Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• bd3b252e0a [CPU] use faster exp in silu_and_mul (#29382) Ma Mingfei 2026-06-29 09:38:30 +08:00
• 909123ddb8 [misc] Use --cuda-graph-max-bs-decode in tests, examples, and docs (#29591) Liangsheng Yin 2026-06-28 18:38:28 -07:00
• 3217410cf6 [CPU] enable fused_sigmoid_mul on CPU device (#29378) Ma Mingfei 2026-06-29 09:38:04 +08:00
• d5abafcc1c [Intel GPU] add pytorch profiling support for XPU in bench offline throughput and enhance num steps (#28308) Polisetty V R K Jyothendra Varma 2026-06-29 06:40:15 +05:30
• abaee4664e [dflash] fa3/fa4: device-side page table; drop seq_lens_cpu D2H sync (#29343) Liangsheng Yin 2026-06-28 18:06:33 -07:00
• bd2a5db987 [HiCache] Detect for double-free in HostKVCache (#29310) Chao Shi 2026-06-29 09:02:21 +08:00
• b9b860652e Fix DSA indexer fusion bug causing excessive memory consumption. (#29576) Brayden ZhongandBrayden Zhong 2026-06-28 14:11:07 -07:00
• ad30a9958e Fix dummy weight init for tensor subclasses (#29229) Aurick Qiao 2026-06-28 10:17:00 -07:00
• c5b9388721 [scheduler] Add scheduler metrics reporter init hook (#29535) Yinghai Lu 2026-06-28 09:59:34 -07:00
• aaa31eb0a1 feat: first-class session identity in SGLang (#29436) ishandhanani 2026-06-28 16:16:38 +02:00
• 828411e6f1 Fix FlashInfer A2A dispatcher during CUDA graph capture (#29461) Lianmin Zheng 2026-06-28 02:21:26 -07:00
• 4cb6d81bda [diffusion] CI: make consistency GT probe robust to transient CDN failures (#29545) Mick 2026-06-28 17:13:53 +08:00
• 6582fc1e5d dflash: drop verify_done barrier; rely on scheduler WAR fallback (#29556) Liangsheng Yin 2026-06-28 01:47:17 -07:00
• b1da2082ca chore: bump sgl-deep-gemm build-time apache-tvm-ffi 0.1.9 -> 0.1.11 (#28481) Martin Hua 2026-06-28 00:26:24 -07:00
• 593f5ba68f [Spec] Replace shared-infra dflash special-cases with capabilities (WAR barrier + seq_lens_cpu) (#29232) Liangsheng Yin 2026-06-28 00:02:32 -07:00
• 97ddc66209 [diffusion] nightly: track SGLang-Diffusion only (#29434) Mick 2026-06-28 14:23:17 +08:00
• 643e1cc779 fix: fix prefill-aware SWA floor tracking (#29520) Mick 2026-06-28 14:18:28 +08:00
• ddc389cf09 [minimax-m3] Split 3/4: disagg K-only index-K transfer (#28714) Xinyuan Tong 2026-06-28 06:54:42 +01:00
• 6eedc8f376 Fix EAGLE draft hidden dim extraction and centralize spec helpers (#29464) Lianmin Zheng 2026-06-27 21:48:26 -07:00
• 5747ed3b19 Fix DP-attention SHM feature finalization race (#29543) Lianmin ZhengandYinghai Lu 2026-06-27 21:36:58 -07:00
• ae09b8302f [CI] Fix GB300 DSV4 Pro FP4 nightly (#29502) Mohammad Miadh Angkad 2026-06-28 12:06:17 +08:00
• 4a76699dfc [diffusion] fix: fix --warmup silently downgrading server-based warmup to request mode (#29514) Mick 2026-06-28 11:46:07 +08:00
• e6cbc8f5fe Fix SWA cache loc slicing for all attention backends (#29460) Lianmin Zheng 2026-06-27 20:37:58 -07:00
• 4e6ce8e4ab [Spec] Capture DFLASH draft greedy sampling inside the draft decode cuda graph (#29395) Liangsheng Yin 2026-06-27 19:34:26 -07:00
• 3d922a51fc test: remove multimodal piecewise CUDA graph gate test (#29537) sglang-botandClaude Sonnet 4.6 2026-06-27 19:03:32 -07:00
