Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• 169fc1e20c [NPU] Acc fix for afmoe model introduced by topk refactor. (#31280) McZyWu 2026-07-27 15:40:49 +08:00
• c0f47a06fc [NPU] Determine the topk norm_type through scoring_func (#31393) McZyWu 2026-07-27 15:12:38 +08:00
• 3d3ba4f746 [BugFix][EPD] Fix Mooncake source-MR lifecycle for multi-TP /send (#32071) Zheng Wengang 2026-07-27 15:06:01 +08:00
• 082b2a10b6 Add local ZIP uploader for whl releases (#32489) Baizhou Zhang 2026-07-27 00:03:59 -07:00
• 5cc273a780 [feat] Opt-in flat response format for prompt top logprobs (#32078) Sam Shleifer 2026-07-27 02:44:34 -04:00
• 4ea17169b0 [UnifiedTree] fix: drop prefetched host refill under an un-backed-up parent (#31902) Yiqi Yangandhzh0425 2026-07-27 13:53:02 +08:00
• ee1736f39a [LoRA] Support LoRA under the breakable/full prefill CUDA graph (#30988) Ethan (Yusheng) Su 2026-07-26 22:10:03 -07:00
• c6a6200a1a [AMD] Fix pip setup in AMD Miles nightly builds (#32469) Xinyu Jiang 2026-07-26 21:32:40 -07:00
• 2abb1d2c37 fix(hisparse): correct DSA KV memory budget (#31992) DAI0818andybyang 2026-07-27 11:19:32 +08:00
• abb8f4b5e3 model: support EmbeddingGemma (#32375) Mick 2026-07-27 10:40:47 +08:00
• a358374ae9 [NPU][Fix Issue]: Send expert weights contiguous tensor across cards during EPLB rebalance (#32001) icarus_zh 2026-07-27 09:20:28 +08:00
• 3863612023 Add oulgen to CI_PERMISSIONS.json (#32453) Lianmin Zheng 2026-07-26 15:04:11 -07:00
• e14068d161 [NPU]Add Ascend transfer version compatibility. (#31189) shadowxz109 2026-07-26 21:07:59 +08:00
• e8a635a412 Load initial expert location metadata on CPU (#32435) icarus_zh 2026-07-26 20:44:25 +08:00
• a76b74cbe0 add fill_draft_extend_prepare_buffers_native for NPU (#32427) ming_wang 2026-07-26 20:42:14 +08:00
• 78d7928296 [UT][NPU] add NPU attention unit tests for ascend_backend and ascend_dsv4_backend (#32294) xdtbynd 2026-07-26 19:59:44 +08:00
• 61057bda6c [BugFix] Prevent TBO crash when return_logprob is enabled (#32180) Wang, FangYuan 2026-07-26 15:07:54 +08:00
• 833e1bc601 [Fix][AMD] Qwen3.5 MoE: disable global-slot shared-expert fusion under per-rank EP backends (MoRI + dp-attention init crash) (#31793) jacky.cheng 2026-07-26 14:59:47 +08:00
• 72e415dfc8 [Bugfix] Fix prefill suspension caused by delayed negotiate_should_allow_prefill invocation (#32389) silencejade 2026-07-26 14:59:26 +08:00
• 1d0cd2e473 [AMD] Nightly Test Coverage - Minimax-M3-MXFP8 Accuracy Test (#30613) YC Yen-Ching Tseng 2026-07-26 14:52:22 +08:00
• 2cbddb842d [DSV4/SM120] Allow fused MHC opt-in with standalone TileLang pre disabled (#30954) ormandjandMohammad Miadh Angkad 2026-07-25 20:02:54 -05:00
• 55c4853487 [comm] Enable multi-node custom-AR v2 on a single NVLink clique (#32339) 3 people 2026-07-25 17:25:18 -07:00
• 2c63a2f12b Fix --hicache-size allocating ~2x host memory on hybrid SWA (#32373) 3 people 2026-07-25 17:19:44 -07:00
• 9989077f24 Use native batched llguidance mask generation (#32412) Lianmin ZhengandAlec S 2026-07-25 16:36:32 -07:00
• fae84ac0f9 Fix token count localization for replicated attention-TP forwards (#32411) Lianmin ZhengandXingyu Liu 2026-07-25 16:36:15 -07:00
