Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• 21ee743f47 [CI][Bugfix] Limit MRV2 sampler JIT warmup registration to ROCm (#58465) Kevin H. LuuandClaude Opus 5.5 2026-09-23 16:11:50 -07:00
• 5fcc6e7c73 [ROCm][Bugfix] Fix TileLang mHC fused RMSNorm on 64-wide wavefronts (#58419) djramicandAndreas Karatzas 2026-09-24 01:03:45 +02:00
• 8b660ce96b [ROCm][Test] Cover MoRI graph replay and output lifetime (#58093) Andreas Karatzas 2026-09-23 16:16:51 -05:00
• 848dbf89ec Merge branch 'main' into qwen4exp-qsa-fused-draft-metadata vr8vr8 2026-09-23 23:37:13 +03:00
• 97b1b12117 [Docs] Fix docstring typos (output_dytpe, kwrags, Abbrivations) (#55936) 陈志谦andmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-09-24 04:36:57 +08:00
• e26547adfb [Bugfix] Pick a KV block size supported by every attention backend (#49845) DivyandClaude Opus 5 2026-09-23 13:36:35 -07:00
• 34b069080d [CI] Select one GPU for the H200 initialized snapshot E2E step (#58351) Thang NguyenandClaude Opus 5.5 2026-09-23 16:08:51 -04:00
• 711fc55c10 [Perf] Use breakable CUDA graphs (no torch.compile) by default under VLLM_BATCH_INVARIANT so the tuned matmul configs see the runtime M (#57586) Lio Einaudi 2026-09-24 02:39:38 +08:00
• 31d78a21b9 [Spec Decode][Qwen3.8] Enable fused draft metadata updates for the QSA side cache bilikasandClaude Opus 5.5 2026-09-23 20:58:12 +03:00
• f9ad9dd6b4 Doc: add DiffusionGemma to supported models (#46466) CBPandMisha Goin 2026-09-24 01:14:30 +08:00
• b5ea0c7107 [Bugfix][ROCm] Fix startup OOM in AITER MLA FP8 prefill workspace sizing (#57923) Simon Danielsson 2026-09-23 18:44:12 +02:00
• c4f6ce4874 [Bugfix][NIXL] Restore successful push completion reporting (#58188) Dao007foreverandCodex 2026-09-23 09:39:04 -07:00
• c4cfd7007b [CI][ROCm] Add an MI355 Kimi-K3 unit test group (#58012) 3 people 2026-09-23 10:25:51 -06:00
• 8033bd08c1 [ROCm][CI] Validate Mooncake and NIXL prefill/decode accuracy (#58095) Andreas KaratzasandOpenAI Codex 2026-09-23 11:16:31 -05:00
• b715699df4 [ROCm][Perf] Extend QK-norm/RoPE/KV-cache fusion to MRoPE (#50212) 3 people 2026-09-23 19:11:57 +03:00
• 0549e8d0ab [Quark] Remove quark-specific silent online quantization (#51800) fxmarty-amdandmgoin 2026-09-23 17:35:30 +02:00
• f69cc7c178 [Quantization][Bugfix] Bump humming-kernels to 0.1.16 (#58054) 3 people 2026-09-23 23:22:45 +08:00
• 9f07d023d0 [Quantization] Enable humming wNaM asymmetric quant (zero_point) with compressed-tensors (#46528) HDCharles 2026-09-23 10:23:11 -04:00
• 88afb9dcac [CPU] Adds support for fp32 attention sinks (#56252) Ankit-Jaiswal-AMDandDouglas Lehr 2026-09-23 19:25:09 +05:30
• 9ed81581a8 [Perf][MoE] Skip top-k slots routed to non-local experts in TritonExp… (#58051) Shuolei Wang 2026-09-23 21:03:21 +08:00
• d95a896ea6 [Feature][Frontend] Add DeepSeek-V4 FIM completion rendering (#44229) 3 people 2026-09-23 21:02:00 +08:00
• 157bcb7c48 [DSpark] Support pipeline-parallel targets in aggregated serving (#56956) 3 people 2026-09-23 19:52:28 +08:00
• 15ed1262e7 [Bugfix][Tool Parser] Migrate Granite to the streaming Parser Engine (#49648) 3 people 2026-09-23 03:37:40 -07:00
• c961121519 [Bugfix] Disable prefix caching for encoder-only before model config hooks (#58287) Tianyu Guo 2026-09-23 17:25:32 +08:00
