Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• 1bc6e072b0 [Rust Frontend] Add per-request timing metrics to chat and completions (#59314) 3 people 2026-10-12 15:39:10 -07:00
• 802feb6770 [Bugfix][KVConnector] Fix MoRI-IO WRITE-mode requests stranded in WAITING_FOR_REMOTE_KVS (#57705) 6 people 2026-10-13 00:29:31 +02:00
• 44f0ddfd7b [Bugfix][Parser] Stop re-parsing the whole GLM tool argument per chunk (#61253) Maroon AyoubandFlora Feng 2026-10-13 00:28:35 +03:00
• 69d54eb078 [Docs] Fix broken llm-d.ai links on the llm-d integration page (#61022) Almog TavorandRobert Shaw 2026-10-13 00:26:43 +03:00
• fe341a53dd [HiSparse] Report host pool in KV cache capacity logs (#60285) Matthew BonanniandClaude Opus 5.5 2026-10-12 16:49:49 -04:00
• deedc7be00 [Perf][Humming] Skip the redundant MoE input copy before the w13 GEMM (#59340) 3 people 2026-10-13 05:24:12 +09:00
• 654b725ea9 [CI] Pace the CRCR nightly report under the relay rate limit (#61269) Andrey Talman 2026-10-12 16:23:53 -04:00
• abad32a718 [ROCm][Perf] Fix Gemma AR+RMS prefill fallback (#58547) Mikko Tukiainen 2026-10-12 22:53:08 +03:00
• fabaecbfa5 [Misc] Expand @ZJY0516 CODEOWNERS to Mamba and KV cache (#61181) Jiangyun Zhu 2026-10-13 03:32:25 +08:00
• 7039c7f17b [Bugfix][Rust Frontend] Omit unset sampling params in render output (#61220) Martin Hickey 2026-10-12 20:31:03 +01:00
• 1bdd3fa2d5 [Bugfix][Core] Check multimodal chunk boundaries before encoder cache reuse (#59721) ztysthandOpenAI Codex 2026-10-13 03:25:35 +08:00
• 08582942d8 [Bugfix][Frontend] Honor request_id and X-Request-Id in chat and completions render (#61236) Martin Hickey 2026-10-12 20:17:12 +01:00
• d0948f754d [Bug] Honor spec decode draft model config (#61240) Wentao Ye 2026-10-12 15:13:23 -04:00
• d6aca837c3 [Attention] Make the sparse MLA prefill workspace size configurable (#61023) Maroon AyoubandClaude 2026-10-12 21:56:51 +03:00
• 56e3b081ae [Bugfix][weight loader] Stop warning when a parameter only widens the weight (#60020) SeongJun LeeandNing Xie 2026-10-13 03:33:31 +09:00
• 8055293353 [ROCm][MLA] Cut ~21x host dispatch from the MLA metadata build (#58381) 4 people 2026-10-12 11:11:20 -07:00
• c9e62f13c2 [CI][ROCm] Add AITER nightly overlay scripts (#60883) 3 people 2026-10-12 13:04:37 -05:00
• e1a5086256 [Attention] Remove cascade attention (#61126) Woosuk KwonandClaude Opus 5.5 2026-10-12 10:41:07 -07:00
• 5b0b8f251e [Bugfix][HiSparse] Fix TP group hang on steps without scheduled tokens (#60941) 4 people 2026-10-12 20:38:12 +03:00
• a7b7bc1fbe [Metrics] Count finished requests by watermark status (#58715) 3 people 2026-10-12 19:21:22 +02:00
• 45bb710eca [CI] Add nightly Python 3.14 smoke test (#60456) Misha GoinandClaude Opus 5.5 2026-10-12 13:11:26 -04:00
• 12e6a585ae [Fix][Compile] Declare group_shape in the static FP8 quant fusion … (#61108) Laith Sakkaandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-10-12 10:09:57 -07:00
• c599b5971f [Bugfix][Frontend] Return all logprobs when completions requests logprobs=-1 (#55717) Chester 2026-10-13 01:05:55 +08:00
• 6e6e86078a [Bugfix][Platform] Don't resolve current_platform during import vllm (#60766) haotiaa 2026-10-12 10:05:34 -07:00
• fa39582756 [Frontend] Serve LiquidAI/d1-3B on /v1/systemone (#60676) 3 people 2026-10-12 13:00:47 -04:00
