Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• becf252e6c AutoWeightLoader support Sglang native models 1: demo (#28671) JD 2026-07-21 14:32:13 -07:00
• 0eae9423d8 Change the FP8 per-tensor GEMM backend on SM120 to cuBLAS (#31961) Brayden ZhongandBrayden Zhong 2026-07-21 14:09:01 -07:00
• cb324cc958 Add wangfakang to CI_PERMISSIONS.json (#31980) Liangsheng Yin 2026-07-21 14:03:00 -07:00
• d03c8cee80 Negotiate PrefillDelayer only after KV-budget admission checks (#31835) Hanming Lu 2026-07-21 12:07:30 -07:00
• 927979e127 Fix pad-row top-k masking with custom_routing_function under DP attention (#31838) Hanming Lu 2026-07-21 11:49:17 -07:00
• 9a6c96083f [Cookbook] Add Laguna-S-2.1 (#31918) 4 people 2026-07-21 10:01:50 -07:00
• 57e5846b90 [LoRA] Guard TMA down path for LoRA hooks (#31608) Yanbin Jiang 2026-07-21 09:47:17 -07:00
• 604b1507d2 Add Inkling to nightly test (#31939) Ke Bao 2026-07-21 23:06:11 +08:00
• 6f55de0468 [EPD] Make encoder register/unregister health-check robust (#31576) Zheng Wengangandsiyu 2026-07-21 10:36:11 -04:00
• 8260ade61b [Bugfix] Fix CUDA import on non-CUDA platforms (#31919) Артем Савкин 2026-07-21 16:47:46 +03:00
• 075bd97952 [Benchmark] Remove obsolete auto-benchmark remnants (#31941) Xiaoyu Zhang 2026-07-21 20:44:52 +08:00
• 0a06cc5317 Fix extra_buffer_lazy guard bypass (#31942) Ke Bao 2026-07-21 20:41:35 +08:00
• e4eea7ce2f Optimize LongCat-Flash router GEMM with the HPC-Ops bf16xfp32 kernel (#30247) 3 people 2026-07-21 20:05:17 +08:00
• 303896a475 [Bugfix] Place empty Qwen encoder-DP embeddings on the communication device (#31663) qyb233 2026-07-21 19:54:36 +08:00
• 7c9257529f [Mamba] Support speculative decoding with extra_buffer_lazy (#30437) YAMY 2026-07-21 04:38:03 -07:00
• 1a53f231d6 [HiCache]: Optimize hybrid/DSA L3 prefetch result sync and usable-prefix clamping (#31443) 4 people 2026-07-21 18:59:51 +08:00
• df39a7b0b6 [CI][PD] Add NIXL disaggregation functional tests (#27894) 3 people 2026-07-21 13:17:06 +03:00
• 2cf2e7377f Fix dropped tool calls when a stream delta carries several (#31860) Ke Bao 2026-07-21 18:07:43 +08:00
• f369a820d4 [HiCache] Remove redundant parameters of build_xxx_stack and others (#31308) Chao ShiandZhangheng 2026-07-21 17:54:06 +08:00
• dcd9014f15 [AMD][MXFP4] Reland "Online MXFP4 quantization 2/N - FP8 to MXFP4 requantization on AMD GPUs" (#28291) 4 people 2026-07-21 11:50:23 +02:00
• 696e8f80d1 [CI] Graceful teardown in kl_mamba hicache tests to release pinned host pool (#31871) Alison Shao 2026-07-21 02:09:26 -07:00
• 66ea79dc2e Update FLA directory in CODEOWNERS (#31916) Yuan Luoandluoyuan.luo 2026-07-21 16:57:10 +08:00
• c0ed009f5b [NPU] Fix LLaDA2 MoE OOM after the FRACTAL_NZ cast, re-enabling the NZ speedup (#31772) wenxuewuhd 2026-07-21 16:09:34 +08:00
• 429f6b6d15 Turn on breakable prefill cuda graph for dp attention by default (#31682) Lianmin Zheng 2026-07-21 00:03:43 -07:00
• 37a830b667 [Feature] Add DWDP (Distributed Weight Data Parallelism) for MoE prefill (#29778) Yuhao Yang 2026-07-21 14:59:54 +08:00