• a22c637ed9 [ROCm][Test] Check GDN prefill numerics and output ownership (#58091) Andreas Karatzas 2026-09-23 04:19:58 -05:00
• e34489a543 [XPU] upgrade to PyTorch 2.14 (#56013) Yan MaandKunshang Ji 2026-09-23 17:18:52 +08:00
• c843f0cab7 [Bugfix][SM120][MLA] Support NoPE sparse MLA (GLM-5.3-Flash) on the FlashInfer SM120 backend (#55277) Gabriel WuandKimi Code 2026-09-23 17:03:20 +08:00
• afec265bca [MRV2] Miscellaneous code cleanup (#57980) Nick Hill 2026-09-23 01:51:52 -07:00
• 36bc694889 [ROCm][CI] Include Python tooling in ROCm CI artifacts (#58271) Andreas KaratzasandOpenAI Codex 2026-09-23 03:44:54 -05:00
• 56b3acb55b [ROCm][Bugfix] Keep zero MiniMax MXFP8 activation blocks finite (#58089) Andreas Karatzas 2026-09-23 03:42:33 -05:00
• 1904ed9402 [Perf][ROCm][Attention] Narrow the Triton prefill-attention KV tile on RDNA3/RDNA4 (#58225) 3 people 2026-09-23 16:39:27 +08:00
• 955bd6abef [CI][Bugfix] Extend groupwise rms_norm scale tolerance to CUDA (#58252) vllm-agentandkimi-no-na-wa 2026-09-23 01:29:11 -07:00
• 6632ed559e [Perf][Attention] Avoid CPU-GPU sync in DCP sequence lengths (#58169) Jiangyun ZhuandOpenAI Codex 2026-09-23 16:28:43 +08:00
• 96e203b08e [ROCm][Bugfix] Register MRV2 sampler JIT warmups (#58092) Andreas KaratzasandOpenAI Codex 2026-09-23 03:17:13 -05:00
• 26879f3260 [ROCm][Compile] Fuse AITER static FP8 attention output (#58099) Andreas KaratzasandOpenAI Codex 2026-09-23 03:16:52 -05:00
• 973a3be780 [Refactor][Quantization] Make FP8 and MLA weight transforms reusable pure functions (#57732) aoshen02andYongye Zhu 2026-09-23 16:16:20 +08:00
• 9488318c6a [Mooncake] Address review nits from #56855 (#57174) Nicolò LucchesiandYifan Qiao 2026-09-23 10:15:45 +02:00
• 153242a314 [ROCm][CI] Add MI355 dense NVFP4 and MoRI kernel mirrors (#58281) Andreas KaratzasandOpenAI Codex 2026-09-23 03:06:15 -05:00
• a9f07d0bc5 [ROCm][CI] Mirror the three TurboQuant evaluation groups on MI355 (#58282) Andreas KaratzasandOpenAI Codex 2026-09-23 02:37:23 -05:00
• 9127170b88 [Quantization] Select per-token NVFP4 MoE backends explicitly (#57176) 5 people 2026-09-23 00:30:58 -07:00
• 0f2a15c927 [Tests] Select V2 for diffusion scheduler unit tests (#58272) Andreas KaratzasandOpenAI Codex 2026-09-23 02:04:22 -05:00
• eb2e91d3ba [ROCm][Perf] Use wvSplitK for single-output GEMMs (#53283) 00CC 2026-09-23 14:38:17 +08:00
• b33c8a657d [Bugfix][Frontend] Keep length finish_reason for max_tokens-truncated streaming tool calls (#46303) Ting SUN 2026-09-23 14:31:10 +08:00
• de6674b002 [MoE] Use GateLinear for all MoE models (#58234) Jee Jee Li 2026-09-23 13:40:13 +08:00
• e975732152 [XPU][CI] Deselect tests/v1/spec_decode/test_mtp.py::test_glm_mtp_defers_lm_head (#58237) xiangdong 2026-09-23 13:20:51 +08:00
• 88aa0d287d [Spec decode] Support variable-length decode for Kimi-K3 adaptive ver (#52988) 3 people 2026-09-22 20:43:32 -07:00
• 355d789f0d [Compilation] Fix QuTLASS compilation with PyTorch 2.13 (#58173) Wentao Ye 2026-09-22 23:43:28 -04:00
• 73c59d365b [XPU] Wire up SYCL apply_rotary_emb kernel in ApplyRotaryEmb (#55721) Michał GanczarenkoandClaude 2026-09-23 05:43:24 +02:00
• 94f4170df3 [Bugfix] Set worker runtime threads before profiling and compilation (#55891) 3 people 2026-09-22 21:12:20 -05:00
• e581e14002 [ROCm][CI] Stage G gating (#50922) Andreas KaratzasandOpenAI Codex 2026-09-22 21:04:54 -05:00
