Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• e806178d2a [BugFix][VL] Fix FA selection on Qwen2.5-VL (#27790) Zhewen LiandRoger Wang 2025-10-30 00:54:44 -07:00
• 5be1bed790 [CI/Build]Add eval config for Qwen3-235B-A22B-Instruct-2507-FP8 (#27113) Huamin Li 2025-10-30 00:50:56 -07:00
• 31b55ffc62 use stringData in secret yaml to store huggingface token (#25685) yitingdc 2025-10-30 15:47:36 +08:00
• ded8ada86a Add more dims for batch invariant shims (#27489) Bram Wastiandgemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> 2025-10-30 01:28:45 -04:00
• 8bff831f0a [Benchmark] Cleanup deprecated nightly benchmark and adjust the docstring for performance benchmark (#25786) Kuntai Du 2025-10-29 21:43:37 -07:00
• b5d70751d8 [BugFix] Reordering extend logic fix (#27739) Lucas Wilkinson 2025-10-30 12:39:34 +08:00
• b8c48c5d72 kernels/moe test pruning (#27053) Fardin HoqueandWentao Ye 2025-10-29 21:10:34 -07:00
• 17d055f527 [Feat] Adds runai distributed streamer (#27230) 3 people 2025-10-30 04:09:10 +00:00
• 2ce5c5d3d6 [BugFix] Handle unscheduled requests properly when async scheduling (#27756) Nick Hill 2025-10-29 21:04:25 -07:00
• b5bae42f91 [XPU] Update latest IPEX 2.8 release (#27735) Kunshang Ji 2025-10-30 11:17:13 +08:00
• d7fb10c574 [Bugfix] mamba-block-size is set for vision language model (#27773) Chen Zhang 2025-10-29 19:39:57 -07:00
• b798e39f93 [XPU][bugfix] fix rope for llama4 and deepseek (#25145) Yan Ma 2025-10-30 09:43:13 +08:00
• 48eb8eba58 [Temp fix] Disable torch.compile for Qwen2.5 VL's VisionBlock temporarily. (#27760) Chenheli HuaandRoger Wang 2025-10-29 16:17:48 -07:00
• b5d90f7400 [Bug] Fix DBO IMA issue for DeepEPHT (#27666) Wentao Ye 2025-10-29 16:28:27 -04:00
• d4aa144343 [BugFix] Fix handling of resumed reqs in SharedStorageConnector (#27719) Nick Hill 2025-10-29 13:16:52 -07:00
• fcb1d570bb [Bug] Fix DeepEP low latency assert self.batched_router_logits.size(-1) == full_router_logits.size(-1) Bug (#27682) Wentao Ye 2025-10-29 14:50:39 -04:00
• accb8fab07 [KVConnector] Add metrics to Prometheus-Grafana dashboard (#26811) Nicolò LucchesiandMark McLoughlin 2025-10-29 19:44:49 +01:00
• 5b0448104f [Bug] Raise error explicitly if using incompatible backend (#27424) Wentao Ye 2025-10-29 13:29:20 -04:00
• f7a6682872 [CI/Build] Test torchrun with 8 cards (#27548) 22quinn 2025-10-29 10:26:06 -07:00
• a9fe0793f2 use_aot_compile should respect VLLM_DISABLE_COMPILE_CACHE (#27698) Boyuan Feng 2025-10-29 10:08:54 -07:00
• 7568a282b9 [FIXBUG] Qwen3VL hallucinations without Contiguous on Torch.SDPA (#27744) JartXandLukas Geiger 2025-10-29 17:55:35 +01:00
• 1da3309ace [Core] Exposing engine sleep & wake_up state as prometheus metrics (#24176) Braulio Dumba 2025-10-29 12:32:01 -04:00
• 5522fb274b [Chore] Optimize P2PNCCLEngine http_address (#27488) Wentao Ye 2025-10-29 12:05:09 -04:00
• 0f95a1c3f2 [CI] Fix flaky test_two_responses_with_same_prev_id test (#27745) Nicolò Lucchesi 2025-10-29 16:10:35 +01:00
• ded24e3e54 [ROCm][Platform] Add MI308X device id in _ROCM_DEVICE_ID_NAME_MAP (#27623) Xiake Sun 2025-10-29 22:44:03 +08:00
• d6704dd099 Fix MiniMax-M2 rmsnorm precision and remove useless code (#27627) Roger Youngandxuebi 2025-10-29 21:01:05 +08:00
• ecca3fee76 [Frontend] Add vllm bench sweep to CLI (#27639) Cyrus LeungandHarry Mellor 2025-10-29 20:59:48 +08:00
• 9a0d2f0d92 [CI/Build] Skip cpu offloading test on AMD (#27690) Zhewen Li 2025-10-29 05:55:51 -07:00
• ad3ec89532 [VLM] Add Qwen3-VL generation test (#25185) Isotr0pyandRoger Wang 2025-10-29 20:19:37 +08:00
• 3481e40743 [chore] Remove models weight on S3 logic (#27725) Kevin H. Luu 2025-10-29 03:29:49 -07:00