• 01f558d905 Sm120 scatter fallback (#31669) JINO ROHITandgithub-actions[bot] 2026-07-21 12:28:27 +05:30
• c4c405a46b [AMD] batch 3: register newly-added JIT kernel benchmarks for jit-kernel-benchmark-test-amd (#31615) Michael 2026-07-20 23:51:51 -07:00
• d6ef68881e [NPU] Adapt MiMo-V2.5-W8A8 (#29131) iridiumine 2026-07-21 09:16:48 +08:00
• fa0ced195e [XPU] Enable breakable prefill CUDA graph on XPU (#30273) Rahul Vijayaraghavan 2026-07-21 06:39:40 +05:30
• bfefdc52d7 [CPU] Fix mxfp4 padding size (#31334) Zaili Wang 2026-07-21 09:03:24 +08:00
• d093c6a4bb [FullCG] Preserve attention LSE through the custom-op boundary (#31050) paulzhang-tm 2026-07-20 21:01:40 -04:00
• 9462c303a5 Fix stop boundaries for grammar-constrained speculative decoding (#31738) Yang Liu 2026-07-20 17:55:27 -07:00
• 4682ded472 vlm: parallelize multimodal preprocessing with customized worker num (#31438) Mick 2026-07-21 08:44:58 +08:00
• e7e8aaa73c Overlap grammar (constrained decoding) with speculative decode verify (#31488) Lianmin ZhengandJason Park 2026-07-20 17:36:05 -07:00
• 8905cbd42f Fix MiniMax-M3 crash on ROCm by making its override fields resolvable (#31837) zijiexia 2026-07-20 16:14:12 -07:00
• 0a2d3ca071 [cookbook] Inkling: add measured accuracy numbers to benchmark cards (#31823) zijiexiaandClaude Opus 4.8 2026-07-20 15:43:35 -07:00
• a82ead53bd Make Q contiguous before TRT-LLM MHA decode (#31667) Mohammad Miadh Angkad 2026-07-21 06:21:00 +08:00
• 7fe9ad25ac [Spec] Extract DFlash compact draft-cache rebuild helpers (#31677) Liangsheng Yin 2026-07-20 15:07:55 -07:00
• 91b210f7b0 [GLM5][MoE] perf: Write FlashInfer TRT-LLM MoE output directly (#28416) Kaixi 2026-07-20 22:45:23 +02:00
• e856eae921 use sgl_kernel_npu rmsrope accelerate llada2 (#27127) 3 people 2026-07-21 03:46:17 +08:00
• ff6c755952 [Refactor] Unify logprob results into a single LogprobResult and rename chunk env vars (#31733) Liangsheng Yin 2026-07-20 12:44:27 -07:00
• e149cdb337 docs(cookbook): revert MiniMax-M3 to dev image (model not yet in a release) (#31819) Douglas YangandClaude Opus 4.8 2026-07-20 12:16:47 -07:00
• 54aaedd76d Clean up prefill CUDA graph runner (#31654) Lianmin Zheng 2026-07-20 12:14:21 -07:00
• 4e8eb1457b [Diffusion] msgpack raw-bytes transport (drop base64/JSON) (#31565) RockduandClaude Opus 4.8 2026-07-20 11:32:02 -07:00
• 5ab3d90b81 Fix dropped Inkling reasoning at stream end (#31787) Ke Bao 2026-07-20 23:23:15 +08:00
• 7fc545b649 Align reasoning_effort schema across chat, tokenize, and responses (#31784) Ke Bao 2026-07-20 23:22:43 +08:00
• b6be150798 [AMD] Split ROCm 7.2 Stage-B large 1-GPU tests into three partitions (#31792) YC Yen-Ching Tsengandbingxche 2026-07-20 23:16:10 +08:00
• c41c573ce9 [GDN] Support ReplaySSM Ring Spec-Verify (#28695) 3 people 2026-07-20 22:06:30 +08:00
• d1c2a1de08 [NPU] memfabric-zbal update (#31777) monkeyLoveding 2026-07-20 22:01:28 +08:00
• 370f454e3d [NPU] fix modelslim quant tensor name (#31456) heziiop 2026-07-20 20:32:57 +08:00
