Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• 22a3cbe152 [ROCm] aiter_unified_attn fp8 q scale refactor (#38296) Divakar Verma 2026-05-06 12:11:36 -04:00
• d5b31c954d [Bugfix] Account for truncate_prompt_tokens when computing max_tokens (#41800) Viktor Pus 2026-05-06 18:10:17 +02:00
• ee38750a75 [Bugfix] Fix spawn_new_process_for_each_test silently swallowing test failures (#41423) David Zheng 2026-05-06 08:17:15 -07:00
• 27e0057aed [Spec Decode] Add Gemma4 MTP speculative decoding support (#41745) Luciano MartinsandLuciano Martins 2026-05-06 11:39:29 -03:00
• f39bcf1e30 [KV Offload] Return None from lookup() for in-flight blocks (#41795) Ronen Schaffer 2026-05-06 17:31:21 +03:00
• 6467213a9f fix(openai): tolerate empty content in forced tool choice (#40148) 3 people 2026-05-06 22:16:03 +08:00
• df8e63f4ed nixl refactor: new transfer design (#40731) zhanqiuhuandNickLucche 2026-05-06 09:16:25 -04:00
• 242afc6bf4 [MM][Gemma4] Respect max_soft_tokens in encoder budget (#41799) 3 people 2026-05-06 21:54:42 +09:00
• 5d0fd87038 [CPU][RISC-V] Auto-bind OMP threads and harden nobind path (#40569) lyd1992 2026-05-06 19:38:08 +08:00
• d8deb5b7ad Fix some legacy checkpoints with deprecated rope_type values (#41734) Harry Mellor 2026-05-06 12:13:12 +01:00
• 2e777d21a8 [Bugfix][Rocm]Aiter MoE re-uses existing tensor addresses after weight update. (#40390) Yuankai Chenandgemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> 2026-05-06 03:32:26 -07:00
• e43a791284 [Bugfix][CI] Fix Disaggregated test area path (#41794) Nicolò Lucchesi 2026-05-06 11:41:24 +02:00
• 66d1cc0c77 fix(rocm): remove workaround causing invalid argument on Qwen3.5 with TP=2 (#40686) 3 people 2026-05-06 16:38:32 +08:00
• 1c58876618 [XPU] Disable CUDA graph memory estimate on XPU platform (#41344) Chaojun ZhangandKunshang Ji 2026-05-06 16:38:18 +08:00
• 51c1ee9b7c [Examples] Resettle Disaggregated examples. (#40759) wang.yuqiandgemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> 2026-05-06 16:20:38 +08:00
• 213f10bfdd [Bugfix] Fix codegen for unqualified names (#40726) Lucas Kabela 2026-05-06 01:11:37 -07:00
• e87e09a50a [Feat] dnnl build for AVX2 W8A8 Int8 (#41318) Tianmu LiandLi, Jiang <jiang1.li@intel.com> 2026-05-06 00:28:02 -07:00
• 809b98e5b7 [CPU] Add FP8 W8A16 linear support (#41186) Yuwen Zhou 2026-05-06 15:05:27 +08:00
• b53c507bc9 [Bugfix] Skip PP sampled-token receive on last rank during async scheduling (#40749) wi-adam 2026-05-05 22:31:14 -07:00
• 2d7d6cf765 [Spec Decode] Allow multimodal models with a warning (#41752) LiandLi Zhang 2026-05-05 22:16:44 -07:00
• 91740ca5ea [ROCm][CI] Refine gating tests (#37243) Andreas Karatzas 2026-05-06 00:05:20 -05:00
• e47c98ef7a [Fix] Add missing stubs from cpu fp8 attention changes (#41387) Tianmu LiandLi, Jiang <jiang1.li@intel.com> 2026-05-05 21:16:27 -07:00
• aee190ac37 [Build] Fall back to system libgomp when torch has no vendored copy (#40575) lyd1992andClaude 2026-05-06 11:42:03 +08:00
• 16e336491e [Mistral Tokenizer] allow more leniency in apply_chat_template (#41658) Julien DenizeandRoger Wang 2026-05-06 04:56:15 +02:00