• da802ddcaf (perf): Shard Kimi-K2.5 Eagle3 draft fc + symm-mem AG (#29223) Khoa PhamandCursor 2026-06-27 17:40:32 -07:00
• 073de15053 Fuse the DSA (V3.2, GLM-5.x) indexer Q/K paths into single kernels (#27705) 3 people 2026-06-27 14:29:56 -07:00
• e4253b39e2 Support JIT fused A GEMM (MLA down projection) and support GLM-5 hidden size, SM120 (#27397) Brayden ZhongandBrayden Zhong 2026-06-27 13:32:06 -07:00
• 592f6c849b [minimax-m3] Split 2/4: mem-cache / HiCache / sparse KV pool (#28713) Xinyuan Tongandhzh0425 2026-06-27 17:19:38 +01:00
• cfd911ad6e docs: refine diffusion cookbook overview (#29507) Mick 2026-06-28 00:10:21 +08:00
• 1589603114 model: support baidu unlimited-ocr (#29186) Aditya KamatandMick 2026-06-27 08:36:19 -07:00
• b030b1a5f3 hisparse: support NIXL DRAM KV destinations for HiSparse (#27563) 3 people 2026-06-27 16:32:29 +02:00
• c1b5c7e499 Fix DeepSeek V4 PP HiCache SWA allocation and layer mapping (#29106) 3 people 2026-06-27 22:19:14 +08:00
• 2f34dbe372 Add native Exa-backed web_search support (#29342) jonah-bermanandXinyuan Tong 2026-06-27 06:41:09 -07:00
• 3306233961 [diffusion] optimize: optimize LTX2.3 CFG path (#28624) Mick 2026-06-27 19:23:14 +08:00
• 91f9e7372d [diffusion] CI: refactor CI (#28762) Mick 2026-06-27 19:22:27 +08:00
• a3c5e286f6 [NPU] [DOC] Fix and update Ascend NPU docs (#29501) amote-i 2026-06-27 18:23:11 +08:00
• e0c0c0a45c [Cookbook] GLM-5.2: tune GB300 NVFP4 recipes + fill benchmarks (#29486) zijiexiaandClaude Opus 4.8 2026-06-27 01:01:28 -07:00
• 9214b9338f [DSA] Enable draft-extend CUDA graph for DeepSeek Sparse Attention (#29413) 3 people 2026-06-26 23:53:46 -07:00
• 81d3c3ce77 [AMD] [GLM5] Guard cuda_runtime.h for ROCm in fused_metadata_copy (#29373) Raiden MakotoandRaiden-Makoto 2026-06-26 23:29:19 -07:00
• c36f166364 [skill] Remove outdated llm-serving-auto-benchmark skill (#29487) Xiaoyu Zhang 2026-06-27 14:19:11 +08:00
• 43435a2f8e [mori] Add a combine-kwargs hook and use_external_inp_buf plumbing (#29097) Chengze Fan 2026-06-26 22:34:27 -07:00
• 19abebcc6a [server_args] Make MoRI per-rank dispatch-token requirement overridable (#29096) Chengze Fan 2026-06-26 22:25:36 -07:00
• 495f13fa12 [KDA-Pilot] Add diffusion residual-gate CUDA fast path for LTX2 (#29361) Xiaoyu ZhangandClaude Opus 4.8 2026-06-27 12:59:41 +08:00
• cd6dedf972 [router] Count requests/responses at the HTTP edge for true intake (#29467) Liangsheng Yin 2026-06-26 21:21:41 -07:00
• 09ca4fc96b Skip FlashInfer FP8 autotune for MXFP8 quantized models (#29462) Lianmin Zheng 2026-06-26 18:15:08 -07:00
• 13b5bd962a fix(spec): track current_token in ReasonerGrammarObject (#29102) IvanShan177andClaude Opus 4.8 2026-06-27 07:43:35 +08:00
• da0f4f6f92 [HiCache] remove large host mem constraint (#28614) 3 people 2026-06-26 16:20:48 -07:00
• 12f76d115c Update GLM-5.2 B300 and GB300 NVFP4 cookbook settings (#29466) Baizhou Zhang 2026-06-26 16:15:33 -07:00
• c470acde2b [AMD] Register 1 kernel unit test for AMD nightly CI (#29333) Michael 2026-06-26 16:06:02 -07:00