• 5f330004bd Fix flaky test_sampling_mask: mask length can legitimately be top_k + 1 (#32410) Lianmin Zheng 2026-07-25 16:05:57 -07:00
• 3da1071d56 [Spec] Hold the grammar bitmask in one GrammarMask type across all decode paths (#32409) Liangsheng Yin 2026-07-25 15:27:43 -07:00
• d3cf4dfbaa Update audio container test time estimate (#32408) Lianmin Zheng 2026-07-25 14:45:08 -07:00
• cd145f840f Radix Cache Split: Spin off TreeCore (#29901) Jialin Ouyang 2026-07-25 14:31:59 -07:00
• a23f6ea090 [Diffusion] offload rollout weights to pinned host memory (#32032) Kangrui DuandYihao Wang 2026-07-25 13:35:53 -07:00
• 91f386a5b2 fix(disagg): support pipeline-parallel hybrid-linear transfer (#32270) YAMY 2026-07-25 13:34:38 -07:00
• 659d349b61 [core/loader] Add presharded load format (#24256) Cheng WanandShu Wang 2026-07-25 13:03:39 -07:00
• 9791fc7090 Add configurable FlashInfer autotune skips (#31389) Mohammad Miadh Angkad 2026-07-26 02:17:58 +08:00
• 953c587adf [Docs] Add Qwen3.6 35B NVFP4 to cookbook (#31413) Mohammad Miadh Angkad 2026-07-26 02:16:53 +08:00
• 69a3c54c70 Fix SWA admission livelock on cached-prefix resumes (#32379) Ke Bao 2026-07-25 22:37:35 +08:00
• e943e609dc [DSPARK] Grammar-constrained decoding, incl. tool_choice=auto (#31753) 3 people 2026-07-25 13:20:25 +01:00
• a678a42033 [KDA] Add target_verify support for speculative decoding (#26888) 4 people 2026-07-25 19:52:44 +08:00
• 7c4b22fae5 [Hicache][1/2]Support Mamba branching in Unified Radix Cache with HiCache (#31181) 3 people 2026-07-25 19:44:40 +08:00
• 1054060ef1 perf: speed up marlin moe with occupancy-aware launch specialization (#31552) Mick 2026-07-25 19:38:11 +08:00
• d021990bf5 [DFLASH] Support grammar-constrained decoding in speculative verify (#30096) Hồ Sỹ Thếandhnyls2002 2026-07-25 18:36:46 +07:00
• 17afd8421f [Spec] Share the grammar mask build and verify-tree staging across spec workers (#32393) Liangsheng Yin 2026-07-25 03:40:21 -07:00
• 3c5bf1f6d2 [Spec] Derive NGRAM grammar tree links on the host instead of reading back retrive_next_token (#32380) Liangsheng Yin 2026-07-25 02:25:16 -07:00
• f5155d9602 [EPD] Fix HTTP dispatch lock blocking cross-request encoder batching (#31275) Yuang Chenand晟海 2026-07-25 14:47:58 +08:00
• 9eb2dccbb7 [Elastic EP] Fix recovery lifecycle and add manual coverage (#31744) Xun SunandShangming Cai 2026-07-25 14:32:47 +08:00
• f9c14e6bd4 [FEAT] Support fast engine recovery through weight cache (#27139) 3 people 2026-07-25 14:31:21 +08:00
• bdd0698541 chore: bump mooncake version to 0.3.12.post1 (#32302) Shangming Cai 2026-07-25 14:12:46 +08:00
• 6a046fad09 [PD] Prevent decode scheduler from blocking on ZMQ sends to a stalled prefill peer (#31144) SovietPowerandShangming Cai 2026-07-25 12:53:41 +08:00
• ebcb74abd4 feat(hicache): Add shared memory allocator for host KV cache (#29326) 3 people 2026-07-24 21:05:37 -07:00
• b83041c3cc Migrate CompressedTensorsW4A4Nvfp4MoE TRT-LLM path onto MoeRunner (#32248) Alex NailsandClaude Opus 4.8 2026-07-24 20:57:38 -07:00
• cff20a2fbb [Spec] Consolidate the grammar sync decision into ScheduleBatch.grammar_needs_sync (#32353) Liangsheng Yin 2026-07-24 20:42:38 -07:00