• 3d82dacd58 Bump CuTe DSL to 4.6.0 (#31714) Mohammad Miadh Angkad 2026-07-20 17:11:59 +08:00
• 97e0647bdc [NPU] [DOC] Fix issues about npu docs found by aidd (#31302) amote-i 2026-07-20 16:31:27 +08:00
• fce5c75a30 [NPU] Fix vit graph tnd cu seqlens (#31701) Jinyan Yi 2026-07-20 16:20:33 +08:00
• 3f48245080 [CI] Temporarily disable GB300 jobs (runner availability) (#31764) Alison Shao 2026-07-20 01:19:24 -07:00
• fafa302e41 [XPU][NIGHTLY] Add 8 XPU nightly tests, enable 1-gpu suite (#30246) 3 people 2026-07-20 13:45:50 +05:30
• 50c118704a [diffusion] disagg: handle numpy arrays in cross-role transfer field extraction (#31325) kangwangamdandBingxu Chen 2026-07-20 15:50:17 +08:00
• 9668d9ea72 Support GPT-OSS zigzag CP with TRTLLM-MHA (#31732) Baizhou Zhang 2026-07-20 00:48:04 -07:00
• 2f14d6c6f2 ci: run nvidia nightly every 2 days at 14:00 UTC (7am PT) (#31749) Alison Shao 2026-07-19 23:58:39 -07:00
• 17fdd8487f [AMD] Update qwen3.5 cookbook (#31737) jacky.cheng 2026-07-20 14:14:14 +08:00
• 02236fa38c Add Inkling model support (#31681) 9 people 2026-07-19 22:57:37 -07:00
• 829e9ce9d5 Lower AutoRound quantization MMLU threshold (#31748) Xiaoyu Zhang 2026-07-20 13:40:48 +08:00
• 1843384c7a Fix LongCat-2.0 real EP (deepep): double all-reduce + ScMoE RoPE crash (#31311) 王鹤男andwhn09 2026-07-20 13:37:16 +08:00
• 49b9c46f41 [dLLM] Reuse block KV/req slots in place across FDFO rounds (#27877) Chenchen HongandXiaoyu Zhang 2026-07-20 13:36:13 +08:00
• b15a83983c [Fix] Release hierarchical cache host pool on graceful shutdown (#31746) Liangsheng Yin 2026-07-19 22:27:29 -07:00
• 2eed35d738 perf: avoid temporary VLM encoder gather padding (#31301) Mick 2026-07-20 12:54:29 +08:00
• 9f8e916131 [diffusion] post_training: run weight update under torch.inference_mode() (#31263) kangwangamdandYC Yen-Ching Tseng 2026-07-20 11:14:13 +08:00
• 1f637a65b9 [NPU] bugfix for W4A8MoE bias 3D dimension mismatch problem (#31707) McZyWu 2026-07-20 11:05:44 +08:00
• 35f2d4f761 Fix no-padding CUDA graph admission (#31273) Mohammad Miadh Angkad 2026-07-20 10:28:35 +08:00
• 8bf2ab9be9 Enable GPT-OSS TinyGEMM on CUDA 13 (#31649) Mohammad Miadh Angkad 2026-07-20 10:16:39 +08:00
• bab1dd0d12 [Intel XPU] Enable (biased) grouped topk for xpu (#31126) gaopengff 2026-07-20 09:35:08 +08:00
• 5325cee7ea Bug fix in compress to support XPU (#31541) Polisetty V R K Jyothendra VarmaandMa Mingfei 2026-07-20 07:04:11 +05:30
• 6a25dd7b5f fix: warm up Kimi VLM vision encoder at startup (#31298) Mick 2026-07-20 08:50:57 +08:00
• b3570a4531 [Refactor] Unify input logprob processing on a single chunked path (#31655) Liangsheng Yin 2026-07-19 16:03:23 -07:00
• 1a317839d7 ci: lower VL PP gsm8k threshold to 0.60 (0.65 has zero margin on H100) (#31702) Alison Shao 2026-07-19 15:41:12 -07:00
• 7d64858093 Add houseroad to CI_PERMISSIONS.json (#31717) Lianmin Zheng 2026-07-19 15:08:36 -07:00
• 555267ed05 Fix ROCm fused KV and KDA paths (#31688) Lianmin Zhengandwangwenchen0407 2026-07-19 15:02:35 -07:00
• 688a6d23f1 [DeepSeek-V4] Fix idle-rank dummy-extend sparse-prefill crash under DP breakable CUDA graph (#31705) Alison Shao 2026-07-19 14:42:04 -07:00