• c7aa186d67 [Frontend] Supports resubmitting output items with missing fields in Responses API (#41355) Chauncey 2026-05-06 10:21:33 +08:00
• f653761252 [CI] Route part of B200 jobs to b200-k8s (#41453) Kevin H. LuuandOpenAI Codex 2026-05-05 19:00:30 -07:00
• 4a8ae26e53 [ROCm][CI] Use vLLM generation defaults for DeepSeek prefetch-offload eval (#41575) Andreas Karatzas 2026-05-05 20:08:12 -05:00
• 1333864408 [CI] Automate Docker Hub release image publishing (#40415) Kevin H. Luu 2026-05-05 17:15:23 -07:00
• 01b9b5af67 [Attention] Minor refactor: layer takes ownership of the MLA prefill backend (#41744) Matthew Bonanni 2026-05-05 19:22:41 -04:00
• 8c57b6e7bc Bump model-hosting-container-standards to >= 0.1.14 (#39755) 3 people 2026-05-05 16:09:57 -07:00
• 79246b5ea6 [Spec Decode] Fix max_model_len logging in speculative config for draft model (#41571) Lanze Liu 2026-05-05 14:56:06 -07:00
• 48954de237 Fix DeepGEMM ep_scatter output address overflow (#39213) Matej SirovatkaandTyler Michael Smith 2026-05-05 11:56:56 -07:00
• c6235ed180 [BUGFIX] Support streamed_args_for_tool in MistralToolParser (#41730) Julien Denize 2026-05-05 19:48:53 +02:00
• 628c436301 [New Model][ROCm] Add AMD support for DeepSeek V4 (#40871) 4 people 2026-05-05 23:55:37 +08:00
• 2228fe6868 [Attention] Move FA3→FA4 upgrade into get_flash_attn_version() (#40815) Canlin Guo 2026-05-05 23:43:03 +08:00
• 84bd8a3c1e Remove unnecessary runtime asserts from linear layers (#41729) Harry Mellor 2026-05-05 15:42:56 +01:00
• b786ec8e74 [Bugfix] Suggest upgrading Transformers for tokenizer class errors (#38099) Lidang Jiang 2026-05-05 22:10:45 +08:00
• 20dcd984f9 [Bugfix] Fix RuntimeError: Already borrowed by adding thread-safe Hugging Face fast-tokenizer wrappers (#41181) yzong-rhandwang.yuqi 2026-05-05 10:04:01 -04:00
• 6fca518157 [BugFix][MyPy]: Module has no attribute "sched_getaffinity" [attr-defined] (#41465) Martin Hickey 2026-05-05 14:20:37 +01:00
• 98661fe012 [Bugfix][KVConnector] Support DCP/PCP in OffloadingConnector (#41549) Itay EtelisandItay Etelis 2026-05-05 14:54:29 +03:00
• b0765bee17 Fix DeepSeek-OCR for Transformers v4 (#41460) Harry Mellor 2026-05-05 12:11:21 +01:00
• 0a201b60cf [Model] support Qianfan-OCR model (#40136) 3 people 2026-05-05 18:51:25 +08:00
• 8b9ea2f881 [Feature] Add Triton kernel JIT compilation monitor for inference (#40137) Artem PerevedentsevandJiangyun Zhu 2026-05-05 13:08:57 +03:00
• 2ceea42958 [XPU] use xpu topk topp sample kernel (#39285) Kunshang Ji 2026-05-05 18:05:17 +08:00
• bee126165f [P/D][Mooncake] Add KVConnectorStats for transfer observability (#40414) Zhewen LiandZhewen Li 2026-05-05 02:17:38 -07:00
• 27cc676be3 [Model] Use AutoWeightsLoader for Plamo2 (#41699) BitToby 2026-05-04 22:56:24 -10:00
• 4845aee6b7 [Benchmark] Add --trust-remote-code flag to multi-turn benchmark (#41661) Dao007foreverandClaude 2026-05-05 01:00:37 -07:00
• 0c620d2e08 [Model] Use AutoWeightsLoader for CohereMoe (#41690) BitToby 2026-05-04 18:44:15 -10:00
• 6bb924bbf3 [Model] Fix Gemma4 MoE activation mismatch (#41574) Luciano MartinsandLuciano Martins 2026-05-05 01:34:11 -03:00