• 95865de24f [diffusion] CI: read consistency GT from ci-data-diffusion at per-platform commit (#32297) Yihao Wang 2026-07-24 18:55:57 -07:00
• ebf5330f71 [diffusion] CI: enhance diffusion GT publishing with per-platform directory resolution (#32282) Yihao Wang 2026-07-24 18:54:46 -07:00
• a690e5e0b3 Add stream label to TTFT metrics (#32363) cctryandJialin Ouyang 2026-07-24 17:44:04 -07:00
• ce705bb6dc Report accelerator type in /v1/loads (#32348) cctryandYinghai Lu 2026-07-24 17:31:28 -07:00
• 1ef2b85ef7 chore: bump docs install version to 0.5.16 (#32347) sglang-botandsglang-bot 2026-07-24 17:19:39 -07:00
• 9402012f0f [Perf] Halve the non-finite sanitization overhead in per_token_group_quant (#32296) DarkSharpnessandClaude Fable 5 2026-07-25 08:17:12 +08:00
• 7de8792758 Fix FP8 Triton dtype selection on A100 (#31340) danielafrimiandDaniel Afrimi 2026-07-25 03:03:19 +03:00
• 3079157175 Fix PyPI release: drop the git-only sgl-eval dep from packaged metadata (#32354) 3 people 2026-07-24 20:00:57 -04:00
• 9c483cccfe Support a same-size mixed q dtype in the fused RoPE kernels (#31834) Pranjal Shankhdharandpranjalssh 2026-07-24 16:19:08 -07:00
• 962c076934 Decode input_audio media containers with PyAV & Update memory profiler (#31832) 3 people 2026-07-24 16:18:45 -07:00
• 14d6e1d3b1 [RL] Support FlashInfer TRT-LLM NVFP4 MoE in the RL weight checker (#31085) Zhihao Wang 2026-07-24 16:01:54 -07:00
• 0a212c6119 [RL] DSV4: add env to quantize SWA KV cache from bf16-rounded values (#31086) Zhihao Wang 2026-07-24 15:52:00 -07:00
• 4ececf2b1d [chore] Add verbose flag to twine upload command (#32349) Qiaolin Yu 2026-07-24 14:54:07 -07:00
• f7986c8603 [RL] DSV4: dispatch indexer topk_transform_512 through DSATopKBackend (#31087) Zhihao Wang 2026-07-24 14:45:44 -07:00
• 8727d105db Add prefill and decode load counters to LoadSnapshot (#32245) cctry 2026-07-24 14:04:54 -07:00
• f15b43242b Bump sgl-deep-gemm to 0.1.5 (#32345) Baizhou Zhang 2026-07-24 14:03:38 -07:00
• be7cc17307 sglang rust server environ fsm error id gen (#32240) Rain Jiang 2026-07-24 13:55:10 -07:00
• 82fe0f041a Fix stale flashinfer-MLA fallback poisoning spec verify capture (trtllm_mla + tc_piecewise) (#32288) Qiaolin Yu 2026-07-24 13:24:26 -07:00
• be7c13af07 [BugFix] Fix DS/Kimi crash on non-first PP ranks when resolving input length (#31752) Zheng Wengang 2026-07-25 03:19:02 +08:00
• 71015f3fea fix(dsa): fail fast on fp8_e4m3 KV with tilelang DSA backend on CUDA (#31346) mosya415andmosya415 2026-07-24 22:10:38 +03:00
• 1e69765bae Add FP4 Indexer for DeepSeek V4 on SM120 (#27059) Jinyan Chen 2026-07-25 02:37:23 +08:00
• 2428f56145 [Bugfix] Fix Kimi-Linear state transfer across heterogeneous TP (#32262) YAMY 2026-07-24 10:31:17 -07:00
• 5da0b6ec39 Write-back policy fix for unified tree (#31845) Zhiqiang Xie 2026-07-24 09:58:18 -07:00
• 448662e85e [mm] Accept per-item embedding lists from DataEmbeddingFunc (#31826) Lu Fang 2026-07-24 08:27:24 -07:00
• fa243fee40 [CI] Skip flaky test in CI for disaggregation group (#32324) Shangming Cai 2026-07-24 21:59:20 +08:00
• dfaf75b1a6 [NPU] bugfix for extra device memory on Ascend (#30112) khalilzhk 2026-07-24 21:13:21 +08:00
