Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• 08e03df940 [Bugfix][ROCm] Use a zero default for masked scales in the MXFP8 GEMM (#59454) Mehmet Cagri 2026-10-01 11:32:25 -05:00
• 9aaef4296f [Bugfix][Frontend] Check reused prompt token ids against the vocab before streaming (#59555) Shijie Lyu 2026-10-01 09:26:55 -07:00
• 83cadd65d9 [Bugfix] Tie lm_head.weight for Nemotron Parse when checkpoint omits it (#53020) aniskumar-nvandCyrus Leung 2026-10-01 09:26:33 -07:00
• cfd54ca881 [ROCm][CI] Drop four no-GPU CPU groups from the legacy AMD pipeline (#59595) stefankoncarevicandAndreas Karatzas 2026-10-01 17:10:49 +02:00
• b538d807ff [Bugfix][MRV2] Keep GDN prefill checkpoint metadata local to each cache group (#59536) 3 people 2026-10-01 07:59:15 -07:00
• 5d9214d6b2 [Bugfix][Frontend] Strip x-anthropic-billing-header billing header from /v1/chatcompletions (#59419) 4 people 2026-10-01 16:44:40 +03:00
• f3b77eff64 [Rust][Benchmark] Warn when temperature is left to the server default (#59247) Idder Ghanbaja 2026-10-01 15:03:03 +02:00
• ab5266769e [CI/Build] Add mooncake auto-label rule and assign topic owners (#59192) Nicolò Lucchesi 2026-10-01 12:19:02 +02:00
• cb6458c5ce [Core] Remove deprecated mamba_cache_mode "all" (#58997) 3 people 2026-10-01 17:59:37 +08:00
• 08d77cad7b [Bugfix] Gate Kimi-K3 KDA warmup on sys.modules to skip Kimi import for non-Kimi models (#59257) 5 people 2026-10-01 17:25:45 +08:00
• bcee730b1a [Docs] Remove references to removed env vars (#59530) Sundri Lai 2026-10-01 01:26:14 -07:00
• 24fb5fc08c [Rust Frontend] Add hf parser for response templates (#59005) Bugen ZhaoandClaude Opus 5.5 2026-10-01 16:16:38 +08:00
• 91fcec1235 [Core][Logging] Add built-in JSON formatter (#58739) Mark McLoughlinandOpenAI Codex 2026-10-01 09:11:46 +01:00
• c4fc3c7d32 [Bugfix][Spec Decode] Per-module LM heads for multi-layer MTP on Model Runner V2 (#58921) 3 people 2026-10-01 16:09:14 +08:00
• e5e38ba9b7 [Model Runner V2] Support randomized dummy inputs (#58411) 5 people 2026-10-01 03:52:26 -04:00
• 5addde2a0a [CPU][Zen] Pass f32 weight scales to the zentorch INT8 MoE (#59434) 3 people 2026-10-01 13:21:53 +05:30
• e12b735109 [Test] Make Anthropic messages test compatible with SDK 1.x via extra_body (#57780) jiangyunfan1andmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-10-01 15:38:29 +08:00
• e12291d733 [CPU][Zen] Add DA8W4 (W4A8) int4 support for dense and MoE layers (#54024) 3 people 2026-10-01 12:48:25 +05:30
• 4d6874d137 [GLM 5.3 Perf] Skip qlnorm calculation for MHA, 4.4~7.7% E2E TTFT Improvement (#58845) Wentao Ye 2026-10-01 03:13:57 -04:00
• aee8fe202a [Bugfix][ROCm] Fix race on AITER MLA FP8 prefill scheduling metadata under async scheduling (#58887) 4 people 2026-10-01 09:02:52 +02:00
• 87a4bf664f [Perf][GDN] Slice pure spec-decode rows instead of a host-mask gather (#58763) 3 people 2026-10-01 08:53:06 +02:00
• 4a9d6f9572 [Bugfix] Preserve sampling masks in DELTA and TITO streaming outputs (#55935) 4 people 2026-10-01 14:43:38 +08:00
• c9578f10ab [Qwen3.8-Flash-Next] Fuse main QK-norm/RoPE/gate and KV-cache write into the QSA pre-indexer launch (#57097) Shuolei WangandThien 2026-10-01 14:21:06 +08:00
• 70be6aae19 [Feature][Spec Decode] Support sampling mask replay for MRV2 MTP (#59359) 4 people 2026-10-01 13:51:39 +08:00
• 37cecc8141 [Bugfix][Tokenizer] Fix off-by-one max_token_id from vocab_size (#59491) Shijie Lyu 2026-09-30 22:43:30 -07:00
