Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• 071e649288 fix(rpc) Synchronize RPC requests only within the TP group. (#25213) twb1235andZhiqiang Xie 2026-07-18 15:44:51 +08:00
• 72c4ed1a3f [Spec] DFlash: remove per-step host syncs so the CPU runs a full step ahead (spec-v2 overlap) (#31468) ThanhhaoandHao Phan 2026-07-18 13:22:07 +07:00
• 7fbe91c6ea [AMD] register 8 JIT kernel benchmarks to jit-kernel-benchmark-test-amd (#31492) Michael 2026-07-17 23:00:00 -07:00
• 7a896215e7 [CP] Migrate MLA prefill CP (DeepSeek V3) to CP-v2 zigzag strategy (#31619) Khoa PhamandClaude Opus 4.8 2026-07-17 22:37:53 -07:00
• 639261f7b2 [Refactor] Move output logprob processing into the logprob_processor layer (#31624) Liangsheng Yin 2026-07-17 22:33:52 -07:00
• 19c53c44a0 [Fix] Account zero-logprob sequences correctly in chunked logprob stitching (#31639) Liangsheng Yin 2026-07-17 22:33:27 -07:00
• ab3d421c30 [plugin] oot torch profiler activity support (#31580) DevashishLal-CBandDevashish Lal 2026-07-17 22:20:54 -07:00
• a5c0b94034 Let CI server launches wait longer for ports held by a dying predecessor (#31281) Alison Shao 2026-07-17 21:45:24 -07:00
• 44e4999ab2 [Diffusion] Use SGLang server for ERNIE-Image prompt enhancement (#31354) 4 people 2026-07-18 06:55:28 +03:00
• 87dc211b87 [MUSA] Fix sglang-kernel build (#31634) R0CKSTAR 2026-07-18 11:51:09 +08:00
• 359009fa00 [Bugfix][NPU] Fix/Refactor routed scaling factor application in MoE routing (#31449) xdtbynd 2026-07-18 10:59:13 +08:00
• 67e7f8d13a [JIT] Refactor dtype traits into DTypeTrait and unify warp reductions (#30838) 5 people 2026-07-18 10:07:18 +08:00
• e48eabbeee ci: add MLX to coverage report backend display order (#31442) Alison Shao 2026-07-17 17:32:20 -07:00
• 238b2b2c9c Remove QServe and FBGEMM FP8 quantization (#31109) Brayden Zhong 2026-07-17 17:10:34 -07:00
• f926c30c57 Revert "Fix mamba track-boundary seqlen under overlap scheduler (#31369)" (#31622) Alison Shao 2026-07-17 17:03:10 -07:00
• 42a058c760 optimize: avoid fla l2-norm recompilation by token count (#31558) Mick 2026-07-18 07:58:36 +08:00
• 01a96720c6 [spec decoding] replace torch.multinomial with several native torch op in rejection sampling (#31620) Qiaolin Yu 2026-07-17 16:58:11 -07:00
• 304a529558 Revert "Bump FlashInfer to 0.6.15 and revert regressions" (#31625) Baizhou Zhang 2026-07-17 16:46:33 -07:00
• a01a8e1ed9 docs(cookbook): replace pinned nightly/dev images with :latest (#31610) Douglas YangandClaude Opus 4.8 2026-07-17 16:33:19 -07:00
• c95026aed3 Upgrade llguidance to 1.7.6 (#31484) Lianmin Zheng 2026-07-17 16:31:44 -07:00
• 632adff9fd refactor logprob processor layer (#20071) Qiaolin Yu 2026-07-17 16:28:10 -07:00
• 0ad0ff2e9e chore: bump sglang-kernel version to 0.4.5 (#31618) sglang-botandsglang-bot 2026-07-17 15:58:36 -07:00
• e2d2e8d07e [spec decoding] fix multi_layer_eagle rotate_input_ids kernel registration (#31614) Qiaolin Yu 2026-07-17 15:18:01 -07:00
• ec6a3163b7 [Feature] Add FP4 KV Cache Design and support SM120 GPUs (#21601) Sam (Kesen Li) 2026-07-18 05:49:43 +08:00
• 7fc3fb9657 Remove deprecated Mamba flags from doc, wrong FP8 GEMM docstrings and change Nemotron image to 0.5.15 (#31094) Brayden ZhongandBrayden Zhong 2026-07-17 14:34:29 -07:00