• 9c3227b689 Fix IPv6 wildcard bootstrap address resolution in disagg (#29459) Lianmin Zheng 2026-06-26 16:05:36 -07:00
• 5eebb4b2e8 [AMD] Fix fused append+remap DeepEP equivalence test on aiter path (#29377) YC Yen-Ching Tseng 2026-06-27 07:04:20 +08:00
• 72812db138 Avoid dynamic Q quantization in trtllm_mha (#29423) Po-Han Huang (NVIDIA) 2026-06-27 06:51:15 +08:00
• 73741f7074 Bypass legacy GLM DSA layer types validation (#29454) Baizhou Zhang 2026-06-26 15:24:16 -07:00
• 267d165ad0 shm_broadcast: retry bind on EADDRINUSE (fix dp-attention port race) (#29455) 3 people 2026-06-26 14:49:39 -07:00
• e745b3af22 [Fix] compressed-tensors block FP8: requantize weight scales to UE8M0 for DeepGEMM on Blackwell (#28662) Jimmy Shong 2026-06-26 14:41:18 -07:00
• 7b02eab7a6 Revert "[DeepSeek V3] Run routed experts on main stream in dual-stream MoE" (#29452) Cheng Wan 2026-06-26 13:59:36 -07:00
• ee77a7d330 [Fix] DSA: size cudagraph page_table to req_to_token width (#29379) Jackey HuaandClaude Opus 4.8 2026-06-26 12:17:34 -07:00
• be1930133a Convert IPC dataclasses to msgspec.Struct with opt-in msgpack transport (#28688) Rain JiangandLianmin Zheng 2026-06-26 12:04:03 -07:00
• 714011a40f [JIT Kernel] Migrate dsv3_router_gemm from AOT sgl-kernel to JIT kernel (#21531) 3 people 2026-06-27 02:52:18 +08:00
• c98d31143d update quantization code owner and document quantization contributions (#26784) Yaochen Hanandronnie_zheng 2026-06-27 00:40:33 +08:00
• 8524678889 Fix MiniMax MSA fallback when fmha plan is unavailable (#29250) Xiaoyu Zhang 2026-06-26 23:14:31 +08:00
• 18b0e5757e [Diffusion] Fuse LTX2 Ada values (#29390) Xiaoyu Zhang 2026-06-26 23:13:15 +08:00
• b91348071e [diffusion] optimize: shard qwen text embed in sp (#29147) Aleksi Vesanto 2026-06-26 16:44:45 +03:00
• b73e57210a [AMD CI] Add nightly Miles ROCm 7.2 MI350X suites (#28853) 3 people 2026-06-26 06:47:00 -04:00
• f5708c3f91 [NPU] Fix mllama cross-attention crash in ascend extend SDPA (#29374) syy-hw 2026-06-26 17:13:17 +08:00
• c0d4d1897c fix: batch BlockRemoved events per radix node (#29265) Yan Ru Pei 2026-06-26 01:13:40 -07:00
• c21c2d9421 [Docs] Cookbook: match playground docker image resolution to deployment (hw|quant) (#29400) zijiexiaandClaude Opus 4.8 2026-06-26 00:56:34 -07:00
• 999199f9ff [DeepSeek V3] Run routed experts on main stream in dual-stream MoE (#29142) 3 people 2026-06-26 00:48:59 -07:00
• aeb4e98108 [diffusion] model: support JoyEcho multi-shot A/V generation support (#27420) 3 people 2026-06-26 15:46:31 +08:00
• dd56a9f069 [Docs] Add NVFP4 quantization to GLM-5.2 cookbook (#29380) zijiexiaandClaude Opus 4.8 2026-06-26 00:40:13 -07:00
• 5996b54bd3 [KDA-Pilot] Add diffusion causal Conv3D cat-pad CUDA fast path for Cosmos3 (#29281) Xiaoyu ZhangandClaude Opus 4.8 2026-06-26 15:06:49 +08:00
• a10a24e9a7 [GDN][KDA] ReplaySSM buffered output-only decode for Linear Attention (#28451) Yuan Luoandluoyuan.luo 2026-06-26 14:17:02 +08:00
• dc113e8804 [Intel GPU] DeepSeek V4 3/N: Support hc_split_sinkhorn on XPU using sgl_kernel (#27783) Polisetty V R K Jyothendra VarmaandMa Mingfei 2026-06-26 11:27:13 +05:30