• b2d3a3b9ad [Rust Frontend] Split argument grammars into schema resolution and per-model rendering (#59395) Bugen ZhaoandClaude Opus 5.5 2026-10-01 13:43:18 +08:00
• ac9126e58a [Rust Frontend] Snapshot structural-tag grammars as readable outlines (#59393) 3 people 2026-10-01 13:09:43 +08:00
• 36229e1337 [Rust Frontend] Token-aware marker parsing for Kimi K3 (#58358) Bugen ZhaoandClaude Opus 5.5 2026-10-01 12:21:50 +08:00
• ec5ef92c87 [Rust Frontend] Anchor tokens after pending UTF-8 bytes at their own first byte (#58357) Bugen ZhaoandClaude Opus 5.5 2026-10-01 12:21:50 +08:00
• 4c2d277643 [Perf][DSv4.1] Faster Engram host lookups: sorted rows, inline big lookups (#59327) Juntian LiuandClaude Opus 5.5 2026-09-30 20:54:44 -07:00
• 4ed438b79c [Perf][Qwen4Exp] Add SM100 low-latency decode GEMM plans (#59214) namgyu-younandThien Tran 2026-10-01 12:40:55 +09:00
• 2eaa3bc5ac [Frontend] Enrich chat_parsing streaming events (#58603) Yoni Gozlan 2026-09-30 23:13:13 -04:00
• a37c402f32 [weight loader] add dtype equality validation between parameter and weight (#51792) Ning Xie 2026-10-01 11:07:31 +08:00
• bfbcde26d5 [Observability] add model initializing duration log (#52141) Ning Xie 2026-10-01 10:54:33 +08:00
• 9e6550bd05 [Bugfix] Register IR providers before hashing configuration (#58756) Nils Matteson 2026-09-30 19:34:28 -07:00
• bc505fc823 [Bugfix] Support CuTe DSL 4.8.0 block-scale API (#59480) 3 people 2026-09-30 22:30:06 -04:00
• 766b4ae54c [Kimi-K3] Add FlashInfer speculative KDA backend (#54255) 4 people 2026-09-30 19:28:50 -07:00
• edaf3e7b61 [Attention][SM120] Occupancy-adaptive Triton split-K segment count (#58482) 3 people 2026-09-30 19:20:33 -07:00
• 76d3929916 [Docker] Compile Python bytecode at image build time (#55422) 3 people 2026-09-30 19:20:17 -07:00
• 7314360c9e [Perf][PP] Skip sampled-token broadcasts whose requests leave the engine (#58542) 3 people 2026-10-01 10:18:55 +08:00
• 34c458d27f [Bugfix] Don't let a structured-output request sample from an unmasked row (#54442) 5 people 2026-10-01 10:11:44 +08:00
• 31ec1c3aa1 [Bugfix][Rust Frontend] Stop vllm-bench chat latency at the last token (#59251) 4 people 2026-09-30 19:04:19 -07:00
• df8fd42116 [CI][Spec Decode] Add async scheduling accuracy tests (#55840) 3 people 2026-09-30 20:06:24 -05:00
• 37d6174096 [Metrics] Add --custom-histogram-buckets to override histogram bucket families (#48867) Guy StoneandClaude Fable 5 2026-09-30 21:03:39 -04:00
• a47e4a5568 [Doc] Score centering via top-k processed logprobs (#59361) 3 people 2026-10-01 08:57:00 +08:00
• c055c1e075 [Core] Include the LoRA path in prefix-cache block hashes (#59335) Nick HillandClaude Opus 5.5 2026-09-30 17:37:09 -07:00
• d7c55116ac [CI] Fix MiniMax-M3 PP aux-state test mock after #58648 (#59508) VenkyandClaude Opus 5.5 2026-09-30 17:34:52 -07:00
• 8bc31bd210 [CI] Deflake the multi-API-server metrics test and Mooncake PD ports (#59507) Kevin H. LuuandClaude Opus 5.5 2026-09-30 17:22:58 -07:00
• 21c08ac458 [CI] Deflake shutdown wait-timeout test by synchronizing on request admission (#58125) jackLei 2026-09-30 17:10:03 -07:00
• 26ce58b48c [ROCm][CI] Sync two AMD test groups between test-amd.yaml and test_areas (#59499) Aarushi JainandAndreas Karatzas 2026-09-30 18:58:07 -05:00
• 8fb16ea5a6 [Docs] Add PR checklist skill for coding agents (#57084) Benjamin ChislettandCopilot 2026-09-30 17:47:28 -04:00
