Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• eb61707851 [Bugfix] Fix stale HPC QK-norm weights after weight refit (#56706) aoshen02 2026-09-15 12:34:41 +08:00
• 7dbc3d62c3 [Cleanup] Remove vestigial tpu_input_batch.py (#56898) Nick Hill 2026-09-14 21:29:15 -07:00
• f84b0c4bce [Model Runner V2] Tolerate lack of pinned memory support (#56908) Nick Hill 2026-09-14 21:28:32 -07:00
• 45779ef99f [Bugfix][NIXL] Fix multiple handles xfer race (#56104) Nicolò LucchesiandNick Hill 2026-09-15 06:27:10 +02:00
• caf56ce9da [CI] Pass the scale-out endpoint flag in EC E2E (#56819) 3 people 2026-09-14 21:25:30 -07:00
• a491b6039d [CI] Drop root privileges for the OTel GPU sampler (#56941) Kevin H. LuuandOpenAI Codex 2026-09-14 21:20:14 -07:00
• 8765952755 [CI/Build] Retry CPU image stream interruptions (#56597) 4 people 2026-09-14 21:18:10 -07:00
• 9d14fd3095 [Kernel][MoE] DeepEP v2: async finalize to overlap shared experts with combine (#52781) 3 people 2026-09-15 12:04:22 +08:00
• a72c046463 [Bugfix] Avoid nested rerank cancellation handlers (#55602) Miguel 2026-09-15 05:19:08 +02:00
• 381c61e008 [Bugfix][NIXL] Report a full prefix cache hit as finished (#56640) Yizheng JiaoandNick Hill 2026-09-14 20:15:19 -07:00
• e6eb0d120c [Model][DSv4.1] Store the whole KV in MXFP8 (FlashMLA V4.1 record) (#56893) 3 people 2026-09-14 20:12:55 -07:00
• a529c1a748 [Bugfix] Handle bare and malformed tool call openers in Gemma4 parser (#53444) wuhangxianandChauncey 2026-09-15 10:52:43 +08:00
• e45bb43984 [Bugfix][KV Offload] Ignore pending chunks in invalid sliding windows (#56486) Liqian Lin 2026-09-15 10:48:38 +08:00
• 567f745c1b [Bugfix] Export weight-cache IPC tensors separately for each client (#56472) 3 people 2026-09-14 19:43:42 -07:00
• 17bc3e106a [KVConnector][P2P] Configurable unbound-store timeout and one-RTT rejection of a late fetch (#53453) liranschour 2026-09-15 05:38:18 +03:00
• a08dffe215 [XPU][CI] update gpt-oss package version (#56783) liuzhenwei 2026-09-15 10:23:14 +08:00
• 3192898754 [ROCm][CI] Prepare TheRock image for CI (#56921) MattandAndreas Karatzas 2026-09-14 18:57:59 -07:00
• ceaa9f95dd [Minor] Invert MRV1 specdec method fallback logic (#56899) Nick Hill 2026-09-14 18:56:04 -07:00
• 84e5f3edac [Bugfix] Validate routed-expert prompt offsets before engine submission (#56844) 3 people 2026-09-15 09:17:41 +08:00
• ef5f7cd119 [ROCm][DSv4.1][Perf] Dequantize the MXFP8 weight once when dot_scaled cannot be used (#56560) JohnQinAMDandClaude Opus 5 2026-09-14 18:07:26 -07:00
• 772e4f05fa [Bugfix][Rust] Fix HF multi-turn dataset integration (#56915) Kevin H. LuuandOpenAI Codex 2026-09-14 18:04:46 -07:00
• a7576447b8 [ROCm][Bugfix] Initialize Ray NIXL agents for sharded RDT (#56687) Andreas KaratzasandCodex 2026-09-14 19:34:21 -05:00
• bbbd0a02c9 [Bugfix] Carry over queued work when materializing the dedicated stream (#56382) zhaoguochun1995 2026-09-15 07:52:20 +08:00
• f0a61bd438 [Rust][Benchmark] Support HF ShareGPT datasets in multi-turn mode (#51104) 3 people 2026-09-15 07:48:23 +08:00
• ba2ae9f239 [MRV2] Run pooling post processing on non-final PP ranks (#56666) Taneem Ibrahimandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-09-14 18:11:11 -05:00
• 3bb0a03f35 [Model Runner V2] Acceptance estimation for adaptive verification (#52228) Giancarlo Delfin 2026-09-14 15:58:48 -07:00