• d79b52b182 [Misc] Consolidate csrc shared headers and prune dead code (#60721) Misha GoinandClaude 2026-10-12 12:46:22 -04:00
• e0317b28ce [Feature] Support batch invariance on ROCm (#52231) 3 people 2026-10-12 08:53:26 -07:00
• 50b21a2667 [Perf] [Feat] [ROCm] Add densemha support to ROCm AITER Sparse MLA (#49263) TJian 2026-10-12 23:50:28 +08:00
• a686c8e1b7 [Triton] Migrate Triton kernels off tl.make_block_ptr for triton-main compat (#60502) Nick Riasanovsky 2026-10-12 11:38:43 -04:00
• ffec221276 [Bugfix][DSv4.1] Match decoder-replay input_ids dtype for FlashInfer MegaMoE (#61041) Shang WangandClaude Opus 5.5 2026-10-12 10:23:00 -04:00
• f3cb10b690 [Model][LoRA] Enable LoRA support for BERT and RoBERTa embedding models (#61033) Hoai VoandClaude Opus 5.5 2026-10-12 10:09:53 -04:00
• 716961cf1f [Bugfix] Drop enable_in_kernel_fc2_reduce from FlashInfer CuteDSL MegaMoE config (#59900) Shang WangandClaude Opus 5.5 2026-10-12 10:05:09 -04:00
• a281ad32f0 [EPLB] Add EPLB support to dspark draft models (DeepSeekV4) (#57549) Ilya Markov 2026-10-12 16:03:09 +02:00
• b3ee560ae8 [Bugfix][MoE] Apply expert_map remapping for CUTLASS FP4 MOE expert parallel (#59042) Ilya Markov 2026-10-12 16:02:56 +02:00
• f022975964 [Bugfix] Reject beam search in completion render (#61212) bzsuni 2026-10-12 21:49:46 +08:00
• 4c36d21280 [Bugfix][LoRA] Apply PEFT rank_pattern / alpha_pattern scaling per module (#59801) 3 people 2026-10-12 17:49:30 +04:00
• 3facf822b7 [Performance] Load HF tokenizers and processors from a local snapshot (#60825) 3 people 2026-10-12 14:45:54 +01:00
• 092c35dd15 [Model] Support Cloudflare/clef-omni on /v1/systemone (#61064) 3 people 2026-10-12 21:27:34 +08:00
• 314ab8cffd [Bugfix][Structured Output] Drain compilation tasks before releasing the backend (#61047) DimensionSTPandArtem Perevedentsev 2026-10-12 22:25:36 +09:00
• ac28877f0e [Bugfix] Fix out-of-bounds lookahead in adaptive block verification (#60945) Chauncey 2026-10-12 21:21:35 +08:00
• 8c175f403a [CI][ROCm] Compare greedy output only in the AITER attn+quant test (#61185) stefankoncarevic 2026-10-12 15:20:35 +02:00
• 7900d2eabc [Test] Accept tied tokens at the k-th value in the Triton top-k test (#60726) Rohan PotdarandClaude Opus 5.5 2026-10-12 08:18:44 -05:00
• c75cb63cd2 [Qwen4Exp] support Sequence Parallelism for Qwen3.8-Flash-Next (#56322) Harry Huangandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-10-12 20:56:44 +08:00
• 258b97ff6b [Bugfix][EC Connector] Size transfers by measured encoder output widths (#57904) Tianyu Guo 2026-10-12 20:32:39 +08:00
• aa1d44df16 [Bugfix][Rust Frontend] Alias qwen3_xml to qwen3_coder and add qwen3_json for the Qwen JSON format (#60136) Liana KolevaandBugen Zhao 2026-10-12 05:20:53 -07:00
• 8d7b93fec2 [Frontend] Report stop_reason on /inference/v1/generate and pass it through derender (#60251) 5 people 2026-10-12 04:37:53 -07:00
• 32b05addc1 [Bugfix][Rust Frontend] Restore DP wave delivery after XPUB reconnects (#60249) 3 people 2026-10-12 19:12:43 +08:00
• f5cda48c90 [Bugfix][Rust Frontend] Export corrupted request metric (#59937) ReidandBugen Zhao 2026-10-12 19:12:39 +08:00
• 3b9e462774 [Bugfix][Rust Frontend] Correct DP routing after request removal (#59631) 3 people 2026-10-12 16:42:35 +05:30