• deb148e730 [Mamba] Add FlashInfer ReplaySSM support for MTP (#52928) Andrii SkliarandAndrii Skliar 2026-09-30 23:41:17 +02:00
• 7e583e615c [Bugfix][Core] Fix mamba prefill checkpoint block reservation and prompt-end eviction in align mode (#59175) kliuaeandNick Hill 2026-10-01 05:26:54 +08:00
• 42f0c17ea7 [Bugfix][PP][Spec Decode] MiniMax-M3 EAGLE3 aux-state relay at PP > 1 and per-stage FlashInfer autotune (#57197) VenkyandClaude Opus 5.5 2026-09-30 14:14:24 -07:00
• 73c7cae4d7 [Bugfix][HiSparse] Stop the host pool feeding device KV cache residency metrics (#58725) 3 people 2026-10-01 04:50:01 +08:00
• 51048b13ce [Docs] Clarify snapshot runtime image support (#59127) Nils Matteson 2026-09-30 13:31:21 -07:00
• 3eb6cec22a [Bugfix][HiSparse] Adopt GPU prefix copies after the hit's allocation (#59282) 3 people 2026-09-30 15:43:09 -04:00
• 84f738f187 [Perf][MiniMax-M3] Triton indexer: decode grid retune + SM12.0 split-K (#56151) 4 people 2026-09-30 12:33:23 -07:00
• b56b54bc54 [ROCm][CI] Relax simple-nemotron-h-8b GSM8K threshold on ROCm (#59477) djramicandAndreas Karatzas 2026-09-30 21:27:21 +02:00
• a21a3f2ec7 [CI] Add supports_multimodal_inputs to test_executor_replace's mock config (#59476) Matthew BonanniandClaude Opus 5.5 2026-09-30 15:04:47 -04:00
• aba01ef1f7 [XPU][CI] Skip test_core_engine_actor_manager.py on Intel CI (#59338) Chaojun ZhangandKunshang Ji 2026-10-01 02:23:02 +08:00
• ed3f6d1a56 [ROCm][CI] Remove duplicate MI355 DPX jobs (#59256) Aarushi JainandAndreas Karatzas 2026-09-30 12:51:20 -05:00
• b33e1cb172 [ROCm]Transpose compressed-tensors MoE weights on device (#59253) Aarushi JainandAndreas Karatzas 2026-09-30 12:44:40 -05:00
• 8f24ab30a3 [ROCm][Kimi-K3][Perf] Fuse MLA decode KV-cache write and Q-prep via AITER (#57640) 3 people 2026-09-30 10:20:05 -07:00
• d2fb35f66e [CI][ROCm] Drop the duplicate OAI Triton MoE run from FP8 MoE Kernels (#59237) stefankoncarevic 2026-09-30 18:24:08 +02:00
• 8873ac83bb [Docs] Add @gau-nernst to CODEOWNERS and committers (#59369) Thien Tranandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-10-01 00:23:40 +08:00
• 0f8b398158 [watermarking] add context deduplication support to speculative decoding (#56807) 3 people 2026-09-30 18:22:10 +02:00
• 73a5831127 [Frontend] Port chat_parsing core from Transformers (#58602) 4 people 2026-09-30 12:21:02 -04:00
• 42a3dd5fec [ROcm][BugFix][The Rock] Update The Rock dockerfile to most recent Triton 3.8 (#59287) rasmith 2026-09-30 11:15:03 -05:00
• ff1b87cca2 [Bugfix][HiSparse] Preserve host prefix publication after request completion (#59007) Yueh-Ting (eop) ChenandMatthew Bonanni 2026-10-01 00:14:51 +08:00
• 279012f94d [Docs] Add @wzhao18 to NVIDIA integration and kv offloading code owners (#59456) Wei ZhaoandCodex 2026-09-30 12:12:12 -04:00
• c32513b99d Add support for unquantized ngram in CT format (#59431) Eldar Kurtić 2026-09-30 18:08:31 +02:00
• c4df37dfcc [ROCm][BugFix][The Rock] Fix mori build for the rock (#59372) rasmithandMisha Goin 2026-09-30 10:51:40 -05:00
• cff08b461e [Security] Fix chat template resource-exhaustion DoS (GHSA-4hhp-h66f-… (#50300) Juan Pérez de AlgabaandCursor 2026-09-30 17:32:00 +02:00
• fc2c801ad9 Disable mypy arg-type and assignment checks in tests (#59428) Harry MellorandClaude Opus 5.5 2026-09-30 16:18:50 +01:00