• 8d16ca6cc2 [CI][Bugfix] Update IPC test caller for #57312's _apply_entries signature (#58107) 3 people 2026-09-22 18:50:06 -07:00
• 0eb42dbbfc [ROCm][Compile] Support BF16 AsyncTP fusion (#58098) Andreas Karatzas 2026-09-22 20:36:38 -05:00
• f9dce295c9 [ROCm][DSv4][Perf] Fuse the inverse RoPE into the sparse decode reduce (#57451) Fangzhou Ai 2026-09-22 21:23:55 -04:00
• 3ee2906f47 [EPD] Support metadata-only audio inputs (#57887) Tianyu GuoandCyrus Leung 2026-09-23 09:03:50 +08:00
• 826e300cde [Bugfix] Skip VllmConfig re-validation for with_hf_config submodel views (#58212) 3 people 2026-09-22 17:50:17 -07:00
• d110c2f19c [ROCm] Use silu_and_mul_with_clamp's torch._C op (#52052) 5 people 2026-09-23 01:30:14 +02:00
• 1c0eee919d [ROCm][Model][Bugfix] Enable GLM-5.2-MXFP4 on the deepseek_v32 path and fix sparse attention correctness (#51915) 4 people 2026-09-22 19:08:51 -04:00
• 9646f53064 [Bugfix][ROCm] Use the platform FP8 range in the concat MLA q test (#58153) stefankoncarevicandAndreas Karatzas 2026-09-23 01:05:33 +02:00
• 111d7d7be2 [ROCm][Build][The Rock] Bump Triton version to 3.8.x tip-of-tree with source build in The Rock image (#58006) rasmith 2026-09-22 18:05:13 -05:00
• d421a2e8fc [Perf] Remove CPU-GPU sync in heterogeneous vocabulary speculative decoding (#57396) Michael Lapshin 2026-09-22 18:34:19 -04:00
• 4edb55169f [CI] Shard (H200 MIG 18GB) Spec Decode Draft Model across whole-directory replicas (#58193) Thang NguyenandClaude Opus 5.5 2026-09-22 17:50:18 -04:00
• 4ec0b95c44 [CI] Split LM Eval TurboQuant KV Cache into per-config jobs (#57113) Thang NguyenandKimi 2026-09-22 17:49:06 -04:00
• 5c3b61e9c4 [Core] Disable JIT warmup in eager mode (#58197) Misha Goin 2026-09-22 17:35:23 -04:00
• 01e15bc94c [Bugfix] Backport Inductor custom-op pattern matching fix (#58189) Misha Goin 2026-09-22 17:26:49 -04:00
• ecdd11d0e0 [CI] Build the torch-nightly image on Ubuntu 24.04 (#58204) Andrey Talman 2026-09-22 17:15:57 -04:00
• 25965ee583 [Kernel] Remove AllSpark INT8 W8A16 GEMM backend (#58001) Misha Goin 2026-09-22 17:07:34 -04:00
• 1b3b88ec2b [Core] structured generation mode for DiffusionGemma model (Jev-like) (#57250) 5 people 2026-09-22 14:51:17 -06:00
• fae5cbfc8b [Refactor] Remove dead code multiple places (#58002) Wentao Ye 2026-09-22 16:36:03 -04:00
• 276db94fb2 [Bugfix][Model][Spec Decode] Defer disposable GLM MTP head (#55442) lucamotz 2026-09-22 22:08:27 +02:00
• 71a2392845 [Bugfix][V1] Read ModelState max_model_len from model config (#58149) hclandmgoin 2026-09-23 03:45:04 +08:00
• d1f63f2af9 [ROCm][CI] Add GELU activation for AiterExperts in the modular-kernel coverage (#58030) Divakar VermaandAndreas Karatzas 2026-09-22 15:39:13 -04:00
• 415e11c6f1 [Bugfix][V1] Honor enable_jit_warmup for V2 kernel warmup (#55146) pengyihangandmgoin 2026-09-23 03:34:31 +08:00
• bc162b3f92 [MRV2] Release weight offloader on shutdown (#57834) Taneem Ibrahim 2026-09-22 15:01:20 -04:00
• f944491db9 [Bugfix] batch_invariant: keep non-AllReduce collectives enabled on NCCL >= 2.31 (#58179) Guanxin LiandFlora Feng 2026-09-22 11:44:00 -07:00
• c024cfdb02 [Fast Start] Support data parallelism in the weight cache daemon (#57386) siyuandIsotr0py 2026-09-23 02:19:40 +08:00
• ca831d1c55 [ROCm][DSv4.1][Perf] Fuse the inverse RoPE into the sparse decode reduce (#57435) 3 people 2026-09-22 14:05:11 -04:00
