: fuse tanh-GELU into the LongCat-Image DiT FFN up-proj (#36322) 3 people 2026-08-26 19:04:06 +08:00
• 8005df61d3 config: spell the parallel config tier at the call site (#36250) Cheng Wan 2026-08-26 03:00:28 -07:00
• 689ade69d1 [kernel] Split the custom all-reduce communicator into push/pull planes (#35735) DarkSharpness 2026-08-26 17:40:28 +08:00
• 58ecbba0bd [Feature] Coordinate FullCG prefill across DP-attention ranks (#35640) Aurick QiaoandYuwei An 2026-08-26 02:16:02 -07:00
• 2511743bd7 [unified-memory] Stop eviction when shared allocation capacity is sufficient (#33091) SuperSongandseokwoosong 2026-08-26 18:06:32 +09:00
• ffc431cd4c [CI] Fix stale tool_call_parser key in Spark2.5 detector test (#36464) Mohammad Miadh AngkadandMohammad Angkad 2026-08-26 16:34:39 +08:00
• a3c4936438 Sync FlashInfer autotune tactic choice across TP ranks (#35343) Mohammad Miadh AngkadandMohammad Angkad 2026-08-26 16:30:47 +08:00
• bede6bc37c Fix unified HiCache PP test import (#36454) Jialin Ouyang 2026-08-26 00:35:18 -07:00
• c3c529c28e [diffusion] docs: distinguish MiniMax H3 checkpoint variants (#36412) Mick 2026-08-26 15:16:18 +08:00
• 7ef49e8f7d Rename Spark3 to Spark2.5 (#36416) KnightYao 2026-08-26 15:04:46 +08:00
• 3ce4f957eb [diffusion] fix: fix MiniMax-H3 dp_size>1 deadlock and cross-request audio determinism (#36398) 4 people 2026-08-25 23:54:08 -07:00
• abe3aeb142 [NPU] [DOC] Update CANN version in NPU installation docs (#36325) amote-i 2026-08-26 14:51:43 +08:00
• 3c9febc68b [Spec][DSA] Add --speculative-dsa-topk-backend (#36313) Ziang Li 2026-08-25 23:35:03 -07:00
• 0c84eaba7f [CI] Grant no-cooldown and tag permissions to seven contributors (#36445) Khoa PhamandClaude Opus 5 2026-08-25 23:23:38 -07:00
• db15976fe9 Fix DSV4 DSpark sample-from-anchor initialization (#36419) weireweireandweireweire 2026-08-26 14:18:58 +08:00
• 04c1036bb3 [Kernel] Skip reserved writes in MLA KV cache (#36003) Tan TrinhandKhoa Pham 2026-08-26 12:42:23 +07:00
• d7baad0116 [diffusion] Keep the Cosmos3 Super DiT resident on high-memory GPUs (#36375) Yihao WangandClaude Opus 5 2026-08-25 21:10:39 -07:00
• 3ec22948c1 [CI] /rerun-failed-ci: rerun cancelled runs and target the newest run per workflow (#34057) Alison Shao 2026-08-25 20:15:54 -07:00
• 34de1fb47f fix(test): stabilize nightly precision regression (#34668) 3 people 2026-08-26 11:04:52 +08:00
• 2e4773aadd Fix NPUMHATokenToKVPool missing k_data_ptrs/v_data_ptrs (#35820) ming_wang 2026-08-26 11:01:11 +08:00
• 5bcc978f2a [AMD][CI] Pass USE_PDL explicitly in the fused MoE gate (#36306) Bingxu Chen 2026-08-26 10:57:59 +08:00
• fa3ac61661 [Diffusion] Bound reusable Ulysses A2A staging buffers across shapes (#36327) Xiaoyu Zhang 2026-08-26 10:57:29 +08:00
• cc3b61873f [Diffusion][minimax-h3] Restrict MiniMax-H3 SubBlock sparsity to video queries (#35850) HuangJi 2026-08-26 10:55:09 +08:00
• 4382947b58 Fix DSV4 shared-fusion CPU unit test after the EP guard landed (#36424) Alison Shao 2026-08-25 19:52:05 -07:00
• 303227951a fix: use a bf16-relative tolerance in the DSA indexer K kernel test (#35795) Shuwen WangandClaude Opus 5 2026-08-26 10:50:29 +08:00
• 07a9de25b4 [AMD][CI] Fix shared-KV verify tests for multi-head GQA (#36296) Bingxu Chen 2026-08-26 10:46:27 +08:00
• 4ae30dc736 fix: detect cross-node multimodal transport by nnodes (#35646) Yuanle Liu 2026-08-26 10:31:48 +08:00
• 2d8484740d Support deepseek v4 and kimi k3 on ssd (#35314) 5 people 2026-08-26 10:05:12 +08:00