• c00206c68c chore: bump sgl-kernel version to 0.4.5 (#31496) sglang-bot 2026-07-17 14:00:55 -07:00
• ae3f62613a docs(cookbook): fix stale/pruned Docker image tags (#31508) Douglas YangandClaude Opus 4.8 2026-07-17 13:24:46 -07:00
• 2c856abbe3 [Fix] Enable chunked input-logprob processing by default to cap peak memory (#31498) Liangsheng Yin 2026-07-17 13:19:03 -07:00
• d389039337 [diffusion] Opt in Qwen and Wan multi-output conditioning expansion (#31233) Andy Ye 2026-07-17 12:00:23 -07:00
• fec6131844 [AMD] Disable DSA fused top-k v2 on ROCm for GLM-5.x / DeepSeek-V3.2 (#30506) fanxingranandkk 2026-07-18 01:22:14 +08:00
• 2db21daf8d Fix Heartbeat Checker in KV Manager Disaggregation (#31584) Muhammad Safiullah MemonandMuhammad Safiullah 2026-07-17 22:21:50 +05:00
• 2c64b7782e [AMD][PD] Fix early-send cached-prefix KV racing the prefill forward on mori (#31368) Zhaoyi LiandMichael 2026-07-17 12:08:04 -05:00
• 53229e88da [AMD] Fix stale imports in test_fused_fp8_kv_write.py (#31515) Bingxu Chen 2026-07-18 00:03:44 +08:00
• c546afc147 [AMD] Register 2 CPU/triton unit + kernel tests for AMD 1-GPU PR CI (#31379) Michael 2026-07-17 09:01:46 -07:00
• 5e7eed4c00 [MLX] Honor --max-running-requests in the model runner stub (#30547) NOOBandR0CKSTAR 2026-07-17 20:54:00 +05:30
• 85ac56c823 docs: simplify diffusion new model guide (#30109) Mick 2026-07-17 19:39:38 +08:00
• 681c223570 refactor: wrap split backends once on full-attention backends (#31439) Mick 2026-07-17 19:15:04 +08:00
• 24a8944e15 fix: enable Kimi multimodal breakable prefill cuda graph replay (#31391) Mick 2026-07-17 19:13:54 +08:00
• 132ade55cd [Kernel] Rewrite JIT custom all-reduce (v2) with a decoupled kernel/storage design (#31049) 3 people 2026-07-17 18:37:22 +08:00
• eaeb779ea4 [Doc] Update GLM5.2 Cookbook with LayerSplit usage (#31577) Baizhou Zhang 2026-07-17 02:25:50 -07:00
• 19f4859b30 [CI] Exclude current process memory from GPU idle check (#31571) Liangsheng Yin 2026-07-17 02:11:41 -07:00
• 8f765bc1c9 [Docs] Inkling cookbook: mark B300/GB300 recipes verified, tune B300 MTP mem fractions (#31550) zijiexiaandClaude Fable 5 2026-07-17 01:07:09 -07:00
• d67aa05697 Bump FlashInfer to 0.6.15 and revert regressions (#31502) Mohammad Miadh AngkadandBrayden Zhong 2026-07-17 15:50:12 +08:00
• e835512303 Add nightly test for GLM5.2 LayerSplit (#31512) Baizhou Zhang 2026-07-17 00:29:14 -07:00
• 619609aa5a [Kernel] Simplify sglang.kernels tests to idiomatic pytest style (RFC #29630) (#31546) Xiaoyu ZhangandClaude Opus 4.8 2026-07-17 15:06:26 +08:00
• 44e3dd2713 [HiCache] Optimize HiCache host pool free-list release (#30658) huangtingweiandZhangheng 2026-07-17 14:52:08 +08:00
• 6e3be088a9 [Spec] Allocate the verify tree-mask scratch on the target backend only (#31527) Liangsheng Yin 2026-07-16 23:23:21 -07:00
• bbd2a3fe4a ✨ [llm][npu][quant] Add W4A4 MXFP4 quantization support for Qwen3 Dense on Ascend NPU (#23795) Junlin Wu 2026-07-17 14:06:30 +08:00
• 1ac1ffea0c [Kernel] Fill non-CUDA coverage: HIP (aiter/rocm-triton) + Ascend NPU backends (RFC #29630) (#31307) Xiaoyu ZhangandClaude Opus 4.8 2026-07-17 14:05:35 +08:00
• 444bbd866d [CI] Fix invalid suite name breaking all PR test lanes (#31539) NOOB 2026-07-17 11:00:05 +05:30