• 8e08cef46e [Bugfix] Redact credentials from benchmark logs (#56662) Taneem Ibrahimandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-09-14 17:58:33 -05:00
• 00972dfd72 [ROCm][DSV4.1][Perf] Fold the mHC post step into the delayed pre projection (#56513) Fangzhou AiandCursor 2026-09-14 17:44:56 -04:00
• bdad63c90a [CI] Check target branch freshness before starting CI (#56169) Andreas Karatzas 2026-09-14 16:40:35 -05:00
• d320a21431 [Bugfix][Rust Frontend] Restore prost test dependency (#56866) Kevin H. LuuandSherlock 2026-09-14 14:37:44 -07:00
• d392ac836e [Doc][Metrics] Fix spec-decode PromQL examples to use the exposed names (#55016) Michele Campi 2026-09-14 23:15:28 +02:00
• dabc4362b4 [ROCm][DSV4.1][Perf] Stride the DSA decode candidate mask over the live context (#56628) Fangzhou AiandCursor 2026-09-14 16:47:32 -04:00
• 7702ee87db [Bugfix][DSA] Write nvfp4_ds_mla from the fused norm+rope kernel (#55538) sychen52 2026-09-14 13:25:33 -07:00
• d4ee7fe7a9 [Quantization] Select linear backends per quantization (#51204) Netanel Haber 2026-09-14 23:24:18 +03:00
• 995e8581f4 [Model] Voxtral Realtime: add support for CUDAGraphMode.FULL_DECODE_ONLY (#51167) Nicolò LucchesiandJorge Gallego Feliciano 2026-09-14 21:16:39 +02:00
• 0a5747d410 [Profiler] Add Proton CUDA graph attribution for MRV2 (#51084) 鐘天楽andNick Hill 2026-09-14 15:02:07 -04:00
• 5372e72a98 [Perf][DSv4.1] Fold the mHC post block into the delayed pre projection (#56633) Yongye Zhu 2026-09-14 11:57:37 -07:00
• 56ca9904ce [CI] Initialize ubatch runner in cudagraph unit test (#56808) Kevin H. LuuandOpenAI Codex 2026-09-14 11:37:08 -07:00
• e85c8826ce [Bugfix] Make KV cache and MFU log lines backend-neutral (#55650) Zhiyao(Ian) YuandClaude Opus 5 2026-09-14 14:00:20 -04:00
• 03dc26e639 [Perf][Kernel][Quantization] Fuse ReLU2 with static FP8 activation quantization (#53793) 6 people 2026-09-14 19:26:15 +02:00
• 435c96f9db [V1][Metrics] Support Sliding Window Attention (SWA) and hybrid layers in MFU/MBU estimation (#55624) Thillai ChithambaramandMark McLoughlin 2026-09-14 12:04:55 -04:00
• b7e8dd8f37 [watermarking hardening] Add e2e test and basic GSM8K quality tests (#56309) 3 people 2026-09-14 17:29:13 +02:00
• 1713a09491 [Rust Frontend] Forward per-request watermarking controls (#56338) Nicolò LucchesiandCodex 2026-09-14 17:17:02 +02:00
• b3124a8237 [Model] Add support for Nanbeige4.2 (transformers backend) (#56071) 3 people 2026-09-14 23:06:01 +08:00
• 4be3dcf0fc [PCP][DCP] Declare FlashMLASparse MTP support at CP interleave > 1 (#56722) Lucas WilkinsonandClaude Opus 5 2026-09-14 10:20:46 -04:00
• ff5f6d41b1 [Bugfix] Scale KV page size for hidden states extraction with TP (#50894) 6 people 2026-09-14 16:07:09 +02:00
• c676e4930b [Spec Decode] Fix Qwen3 DSpark d2t requirement for padded-vocab drafts (#55133) 4 people 2026-09-14 16:07:07 +02:00
• 767d1c4d47 [Docs] Move russellb to emeritus committer (#56467) Russell BryantandClaude Opus 5 2026-09-14 09:44:50 -04:00
• 9d4d9aa5bc [LoRA][Refactor] Unify multimodal LoRA token count hooks (#55071) linitra24 2026-09-14 21:19:15 +08:00
• 6623fe5b4e [Bugfix] MiniCPM-V 4.6: fix ViT self-attn qkv weight loading (#49417) arkohutandHarry Mellor 2026-09-14 21:18:55 +08:00