• eeee3abbbf [CI] Fix false spec accept length failure after profiling phase (#29329) Khoa PhamandCursor 2026-06-25 22:19:30 -07:00
• ff89e0fe9a Fix KV event publisher bind conflict under PP (#29044) ziang663 2026-06-26 12:50:49 +08:00
• 8fd6e9e017 Fix fixed-size HiCache capacity under PP (#29258) ziang663 2026-06-26 12:49:23 +08:00
• b6ebdcc92e [test] Split perf table out of fwd occupancy kit report (#29348) Liangsheng Yin 2026-06-25 20:23:42 -07:00
• cc294829aa [NPU] fix best practicce docs (#29303) jianzhao-xu 2026-06-26 11:17:50 +08:00
• 30ea4c0f4b build(sgl-kernel): bump FlashMLA pin + fix cccl include for CUDA 13 (#29067) Xinyuan Tong 2026-06-26 04:00:15 +01:00
• 7f376644e0 [AMD] [GLM5] Mark EAGLE verified on MI300X/MI325X (gfx942) in GLM-5.1 cookbook (#29313) Raiden MakotoandRaiden-Makoto 2026-06-25 19:49:58 -07:00
• 413aeac0c9 [AMD][DI][CI] 1/N: MI355X disaggregation nightly benchmark (#29084) 4 people 2026-06-25 21:21:28 -05:00
• cfc0a0e0e0 Add Intel Quantization Support in SGLang (#18139) 4 people 2026-06-26 09:54:35 +08:00
• 10ff3c1dcb [NPU] [DOC] Add environment prerequisites to model tutorials (#29293) amote-i 2026-06-26 09:51:06 +08:00
• 6c92f9f328 fix(bench): pass DCP_RANK/DCP_WORLD_SIZE to set_mla_kv_buffer_kernel (#29356) Khoa PhamandClaude Opus 4.7 2026-06-25 18:13:57 -07:00
• 781537b61d [NPU] Nightly CI refactor and enhancement (#27433) hhhh1252023 2026-06-26 08:52:06 +08:00
• f83cbc2516 Add LFM2.5-230M to the LFM2.5 cookbook (#29321) Yi Zhong 2026-06-25 17:27:57 -07:00
• 4b0b0ea55f [sgl-kernel/cpu]: exclude amx gemm source from arm build (#29286) Yibo Cai 2026-06-26 08:23:43 +08:00
• ed71fb8f95 fix(anthropic): detect-and-passthrough mid-conversation system messages (#28906) Xinyuan Tong 2026-06-26 01:14:12 +01:00
• 4ce1c180bd [sgl-kernel/cpu]: fix arm64 w8a8 moe kernel signature (#29270) Yibo Cai 2026-06-26 08:11:57 +08:00
• 623300a589 [MLX] Fix FutureMap relay unit test to use RelayPayload (#29311) Lijuan Tang 2026-06-25 16:53:58 -07:00
• 38e857f423 [MUSA][diffusion] Bump torchada version to 0.1.68 (#29236) R0CKSTAR 2026-06-26 07:53:43 +08:00
• 1ba7c79761 [CPU] add indices in chunk_gated_delta_rule (#29267) Ma Mingfei 2026-06-26 07:51:07 +08:00
• c8c6757bed test(dp-attn): drop --enable-torch-compile from TestDPAttentionDP2TP2 (#29345) Khoa PhamandClaude Opus 4.7 2026-06-25 16:41:35 -07:00
• 65f31acc6c Add yueming-yuan to CI_PERMISSIONS.json (#29341) Yueming Yuan 2026-06-25 16:18:54 -07:00
• ea8f4e9f3f [feature] implement dcp for deepseek_v2 (#14194) Augusto Yao 2026-06-26 06:15:04 +08:00
• e4696ed62d [test] Report token tps in fwd occupancy kit and force ignore_eos (#29332) Liangsheng Yin 2026-06-25 15:08:33 -07:00
• b7d3c3016d [AMD] Feat/dsv4 aiter reduce scatter decode (#29103) kkandwunhuang 2026-06-26 04:45:18 +08:00
• 212c30d008 [MoE Refactor] Centralize FlashInfer CUTLASS MoE runner (#28211) Mohammad Miadh Angkad 2026-06-26 04:40:33 +08:00
• dbe9e3b706 [PD] Early-send cached-prefix KV overlapping uncached prefill forward (#29316) cctry 2026-06-25 13:31:53 -07:00