• 96dd96c02b [DCP] Auto-disable tc_piecewise and breakable prefill CUDA graphs under DCP (#31532) Khoa PhamandCursor 2026-07-16 22:17:50 -07:00
• d310fce85f [Fix] Update PR #25015 revert for fused topk=1 draft postprocess (#31519) Khoa Phamandkpham-sgl 2026-07-16 21:51:16 -07:00
• ead703815e [DCP] Enable decode context parallel for Kimi K2.5 NVFP4 (#31514) Khoa Phamandkpham-sgl 2026-07-16 21:50:42 -07:00
• bf417440e9 [Scheduler] Add SGLANG_MAX_NEW_TOKENS_LIMIT to cap per-request max_new_tokens (#22591) AuFlowandAuFlow 2026-07-17 12:34:10 +08:00
• 45824c69ca fa3: build the topk>1 verify replay page table on-device (#31381) Liangsheng Yin 2026-07-16 21:21:01 -07:00
• 27ad9d11b1 [CI] Wait for GPU memory release before each test class setUpClass (#31509) Liangsheng Yin 2026-07-16 21:06:29 -07:00
• 12af7e6c34 [NPU] Fix DSA top-k seed buffer shape for MTP IndexShare (#31005) Peng Xingchen 2026-07-17 12:05:57 +08:00
• dfa6278370 [Fix] Publish idle scheduler metrics immediately when the running-reqs gauge is stale (#31517) Liangsheng Yin 2026-07-16 21:04:10 -07:00
• 486a56be56 [CPU] improve silu performance by replacing fp32 div with rcp14 (#31304) Ma Mingfei 2026-07-17 12:02:37 +08:00
• 27a52d2530 [Docs] Tune DeepSeek-V4 HiCache for MI355X PD (#31452) Duyi-Wang 2026-07-17 11:50:58 +08:00
• 40a3bd7659 [Docs] Mistral Medium 3.5 cookbook: replace stale day-0 dev images with latest (#31507) zijiexiaandClaude Fable 5 2026-07-16 20:19:01 -07:00
• 7355e0cb87 [NPU] custom-ops adapt (#30731) monkeyLoveding 2026-07-17 10:39:00 +08:00
• 9910ef8167 flashmla: sync-free spec via device-side draft-extend (#31090) Liangsheng Yin 2026-07-16 19:38:12 -07:00
• 8432eafd3d [Kernel] Decouple KernelBackend from device + device-based CapabilityRequirement (RFC #29630) (#31292) Xiaoyu ZhangandClaude Opus 4.8 2026-07-17 10:35:34 +08:00
• 68d324d697 [PD] Fix send_multipart blocking after prefill failure (#31306) amote-i 2026-07-17 10:33:29 +08:00
• 0675d3033f Fix mamba track-boundary seqlen under overlap scheduler (#31369) Sam Shleifer 2026-07-16 19:22:07 -07:00
• 37f94cb7a0 [Intel GPU] DeepSeek V4 13/N: use sgl-kernel implementation of kernels in V2 Compressor to run on XPU (#28439) Polisetty V R K Jyothendra VarmaandRahul Vijayaraghavan 2026-07-17 06:46:04 +05:30
• 302c3b97d2 fix: Qwen3.5-35B-A3B-AWQ w2_weight KeyError and related params for CPU (#31174) ZiruandNiu Ziru 2026-07-17 09:14:55 +08:00
• 27ea5ed070 [XPU] Fix dtype mismatch in MRotaryEmbedding.forward_xpu by calling _… (#31441) jiayisunx 2026-07-17 09:14:25 +08:00
• dc0b3eb68f [2/3] [EAGLE] perf: Fuse TP vocab-parallel embedding (#30948) Kaixi 2026-07-17 00:57:51 +02:00
• d539bf2cda [1/3] [EAGLE] perf: Fuse topk=1 draft postprocess (#30947) Kaixi 2026-07-17 00:57:27 +02:00
• 77d23a796e [1/N] elastic-ep: Add runtime EP scale-up (#30164) Yoray Zack 2026-07-17 01:53:44 +03:00
• d28e35b1a1 Fix num_running_reqs gauge on disagg prefill servers (#31495) cctryandZhiqiang Xie 2026-07-16 15:43:06 -07:00
• 640101da13 [Fix] Deflake score engine determinism tests (#31494) Liangsheng Yin 2026-07-16 15:42:47 -07:00
• 7cd55c6818 [HiCache] Optimize L2 mem allocation when cache miss in L3 (#19320) Bob ChenandZhiqiang Xie 2026-07-17 06:00:53 +08:00