• 47fbd36e6d [Bugfix] Trim stale consequence claims from unannotated-eagle warning (#56791) Jiangyun ZhuandKimi Code 2026-09-14 21:06:26 +08:00
• e0c04c7b4d [Bugfix][Models] Fix OpenPangu sleep mode with static sinks (#53696) Ronald 2026-09-14 21:05:47 +08:00
• 1d0d1081c4 [ROCm][Connector] SWA+HMA-support in MoRI-IO connector (Gemma4) (#53721) Simon Danielsson 2026-09-14 14:52:56 +02:00
• 3f55ad2f07 [Qwen3.8-Flash-Next] Fuse PLE residual and QSA output gate (#55309) 3 people 2026-09-14 20:16:25 +08:00
• c612e2bff0 [Bugfix][KV Offload] Reuse in-flight async lookup probes (#55823) 3 people 2026-09-14 20:02:45 +08:00
• a5f6f61a8a [CI][CPU] Add speculative-decoding coverage to CPU CI (#54934) 3 people 2026-09-14 17:21:38 +05:30
• a6c5d6d0fc [ROCm][Perf] Enable CSA multi-stream overlap for DeepSeek-V4 (#51794) Shanshan Shen 2026-09-14 19:21:30 +08:00
• e82794f48f [CI][Test] Mock CPU backend block sizes in kv_connector unit conftest (#56671) Xiaozhen BiandClaude Opus 5 2026-09-14 03:48:53 -07:00
• 7b1ea3f524 [Bugfix][Rust Frontend][Multimodal] Align DeepSeek V4.1 and Kimi K3 media with rendered placeholders (#56366) 3 people 2026-09-14 18:37:59 +08:00
• 0ca7aef4e3 [Bugfix][NixlPush] Guard _remote_agents read in _do_send_reg_notif (X1) (#56317) Nicolò Lucchesi 2026-09-14 12:37:12 +02:00
• ea723c81c3 [Security] Cap Qwen-VL video sampling knobs (#56729) Juan Pérez de AlgabaandCursor 2026-09-14 12:22:18 +02:00
• 1678b39627 [Bugfix][EPD] Preserve media processing options in encoder requests (#56786) jiangkuaixue123 2026-09-14 18:05:13 +08:00
• 934b1fcbcb [Frontend] Replace VLLM_ENABLE_SCALE_OUT_ENDPOINTS with --enable-scale-out (#55176) Martin Hickey 2026-09-14 11:02:14 +01:00
• 2c7ee87223 [Bugfix][Mooncake] Report failed remote KV loads to the scheduler (#50984) 3 people 2026-09-14 17:55:39 +08:00
• 5236bef721 [CI] Keep OTel bytecode out of mounted checkouts (#56766) Kevin H. LuuandOpenAI Codex 2026-09-14 02:52:45 -07:00
• dc2e8f1157 [ROCm][Perf] W4A16: keep skinny GEMM zero-points packed 4-bit (#54965) Matthias Gehre 2026-09-14 11:40:25 +02:00
• 238cb2b191 [Perf][GLM-5.3-Flash] Dense/masked-MHA sparse prefill for the NoPE (256, 0, 256) layout + skip the NoPE K concat (#55738) 3 people 2026-09-14 17:36:39 +08:00
• 79f0be21ff [Bugfix][CPU] Fix DeepSeek-R1 (FP8 MLA + MoE) correctness on CPU backend (#56773) Li, Jiang 2026-09-14 17:30:28 +08:00
• dc89fdfb0e [CPU][Profiler] Group torch profiler tables by input shape when record_shapes is on (#56016) Chinmay-Kulkarni-AMDandLi, Jiang <jiang1.li@intel.com> 2026-09-14 14:46:48 +05:30
• b443c1cc4e [Perf][GLM-5.3-Flash] Use FlashKDA for KDA chunked prefill (1.7-3.8x faster than the Triton chunk path) (#55737) 3 people 2026-09-14 16:36:19 +08:00
• 23cfaad497 [CI] Update entrypoints CI (#56763) wang.yuqi 2026-09-14 15:59:51 +08:00
• 78e84261ab [Performance] Add Triton kernel for Gemma3n sparse GELU (#48498) 3 people 2026-09-14 16:42:15 +09:00
• e6b4e47d2d [Bugfix][Rust Frontend] Preserve selected-token logprob mode (#56406) AlecandOpenAI Codex 2026-09-14 00:05:33 -07:00
• 663d7f679e [Bugfix] Fix --lora-modules name=path parsing when path contains '=' (#53353) Hung Hoang (黄兴） 2026-09-14 13:55:21 +07:00
• dbf49dad11 [Fast Start] Use GPU uuid as socket folder identifier (#56669) Isotr0py 2026-09-14 14:28:13 +08:00