• eaec7be446 [BugFix] Preserve max_seq_len in ubatch metadata during CUDA graph capture (#40961) czhu-cohere 2026-05-04 21:29:34 -07:00
• f04fd1677b [Ray] Enable RayExecutorV2 by default (#41421) Jeffrey Wang 2026-05-04 21:27:34 -07:00
• 420b0a5c95 [Hardware][Power]Add Power VSX Attention Backend and fix l2 Cache Crash (#40451) 4 people 2026-05-05 09:21:09 +05:30
• 1e9500410a [ROCm][Quantization][2/N] Refactor quark_moe w4a8 w/ oracle (#39136) Bowen Bao 2026-05-04 19:50:38 -07:00
• 416f9cdede [Perf][2/n] Eliminate GPU<->CPU syncs in pooling code (#41433) Nick Hill 2026-05-04 19:43:25 -07:00
• 685bf811d6 [XPU] enable is_act_and_mul for xpu (#37481) Chendi.XueandKunshang Ji 2026-05-04 20:07:39 -05:00
• e1e4646b06 [Model Runner V2] Rebuild attn metadata between draft decode steps (#41162) Giancarlo Delfin 2026-05-04 17:44:55 -07:00
• 4f2af1a7c0 [Feature] TurboQuant: support hybrid models and uniform quantization (#39931) 4 people 2026-05-05 02:14:01 +02:00
• 577b9623e6 [Bug] Fix status update address for non-MOE model within external dp mode (#40839) Wentao Ye 2026-05-04 19:37:16 -04:00
• 1cb0838721 [ROCm][CI] Fix MLA prefill scale for DeepSeek GSM8K (#41569) Andreas Karatzas 2026-05-04 18:32:55 -05:00
• be5983b874 [Docs] Add non-causal support to attention backend docs (#41643) Matthew Bonanni 2026-05-04 16:35:15 -04:00
• 9c07342fdc [NVFP4][fix] Fix layer.weight -> w13 typo in NVFP4 MOE emulation kernel preparation (#41630) fxmarty-amd 2026-05-04 22:13:37 +02:00
• 844df54269 feat: update xgrammar==0.2.0 to use structural tags for strict tool calling + reasoning for more models (#40894) 4 people 2026-05-05 03:45:24 +08:00
• 422dd02598 [bugfix] Fix prompt logprobs on request eviction during chunked prefill (#41411) Joachim StudniaandClaude Opus 4.6 2026-05-04 11:46:00 -07:00
• 8c780943b4 Fix Nano Nemotron text-only weight loading (#41205) 4 people 2026-05-05 03:43:07 +09:00
• e724b0ea8d [ROCm] ROCm7.2.2 + profiler fix + AITER 0.1.12.post2 (#41386) Gregory ShtrasbergandRohan138 2026-05-04 13:07:19 -05:00
• 712ad0286c [Bugfix] KimiK2ReasoningParser: guard against buffered end-token in streaming (#41068) 4 people 2026-05-04 10:42:05 -07:00
• 321fa2d6d1 Limit gpu utils and lower max BS on test_transcription_api_correctness.py (#41649) Ekagra Ranjan 2026-05-04 13:30:02 -04:00
• 3e1ad4435f [Bug] Fix tests/compile/test_config.py AttributeError: 'NoneType' object has no attribute 'dtype' (#41288) Wentao Ye 2026-05-04 12:22:07 -04:00
• 8decbfa02c Test nemotron nano-v2 and nemotron nano-v3 separately, disable super-omni redundant tests (#41616) Netanel Haber 2026-05-04 16:31:37 +03:00
• 62ba7516e8 Revert "[Doc] Fix RTD build: pytorch.org/docs/stable/objects.inv returns 404" (#41618) Stefano Castagnetta 2026-05-04 13:47:42 +02:00
• 6f53753fc9 [Bugfix] Apply ruff-format to hyperclovax.py (#41620) Stefano Castagnetta 2026-05-04 12:37:16 +02:00
• 6ec9bbec38 [CI] Stabilize cpu offload compressed tensors test (#41102) Andreas Karatzas 2026-05-04 00:22:42 -05:00
• 01d4d1ad37 [ROCm][CI] Align spec decode logprob test prefill settings (#41335) Andreas Karatzas 2026-05-03 23:33:29 -05:00
• c103c02a1a [Transformers v5] Vendor HCXVisionConfig for compatibility (#38447) Fang HanandClaude Opus 4.7 2026-05-03 21:19:52 -07:00