• 17e9295dd5 [BugFix][Multimodal] Pick worst-case DeepSeek-V4 VL dummy image size (#59271) (#59373) Yida WengandGrok 2026-09-30 23:09:11 +08:00
• 4e1182a3a6 [ROCm][MoE] Pad the AITER MoE intermediate size at allocation time, and round the expert-group count to a kernel that exists (#55368) Sergey Shlyapnikov 2026-09-30 19:08:21 +04:00
• 2df122e65e [ROCm][Perf][GLM-5.3-Flash] "Fit kpool top-k indices to AITER" with a single Triton kernel (#58008) Simon Danielssonandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-09-30 16:45:00 +02:00
• 643bbef55c [Refactor] Share workspace and model runner init between GPU and XPU workers (#59200) aoshen02andClaude Opus 5.5 2026-09-30 22:29:12 +08:00
• eef8b15d7c [Security] Bump pyjwt and rand for remaining GHSAs (#59427) Juan Pérez de AlgabaandCursor 2026-09-30 16:26:16 +02:00
• 98b45c1a94 [Feat] Support PP with PCP in GPU Model Runner V2 (#59139) Qiu ChunshuoandLucas Wilkinson 2026-09-30 07:21:56 -07:00
• 2e52a558f0 [Bugfix][Core] Schedule encoder-only prompts larger than one step (#59029) Giulio De PasqualeandIsotr0py 2026-09-30 15:04:28 +01:00
• 02a3c6dfc5 [Test] Skip IPC weight-transfer test on XPU platforms (#59379) Chaojun ZhangandKunshang Ji 2026-09-30 21:50:40 +08:00
• 677e3f5d50 [CI/Build] Fix pre-commit (#59426) Cyrus Leung 2026-09-30 21:45:28 +08:00
• 5463fe4962 [KV-Offloading][TP] : Expand replicated_layout detection to multi-group MLA (#57652) Varun Sundar Rabindranath 2026-09-30 09:11:03 -04:00
• f28a508162 [Feature][Rust Frontend] Add Shutdown control RPC (#59316) 3 people 2026-09-30 05:49:25 -07:00
• b22494cc0c [Docs] Reinstate docs build gate for PRs (#59411) Harry MellorandClaude Opus 5.5 2026-09-30 13:41:53 +01:00
• effdea9a53 [MM][Mistral] Add compile support for Pixtral vision encoders (#57168) 3 people 2026-09-30 14:27:12 +02:00
• 93dd80f636 [CPU] Conv1d optimised kernel for aarch64 (#54093) almayneandLi, Jiang <jiang1.li@intel.com> 2026-09-30 13:23:42 +01:00
• df5668e5c9 [XPU] Dispatch nn.LayerNorm to fused SYCL kernel via CustomOp (#57172) 3 people 2026-09-30 14:19:33 +02:00
• a4c4bb418e [Security] Bump remaining Dependabot packages (excl. ignored) (#59315) Juan Pérez de AlgabaandCursor 2026-09-30 14:16:31 +02:00
• 3d5625cc3c [Security] Authenticate shared-memory multimodal cache handles (#59357) 3 people 2026-09-30 08:16:26 -04:00
• 72e7874fa6 [Hardware][Power] Enable W4A16 (AWQ & GPTQ) quantization on POWER10 using VSX (#59149) Rukhaiya2004andLi, Jiang <jiang1.li@intel.com> 2026-09-30 17:14:48 +05:30
• 863475eb9c docs: add gemma-2, SmolLM2, Qwen2.5-Coder, Llama-3.2-1B to batch invariance tested models (#54441) yuvalluriaandClaude Sonnet 4.6 2026-09-30 14:13:23 +03:00
• 5a4351ab09 [MM] Keep device input normalization fused when encoder compilation is enabled (#59195) Jakub ZakrzewskiandCodex 2026-09-30 13:08:24 +02:00
• 5225f1bacb [MyPy][3/N] Fix MyPy errors in test groups (part 3) (#55939) 6 people 2026-09-30 19:06:07 +08:00
• c4fa6f36a7 [Tests][Multimodal] Cover scoped processor kwargs precedence (#59399) apex-mochen 2026-09-30 18:54:32 +08:00
• 0103a9b96f [Quantization] Add per-token NVFP4 CuTe-DSL MoE backend (#50030) 4 people 2026-09-30 03:42:03 -07:00
• d5ee309314 [UX] Tag torch.compile log lines with the component being compiled (#48133) 3 people 2026-09-30 12:34:31 +02:00
• d71f662606 [Bugfix][Model][CPU] Fix Mamba2 quantized in_proj weight and scale loading for TP >1 (#58083) 4 people 2026-09-30 15:29:02 +05:30