• 72675c4706 [Bugfix][ROCm] Dispatch the QuantFP8 CUDA fallback on the class (#58136) stefankoncarevicandAndreas Karatzas 2026-09-22 19:49:11 +02:00
• 74370a0e30 [Bugfix] hadacore_transform: respect inplace parameter to fix garbage outputs with QuIP transforms (#43462) Gilles TurpinandWentao Ye 2026-09-22 18:48:20 +02:00
• a6131b0e65 [Bugfix][Engram] Fall back when /dev/shm is absent before sharing tables (#57914) 3 people 2026-09-22 09:19:09 -07:00
• 066ee91197 [Bugfix][Attention] Avoid NaN in the Triton softcap for large attention logits (#56579) 3 people 2026-09-23 01:11:05 +09:00
• 045467c102 Merge remote-tracking branch 'origin/main' into sync/lesj/qwen4exp-bugfixes lesj0610 2026-09-23 00:40:48 +09:00
• c42f5285ec [Bugfix] prioritize architecture capability before DeepGEMM availability check (#58073) Tony Lin 2026-09-22 23:38:04 +08:00
• feb87b1934 [Bugfix] Narrow AuxOutput KV restrictions to known PD connectors (#58150) aoshen02 2026-09-22 23:09:39 +08:00
• 5c121e962a [SpecDecode] Restore residual-logits comments in _resample_kernel (#58166) Lucas WilkinsonandClaude Fable 5.1 2026-09-22 11:07:31 -04:00
• e4340e41c9 [Feat][XPU] VLLM_BATCH_INVARIANT support for Dense/MoE models (#55881) Tony LinandFlora Feng 2026-09-22 23:06:54 +08:00
• 1ea7c63f4a [Bugfix][Structured Output] Disallow MRV1 + PP>1 + async sched + structured output (#56250) Artem PerevedentsevandCNE Pierre FICHEPOIL 2026-09-22 17:00:05 +03:00
• c9b34fdb2d [Bugfix][GLM-5.3-Flash] Run the dense MLP layers on the sequence-parallel shard (#58061) 3 people 2026-09-22 21:58:21 +08:00
• d90f0eade5 [Build] Fix CUDA 12 KV connector dependency selection (#57945) Jee Jee Li 2026-09-22 21:13:07 +08:00
• e548298f81 [Bugfix][Structured Outputs] Reject empty structural_tag at request validation (#47450) 3 people 2026-09-22 21:09:39 +08:00
• 362a64b21d [XPU][CI]Remove model_runner_v2 test from Intel GPU CI (#58050) xiangdongandKunshang Ji 2026-09-22 19:56:01 +08:00
• 9815732c16 [BugFix][Core] Make the structured-output grammar poll non-blocking (#55931) 3 people 2026-09-22 19:39:47 +08:00
• 81d7293c21 [Rust Frontend] Construct model-owned vision processors through specs (#58109) Bugen ZhaoandCodex 2026-09-22 18:01:53 +08:00
• 6dc34b6334 [CPU] Add device-memory-utilization CLI alias (#56547) Louie TsaiandHarry Mellor 2026-09-22 02:37:15 -07:00
• 91d7324cb1 [perf] wire FA and FlashMLA for sm90 GLM5Next NoPE SparseMLA (#55385) 4 people 2026-09-22 17:03:59 +08:00
• 64a48b19b4 [CI] Emit a kernel symbol map from the csrc build (opt-in, for test selection) (#58097) Kevin H. LuuandClaude Fable 5.1 2026-09-22 01:37:34 -07:00
• f92b78f6ef [Kernel][DSV4.1] Fuse MXFP8 wo_b GEMM with sequence-parallel reduce-scatter (#57428) 5 people 2026-09-22 16:07:17 +08:00
• a6c47fbbf4 [Rust Frontend] Separate multimodal instrumentation from request timing (#58084) Bugen Zhao 2026-09-22 14:57:56 +08:00
• 2af05512cf [Kernel] Add opt-in load-time MXFP4 dequantization (#50814) yinfengLiuandShanshan Shen 2026-09-22 14:51:12 +08:00
• 68363d64ec [ROCm][Perf] Avoid extra reshape kernel in Qwen GDN output norm (#47842) Mikko TukiainenandCursor 2026-09-22 09:38:16 +03:00
• 496c6472cb [Rust Frontend] Build full-output grammars from initialized reasoning parsers (#57340) Bugen ZhaoandCodex 2026-09-22 14:11:55 +08:00