• 67058ca326 [CI] Clean up remote servers on pytest parent exit (#41570) Andreas Karatzas 2026-05-03 22:11:22 -05:00
• 894a02500b [Bench] Forward --seed to CustomDataset and CustomMMDataset shuffle (#40788) Akim Tsvigun 2026-05-04 02:39:10 +02:00
• 66dfee7121 [Bugfix] Fix degenerate KV cache stride causing TMA cudaErrorIllegalInstruction (#40737) 4 people 2026-05-03 16:52:18 -07:00
• db9a84e0cd [Bugfix] Fix FP8 Bias Loading (#41424) Alex Brooks 2026-05-03 14:30:04 -06:00
• cb03fee32b [Bugfix][Ray] Fix RayExecutorV2 actor name collision with DP > 1 (#40398) tomeras91 2026-05-03 23:00:41 +03:00
• c51df43005 Disable flashinfer autotune temporarily due to correctness issues (#41524) Wei Zhao 2026-05-03 12:19:59 -04:00
• 54dc64d5d3 [Doc] Add Qwen3-30B-A3B-Thinking-2507-FP8 to batch invariance verified models (#41513) Taneem Ibrahim 2026-05-03 07:47:55 -05:00
• e6ff3e9c83 [MRV2] Add shutdown() method (#41297) Woosuk Kwon 2026-05-02 21:06:30 -07:00
• 08834cc3ce [Quantization] add humming mxfp4 moe backend (#41083) Jinzhen Lin 2026-05-03 09:36:03 +08:00
• 856ec4804a [DSv4] Tune default value of VLLM_MULTI_STREAM_GEMM_TOKEN_THRESHOLD (#41526) Roger WangandCopilot 2026-05-02 18:32:09 -07:00
• 1c607d7b2c [DSV4] Guard megamoe flag with Pure TP (#41522) Yongye Zhu 2026-05-02 19:41:40 -04:00
• 4f7309fcc0 [CI] Add ci-fetch-log.sh helper for Buildkite job logs (#41517) Michael GoinandClaude Opus 4.7 2026-05-02 18:23:59 -04:00
• 0a9362d6ab Revert "[Build] Make bundled DeepGEMM wheel portable across Python versions" (#41512) Michael Goin 2026-05-02 12:42:41 -04:00
• cfd2573f23 [Build] Switch CUDA 13.0 wheel builds to PyTorch manylinux_2_28 base (#41416) Michael GoinandClaude 2026-05-02 08:51:28 -04:00
• c3ad791e1a [Bugfix][Gemma 4] Clamp soft-token estimate to max_soft_tokens (#40796) 4 people 2026-05-02 13:34:59 +07:00
• 8586369f61 Refactor Step3Text loading to use AutoWeightsLoader (#41492) Matthew Santiago 2026-05-02 01:22:14 -05:00
• ae3b4deb8a [Doc] Add Codex usage example (#41358) Chauncey 2026-05-02 13:27:43 +08:00
• c293ccc58e [ROCm][Bugfix] Fix init-time bias dtype cast when gate.out_dtype is None (#41405) Rita Brugarolas 2026-05-01 21:13:15 -07:00
• d58c42e19c [vLLM IR] 2/N fused_add_rms_norm and maybe_inplace overload (#36823) Luka Govedič 2026-05-01 23:41:15 -04:00
• 3e49479c4b Limit concurrency on test_transcription_api_correctness.py (#41478) Ekagra Ranjan 2026-05-01 23:19:07 -04:00
• 964a4bc2a5 [MM][CG] Support ViT CG for Qwen2.5-VL (#40830) John Calderon 2026-05-01 23:10:14 -04:00
• c408fdd663 [Fix] Sync gemma4 chat template from hf (#39570) FredericOdermatt 2026-05-02 05:06:54 +02:00
• 5737770c6c Re-enable allreduce rms fusion for DP / PP (#41458) Andy Lo 2026-05-02 00:01:37 +01:00
• 0c99629ede [Build] Make bundled DeepGEMM wheel portable across Python versions (#41476) Michael Goin 2026-05-01 17:45:03 -04:00
• edd60ac93a [Bugfix] Fix persistent_topk inter-CTA init race on RadixRowState (#41444) Yongye Zhu 2026-05-01 17:42:52 -04:00
• bcf5cac9fb [DSV4] Add knob to enable pre-attn gemm (#41443) Yongye Zhu 2026-05-01 15:23:17 -04:00