• cf1584f373 [Refactor] Normalize DeepSeek V4.1 model package naming (#56741) Bugen Zhao 2026-09-14 14:17:50 +08:00
• 10e8d5614c [Bugfix][Beam Search] Respect skip_special_tokens during decoding (#56211) Alexander LeeandOpenAI Codex 2026-09-14 01:49:44 -04:00
• 73d2a8cf86 [2/2][Model Runner V2] FULL CUDA graph capture for microbatched steps (DBO) (#51700) Anjie HouandCodex 2026-09-14 13:42:33 +08:00
• a66a3e1b4e Merge remote-tracking branch 'origin/main' into sync/lesj/qwen4exp-bugfixes lesj0610 2026-09-14 14:36:05 +09:00
• 67111973ee [Frontend] Move grpc_server to launchers. (#56746) wang.yuqi 2026-09-14 13:03:08 +08:00
• dc36fcce90 [CI] Collect GPU memory telemetry for MIG slices (#56747) Kevin H. LuuandOpenAI Codex 2026-09-13 21:30:29 -07:00
• 214248c7d8 [CI] Move smaller H200 workloads to 18GB MIG slices (#56695) Kevin H. LuuandOpenAI Codex 2026-09-13 21:30:11 -07:00
• a2685f2cda [Bugfix][Multimodal] Validate base64 video payloads, matching image and audio (#54323) Hotragn PettuganiandIsotr0py 2026-09-14 00:14:26 -04:00
• 39545e475d Revert "[CI][XPU] Disable model runner V2 for XPU quantization test for some partially pre-quantized models" (#56394) Chaojun Zhang 2026-09-14 11:40:13 +08:00
• 7632f767f0 [Rust Frontend] Support HTTP RL weight synchronization (#56567) 3 people 2026-09-14 11:37:04 +08:00
• 2d9c30a7b5 [CI][Bugfix] Complete FSE fixture R-SWA contract (#56710) Kevin H. Luu 2026-09-13 20:30:25 -07:00
• 52dd0d7562 [Fast Start] Fast loader support nnode>1 (#55468) siyuandIsotr0py 2026-09-14 11:05:54 +08:00
• eb42686a30 [Bugfix][Frontend] Support Responses text types in DeepSeek V4.1 (#56299) CedricHuangandOpenAI Codex 2026-09-14 10:43:09 +08:00
• dfa1984e58 [ROCm][Perf] Tune MiniMax-M3 decode top-k for short contexts (#55235) Fangzhou Ai 2026-09-13 22:36:27 -04:00
• 58d45fd767 [Bugfix][KV Connector] MooncakeStore: exclude non-prefix-cacheable (QSA ring) groups; fix align-mode check (#55027) 4 people 2026-09-13 19:18:16 -07:00
• fe49665eb3 [XPU][CI] update test requirements (#56704) liuzhenwei 2026-09-14 10:01:50 +08:00
• 3b533197a9 [Perf] Parallelize mHC pre-norm JIT warmup (#56683) Woosuk Kwon 2026-09-13 18:58:20 -07:00
• 9f03b510c3 [DSv4 Bug] fix dsv4 start up error NotImplementedError: DeepSeek V4 MegaMoE currently requires expert parallel (#55914) Wentao Ye 2026-09-13 20:22:08 -04:00
• 3b1911c0ca [Bugfix][Model] Fix Qwen4Exp MTP buffer placement, config parsing, and Triton warmup lesj0610andClaude Opus 5 2026-09-14 07:25:01 +09:00
• 319cc5ef19 [Performance][EPD] Reduce Python proxy serialization overhead (#56657) Tianyu Guo 2026-09-14 06:07:12 +08:00
• b23433088b [Attention] Remove DCP indexer interleave guard and test TP1 output parity (#55802) 6 people 2026-09-13 13:56:28 -07:00
• 186a1e6222 [Warmup] Gemma 4 de-JITification (#55768) Roberto L. CastroandCodex 2026-09-13 22:47:21 +02:00
• 586f652f8d [Bugfix][PCP][DCP] Respect interleave in indexer KV gather mapping (#56715) Lucas WilkinsonandOpenAI Codex 2026-09-13 16:45:17 -04:00
• 7fe8fc803d [NIXL][PCP][DCP] Expose PCP producer KV shards as transfer ranks (#56645) Lucas WilkinsonandOpenAI Codex 2026-09-13 16:23:15 -04:00
• 71888f507a [Perf] Avoid redundant conversions in FP8 dummy initialization (#56688) Woosuk KwonandOpenAI Codex 2026-09-13 12:06:13 -07:00