• d4801be447 fix: fix vlm cuda graph shape stability (#30868) Mick 2026-07-19 22:35:51 +08:00
• a03ca46a28 Fix KDA prefix caching under mamba extra_buffer and enable it for kimi_linear (#31474) Yuhao Yang 2026-07-19 20:09:03 +08:00
• 7a03d30149 Support MiMo V2.5 with zigzag context parallelism (#29972) Baizhou Zhang 2026-07-19 02:36:40 -07:00
• 377c93d54e [AMD] Gate TP4 o_proj/qkv CK block-FP8 GEMM shapes to Triton (ROCm 7.0 Qwen-3.5) (#30940) YC Yen-Ching Tseng 2026-07-19 16:45:44 +08:00
• c68392c535 [AMD] Fix DeepSeek MLA prefill shape mismatch on HIP eager fallback (missing mha_companion_layers) (#31675) YC Yen-Ching Tseng 2026-07-19 16:40:41 +08:00
• 942bf04ef9 [Scheduler] Add SGLANG_FORCE_COARSE_WAR_BARRIER opt-in for a whole-forward WAR barrier (#29353) Feng Yao 2026-07-19 00:28:49 -07:00
• 609fe1c0d1 fix(gemma4): prevent attention mask offset overflow (#31672) Kevin FlansburgandJimmy Shong 2026-07-18 23:24:03 -07:00
• b8ec544946 [DSA] Integrate Q8KV8 FP8 Sparse MLA Prefill into the DSA Backend (DeepSeek-V3.2) (#30514) 3 people 2026-07-18 20:58:16 -07:00
• cce5fe7696 [Scheduler] Move the WAR barrier to right after each run_batch launch (#31687) Liangsheng Yin 2026-07-18 20:38:34 -07:00
• 99f5a6f46b [flashinfer] Pass window_left at plan time for the SWA paged prefill wrapper (#31501) Lucia Fang 2026-07-18 13:33:24 -07:00
• 10908a6793 [Fix] Respect cache_protected_len in ChunkCache and disabled-radix release paths (#31662) Liangsheng Yin 2026-07-18 12:04:25 -07:00
• b3a0185cab model_runner: extract post-memory-pool wiring into _init_post_memory (#31601) Sam Shleifer 2026-07-18 12:57:54 -04:00
• d86ae51fcf metrics: allow extra labels on HTTP request/response Prometheus metrics (#31530) Sam Shleifer 2026-07-18 12:57:26 -04:00
• 5609f8e509 Reset only the used mamba state on radix cache hit (#31643) Ke Bao 2026-07-19 00:40:02 +08:00
• ece02ffc9c [NPU] FIX CMB illusion of garbled characters acc problems, in prefix cache mtp scenarios. (#31659) McZyWu 2026-07-18 21:49:37 +08:00
• 48ae829f6e Reset only the used mamba state on unified radix cache (#31648) Ke Bao 2026-07-18 21:43:13 +08:00
• d7b9425529 [NPU] fix: skip Triton embedding kernel on NPU to avoid kernel launch failure (#31636) chx96642264 2026-07-18 19:32:03 +08:00
• 216b750c8f [Kernel] Sweep decoupled scattered kernels into sglang.kernels.ops (RFC #29630) (#31582) Xiaoyu ZhangandClaude Opus 4.8 2026-07-18 19:07:07 +08:00
• 9306278fbc vlm: batch cross-request vit encoding and reuse attention metadata (#24013) 3 people 2026-07-18 18:50:33 +08:00
• 573c075fef CI: synchronize prefill graph test fixtures (#31665) Mick 2026-07-18 18:48:05 +08:00
• faf6894093 Implement SM120 DeepSeek V4 flashinfer_mxfp4 moe runner backend + TP2 (#30272) Gabriel Wu 2026-07-18 18:01:06 +08:00
• 6c6175fabd perf: avoid excessive prefill CUDA graph padding (#31487) Mick 2026-07-18 16:25:30 +08:00
• 38b29dcd6c Fix SM120 NVFP4 KV cache test OOM (#31653) Baizhou Zhang 2026-07-18 00:53:47 -07:00