• 3d91a569ce [MoE Backend] Add HPC-Ops FP8 MoE runner backend (#30541) 3 people 2026-07-24 19:46:11 +08:00
• 4d5917e744 Add DeepSeek-reference 1e-20 epsilon to top-k renormalization to prevent 0/0 NaN (#31017) Jun LiuandXiaoyu Zhang 2026-07-24 20:45:44 +09:00
• 8389d79e43 ci: add LongCat-Flash-Lite-FP8 8-GPU nightly test + fix NextN rope_theta (#32125) 3 people 2026-07-24 19:45:28 +08:00
• 841fa293b5 [Fix] Reject online weight updates while the HPC-Ops router GEMM split cache is active (#31943) 3 people 2026-07-24 19:30:13 +08:00
• a31542ebd9 [Feature] Add leveled invariant-check primitive for nan/inf/oob validity checks (#32308) Liangsheng Yin 2026-07-24 04:27:20 -07:00
• de816e1eb5 [Disagg][StagingBuffer][1/2] Robustness and failure handling (#31217) YAMY 2026-07-24 02:22:55 -07:00
• 34d02bae47 Add CI permissions for Elastic EP contributor UNIDY2002 (#32306) Shangming Cai 2026-07-24 17:20:15 +08:00
• f4f15162bc [Fix] Fail fast when a safetensors index references missing shard files (#32279) Liangsheng Yin 2026-07-24 02:18:56 -07:00
• b954e9cf3d [6/6][kimi-deterministic] Use deterministic seeded coins for EAGLE rejection sampling (#30822) Yuzhen Zhou 2026-07-24 02:11:21 -07:00
• 3849beb7e3 [CI] Fix XPU platform test on machines without the XPU sgl-kernel op (#32298) Ethan (Yusheng) Su 2026-07-24 01:42:36 -07:00
• 364b5f23e6 [BugFix][EPD] Harden zmq_to_scheduler receiver failures; sync error info across TP (#31592) Zheng Wengangandsiyu 2026-07-24 04:37:52 -04:00
• 58f417049d [PD] Fix multi-tokenizer disaggregation metrics labels (#30412) Jun LiuandXinyuan Tong 2026-07-24 17:27:21 +09:00
• b8bb1b4e5a [lora] Fix WAR race: never write MoE runner output into hidden_states in place (#31870) Sam Shleifer 2026-07-24 01:58:55 -06:00
• d059b0f56e [Fix] Clamp degenerate all-sentinel draft rows to token 0 in dspark _online_combine_kernel (#32277) Liangsheng Yin 2026-07-24 00:53:56 -07:00
• 35e25f5356 [Feature] DCP: A2A + FlashInfer-MNNVL comm backends and q-replicate (Helix) (#21637) 6 people 2026-07-24 14:21:53 +07:00
• 39955d5314 [MoE] Make DeepEP auto serve flashinfer_cutedsl FP4 (coerce to low_latency) + guard (#29523) Xinyuan Tong 2026-07-24 14:59:18 +08:00
• 15d73f1e03 Fix dynamo recompile limit in allreduce and bf16 gemm (#32239) Qiaolin Yu 2026-07-23 22:24:01 -07:00
• 319055c191 [Intel GPU] Add XPU Platform support (#31949) Polisetty V R K Jyothendra VarmaandMa Mingfei 2026-07-24 10:19:29 +05:30
• f0f78a6c93 Add deterministic inference for eagle parity test (#30026) ANSHUMAN TRIPATHY 2026-07-24 10:18:21 +05:30
• 99b29bf188 [Fix] Support ENCODER_ONLY target-verify in the trtllm_mha backend (#32178) Liangsheng Yin 2026-07-23 20:33:39 -07:00
• eac7c7d7cd fix(attention): read per-runner kv cache dtype off model_runner (#32251) Cheng Wan 2026-07-23 20:08:57 -07:00
• bd3f6a7935 [CI] Remove redundant Rust cache save-if settings (#32243) Lianmin Zheng 2026-07-23 18:48:54 -07:00
• 1e10ec93b3 [XPU] Add XPU device support for LMCache radix cache integration (#23534) 4 people 2026-07-23 17:41:01 -07:00
• 2f823a2eee [Intel GPU] calculate free memory based on allocated memory for XPU (#32044) Polisetty V R K Jyothendra VarmaandMa Mingfei 2026-07-24 06:08:52 +05:30