• 21445806bd [Rust Frontend] Support logprob_token_ids on OpenAI endpoints (#59549) 4 people 2026-10-12 19:12:30 +08:00
• ee8f7ea449 [CI] Enroll a fourth batch of five steps in automatic sharding (#60858) Thang NguyenandClaude Opus 5.5 2026-10-12 06:46:07 -04:00
• f7a3a7265f [Bugfix][Multimodal] Fix Qwen3-ASR audio preprocessing parity (#57146) MatthewCYM 2026-10-12 18:42:51 +08:00
• 2534a96fcf [Docs] List additional unauthenticated endpoints (#61081) Jimmy WangandJuan Pérez de Algaba 2026-10-12 17:47:52 +08:00
• 5a09062632 [Rust Frontend] Compose answer constraints into parser-owned output grammars (#60781) Bugen ZhaoandClaude Opus 5.5 2026-10-12 17:43:48 +08:00
• 0466bd36ea [Rust Frontend] Build the Kimi K3 full-output grammar from its parser mode (#59190) 3 people 2026-10-12 17:43:47 +08:00
• 83773ecdf3 [Frontend] Support CustomToolCall in the non-Harmony Responses API (#57322) Shijin ZhangandChauncey 2026-10-12 17:22:55 +08:00
• a33d46b471 [Bugfix][XPU] Forward softcap and alibi_slopes in xpu_ops.flash_attn_varlen_func (#58936) zhangchengzhucufe-devandKunshang Ji 2026-10-12 17:10:55 +08:00
• e05201aa2e [XPU] refine memory release for sleep mode (#61158) Yan Ma 2026-10-12 16:59:38 +08:00
• 99b44c7a42 [Metrics] Add admission rejection metrics (#55812) Thuongvu HoandNicolò Lucchesi 2026-10-12 01:31:07 -07:00
• 5bed289707 [DX] Update CODEOWNERS for entrypoints (#60984) Cyrus Leung 2026-10-12 16:06:47 +08:00
• 4163a80142 [Rust Frontend] Return weight-version spans on final outputs (#61070) 4 people 2026-10-12 15:56:59 +08:00
• f4114e068e [Bugfix][Frontend] Preserve token IDs in chat derender responses (#61065) Ting SUN 2026-10-12 15:55:06 +08:00
• e3bbaf2369 [ROCm][DSv4] Fuse the shared expert into AITER MegaMoE (#61004) AMD-yanfeiwangandaoshen02 2026-10-12 15:53:16 +08:00
• 36f3972812 [Spec Decode] Skip metadata build for dflash/dspark when all model backends support it (#60692) Giancarlo DelfinandRishi Puri 2026-10-12 00:09:10 -07:00
• 6a70913590 [XPU] bump up auto-round-lib to 0.16.0 (#60976) Zhenzhong XuandKunshang Ji 2026-10-12 14:47:59 +08:00
• 74f4404f58 [Bugfix][Distributed] Own SHM out-of-band buffers before releasing read slots (#61000) 3 people 2026-10-12 07:44:12 +02:00
• fe4df3df0e [Bugfix][GLM-5.3-Flash] Restore MTP index sharing across draft iterations (#61079) Canlin GuoandClaude Opus 5.5 2026-10-12 13:41:52 +08:00
• d4e92ccc37 [Bugfix][Frontend] Honor endpoint plugin route overrides (#58839) 3 people 2026-10-12 11:58:15 +08:00
• bfa7e5b7af [XPU][CI] Enable XPU graph in graph tests (#60784) liuzhenweiandKunshang Ji 2026-10-12 10:39:40 +08:00
• f147949be8 [XPU] limit FA kernel block size to multipleOf 64 (#57297) Yan Ma 2026-10-12 10:25:32 +08:00
• 124b89d5b7 [Bugfix][RecoverSSM] Correct aligned state indices at exact block boundaries (#59962) dingdangmaoandCodex 2026-10-12 10:08:14 +08:00
• 03091e28b8 [Bugfix] Read Ling quantization_config from top-level config for Bailing V3 VL (#59026) Michał GanczarenkoandClaude Opus 5.5 2026-10-12 03:25:29 +02:00
• cc9b0f097d [Fix] MiniMax: fix CUDA IPC handle exchange on cuda-bindings >= 13.4 (#60917) Laith Sakka 2026-10-11 17:38:56 -07:00
• 3fe917abb1 [Bugfix][MRV2] Fix FlashInfer attention and ReplaySSM warmups on MRV2 (#61013) Nick Hill 2026-10-11 16:04:41 -07:00