• 9a4d640244 [Perf] Cache uniform ragged-verify layout for DSpark verify-all compact (#31434) Liangsheng Yin 2026-07-16 15:00:41 -07:00
• fc1e3797b7 [Spec] Split the capture width from num_tokens_per_req and gate replay on it (#31255) Liangsheng Yin 2026-07-16 14:59:11 -07:00
• 059ac7efe8 Don't fail server startup when psutil can't parse /proc/meminfo (GB300 ShadowCallStack) (#31098) Alison Shao 2026-07-16 14:57:34 -07:00
• bff489284b [Feature] Support DeepSeek-V4 Wint4Abf16 and Win4Afp8. (#25763) yuyu5333andzekai 2026-07-17 05:24:34 +08:00
• 68f4de162d [Docs] Remove Inkling H200 LoRA BF16 cookbook command (#31489) zijiexiaandClaude Fable 5 2026-07-16 12:57:37 -07:00
• 8c9833f9a9 cookbook(qwen3.5): bump AMD ROCm docker images to v0.5.15.post1 (#31454) ChangLiu0709 2026-07-16 20:41:24 +01:00
• 44f4ea917c fa3: sync-free eagle spec via fixed-window draft-extend metadata (#31364) Liangsheng Yin 2026-07-16 12:31:50 -07:00
• 1f34911de7 [NemotronH] Load shared embed_tokens/lm_head in MTP draft weights (#30456) Serge Panev 2026-07-16 10:20:38 -07:00
• 3bc7c54111 [NPU] Updated baselines for GLM-Image (#31467) Elizaveta Martirosianandronnie_zheng 2026-07-16 20:16:07 +03:00
• 4ad418d2c3 Push test case scripts from test repo to main upstream community repository (#31114) pllimax 2026-07-16 21:24:01 +08:00
• 7d0fd5101d optimization: shard kimi dp image feature transport and misc optimizations (#31227) Mick 2026-07-16 20:31:42 +08:00
• b296e1a503 [hicache]: add mamba concurrency io transfer kernel (#30535) jojoandhzh0425 2026-07-16 18:13:02 +08:00
• e2d021d4ab [AMD] Support two batch overlap with MTP on DeepSeekV4 (#30238) Wang, FangYuan 2026-07-16 17:34:04 +08:00
• e5f9804e26 Refining fused A GEMM dispatch (#31241) 3 people 2026-07-16 01:57:01 -07:00
• a798a2aeea [GDN/KDA] Fuse SM100 CuteDSL prefill state I/O into the chunk h kernel (#30169) Yuan Luoandluoyuan.luo 2026-07-16 16:40:33 +08:00
• db7e6807de [BugFix] Preserve tokenizer worker fanout when skip_tokenizer_init is enabled (#30682) 3 people 2026-07-16 16:29:46 +08:00
• 4d60c4540c [AMD] Skip AMD nightly local registry push (#31436) YC Yen-Ching Tsengandbingxche 2026-07-16 16:15:05 +08:00
• 3264477a07 [Docs] Add AMD-specific HiCache config for DeepSeek V4 playground (#31122) seungrokjandClaude Opus 4.6 2026-07-16 17:12:29 +09:00
• 1b9f228838 [Docs] Playground: migrate CP knob to canonical prefill-CP flags, align gating with runtime semantics (#31411) zijiexia 2026-07-16 01:07:34 -07:00
• 01b003255a [AMD]Reuse fused FP8 KV cache write on standard aiter prefill/decode (#26852) zhengyao 2026-07-16 16:05:04 +08:00
• b55228cfdb [Spec] Consolidate the verify step into eagle_worker_common.run_eagle_verify (#31380) Liangsheng Yin 2026-07-16 00:56:54 -07:00
• 5cbea10e2f Fix LongCat n-gram embedding in PD-disaggregated scheduler loops (#31134) 3 people 2026-07-16 15:42:51 +08:00
• a614821341 [Docs] Align B200 DeepSeek-V4-Pro balanced recipe with MegaMoE (#31373) Cameron Quilici 2026-07-16 02:24:02 -05:00
• 095a817612 [Bugfix][HiCache] measure load-back duration with CUDA events (#26411) 4 people 2026-07-16 15:11:04 +08:00
• f8eac995aa [diffusion] test: fix GLM-Image AR model-path to resolve local snapshot subfolder (#31313) kangwangamd 2026-07-16 14:57:34 +08:00