• f26d9f432c [Bugfix][GLM-5.3-Flash] Honor --mamba-ssm-cache-dtype for the KDA recurrent state (#59147) 3 people 2026-10-12 01:55:00 +08:00
• c059bee812 [MRV2] Switch CI tests still running on MRV1 unnecessarily (#61014) Nick Hill 2026-10-11 09:57:07 -07:00
• b99fdbe1f8 [Bugfix] Respect the configured offline chat template (#61061) Ting SUNandCyrus Leung 2026-10-12 00:27:21 +08:00
• 93d2cab35e [Security] Fix P2P ZMQ socket exhaustion via unbounded peer sessions (#51504) 3 people 2026-10-11 17:39:11 +02:00
• 239a3fa3a8 [Sleep] Retain draft-private weights with existing parameter selector (#61046) aoshen02andCodex 2026-10-11 23:30:56 +08:00
• 3e8e1364ff [Perf][GLM-5.3-Flash] Shard long-context indexer prefill rows across TP (#54951) 6 people 2026-10-11 22:43:37 +08:00
• bd5bbce6e4 [RL] Release weight-transfer communicators without stopping inference (#61060) aoshen02 2026-10-11 21:15:14 +08:00
• c85616efc6 [Model] Support MiniCPM-V 4.7 (#58674) tc-mbandCyrus Leung 2026-10-11 18:21:10 +08:00
• f425a21208 [Bugfix][Frontend] Support scalar priority and wire priority into LLM.chat() (#57243) Subhash Polisetti 2026-10-11 03:03:33 -07:00
• 2339bd39d6 [Bugfix][Mooncake] Use the EAGLE replay boundary when retaining sliding-window tails (#61012) 4 people 2026-10-11 02:15:18 -07:00
• 1789f09588 [Frontend] Explain disabled scale-out endpoints with opt-in hints (#61042) aoshen02andOpenAI Codex 2026-10-11 16:58:07 +08:00
• fcae5c9887 [Security] Restrict default CORS origins (#61009) Juan Pérez de AlgabaandCyrus Leung 2026-10-11 10:26:02 +02:00
• 26a488ad93 [Fast Start][Bugfix] Snapshot GPU memory after external weights are ready (#60966) Xun Sun 2026-10-11 15:10:04 +08:00
• ef42b4385f [Bugfix] Reclaim cached GPU memory during weight loading (#57516) 3 people 2026-10-11 00:03:06 -07:00
• 61f084b5f3 [Bugfix][Multimodal] Fix per-video cache option reuse (#55203) Bruno VolpatoandCodex 2026-10-11 02:21:55 -04:00
• 2eb491e18b [Perf][MiniMax-M3] MSA prefill attends with the E4M3 query; bump fmha_sm100 to 968061a (#60684) Yongye ZhuandClaude Opus 5.5 2026-10-10 22:40:45 -07:00
• f8471164bf [MRV2] Detatch weight-reload logic from MRV1 (#60937) Nick Hill 2026-10-10 21:15:05 -07:00
• 01abe975b5 [Bugfix][KV Connector] Fix speculative block reservations across KV cache groups (#59880) bzsuni 2026-10-11 11:55:41 +08:00
• b89852aad0 [RL][Frontend] Return token-level weight-version spans on final outputs (#59582) Shuolei Wangandaoshen02 2026-10-11 11:33:29 +08:00
• 3c062d391d [Bugfix] Support 0-dim tensors in packed weight transfer (#52235) 3 people 2026-10-10 18:41:06 -07:00
• 276fbcff27 [Deprecation] Remove scheduled deprecated items (#60923) Wentao Ye 2026-10-10 15:45:33 -04:00
• 187a0eb98a [Bugfix][Frontend] Forward --enable-log-outputs to the Anthropic handler (#60982) Yufeng He 2026-10-11 00:36:03 +08:00
• f2376cdc4d [Fast Start] Preload the FlashInfer autotune table on the weight cache daemon (#60085) 3 people 2026-10-10 23:48:06 +08:00
• 3709632ff2 [Refactor] Remove dead code multiple places (#60120) Wentao Ye 2026-10-10 11:41:10 -04:00
• 838526c7e5 [Compile] Fix sampler compile warning (#60680) Wentao Ye 2026-10-10 11:09:45 -04:00
• bf16c9c139 [Bugfix][Model] K-EXAONE: restore loading of pre-2.0 configs and apply RoPE to the MTP layer (#60187) Mook 2026-10-11 00:05:55 +09:00