• 5e72216d17 Feature/video support in random mm dataset (#25963) Eugene KhvedchenyaandRoger Wang 2025-10-29 12:24:52 +02:00
• 1a33aacf82 [Misc] Raise error for missing video metadata in MultiModalDataParser (#27664) Isotr0pyandgemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> 2025-10-29 18:06:42 +08:00
• 7ba6aa8f56 [Fix] import get_kv_cache_torch_dtype error in LMCacheConnector integration (#27670) Yue Zhang 2025-10-29 18:03:54 +08:00
• ab2eb27b74 [Frontend] [gpt-oss] Mcp type bug (#27689) 4 people 2025-10-29 06:01:32 -04:00
• 3c7fefdeba [Frontend] [gpt-oss] Tool json call parsing error retry (#27675) 4 people 2025-10-29 05:42:44 -04:00
• 1891cf605a [Bugfix] Fix modular kernel tests (#27707) bnellnm 2025-10-29 04:14:33 -04:00
• 8df98c2161 [perf] Enable concurrent execution of "shared_experts" and "selected_experts" in qwen3-next (#27578) Jiangyun Zhu 2025-10-29 16:12:54 +08:00
• 4fb8771cc0 [CI/Build] Move pre-commit only scripts to tools/pre_commit (#27657) Cyrus Leung 2025-10-29 16:04:33 +08:00
• 413ef7a3b4 [Speculators] Move tests + fix integration (#27308) 3 people 2025-10-29 03:54:21 -04:00
• 8b62495076 [Bugfix] Fix non-contiguous tensor error in rocm_unquantized_gemm_impl (#27605) Zhewen Li 2025-10-29 00:00:15 -07:00
• 83fd49b1fc [CI/Build][Bugfix]Fix Quantized Models Test on AMD (#27712) Zhewen Li 2025-10-28 23:27:30 -07:00
• a4a4f0f617 [KV Connector] Update lmcache connector with latest compatibility (#27681) ShaotingandSamuel Shen 2025-10-28 22:38:37 -07:00
• 0d8161b075 [Model] Fix Qwen3VL and Qwen3Omni after torch.compile changes (#27705) Lukas GeigerandRoger Wang 2025-10-29 05:28:20 +00:00
• d2c33c397a [NIXL][XPU] update name of nixl wheel (#27631) liuzhenwei 2025-10-29 12:43:29 +08:00
• f6d5f5888c [Build] Revert triton_kernels requirements (#27659) Varun Sundar Rabindranath 2025-10-29 00:07:09 -04:00
• 9007bf57e6 Revert "Install pre-built xformers-0.0.32.post2 built with pt-2.9.0" (#27714) Simon Mo 2025-10-28 20:58:01 -07:00
• f257544709 Install pre-built xformers-0.0.32.post2 built with pt-2.9.0 (#27598) Huy DoandRoger Wang 2025-10-28 19:39:15 -07:00
• 0b51c9bd8b [Core] Early return in SlidingWindowManager.remove_skipped_blocks (#27673) Jialin Ouyang 2025-10-28 18:32:33 -07:00
• d3ab240f39 [Bug] Fix deepep low latency use nvlink by default (#27677) Wentao Ye 2025-10-28 19:53:12 -04:00
• 94666612a9 [Misc][qwen2_5_vl][torch.compile] Enable supports_torch_compile on generic nn.Module and demonstrate speedup on Qwen Vision model (#23207) Lucas Kabela 2025-10-28 15:36:43 -07:00
• 4fe5895361 [AsyncScheduling] Make async overlap work with logprobs (#27615) Nick Hill 2025-10-28 15:35:54 -07:00
• 111faf1118 [Core] Scheduler: Publish connector events after output (#25875) Or Ozeri 2025-10-28 23:01:33 +02:00
• 6afc28a9ba [Test] Batch Invariant: Unit test using parameterized backend (#27478) Wentao Ye 2025-10-28 16:51:35 -04:00
• 141e6a0505 [Misc] Make reorder batch also separate extends (#27367) Lucas Wilkinson 2025-10-29 01:55:10 +08:00
• 130aa8cbcf Add load pattern configuration guide to benchmarks (#26886) Matvei PashkovskiiandHarry Mellor 2025-10-28 19:49:15 +02:00
• e3d8186666 [compile] Add fallback path to AOT compile when serialization fails. (#27350) Zhengxu ChenandCyrus Leung 2025-10-28 12:54:26 -04:00
• f5710ef02a [Misc] Make LayerBlockType a Literal instead of Enum (#27658) Cyrus Leung 2025-10-29 00:23:35 +08:00
• a8c02fb5bf [Bugfix][CI] Fix v1 attention backend tests and add CI coverage (#26597) Mohammad Miadh AngkadandYe Qi 2025-10-28 23:42:05 +08:00
• 02af36df36 [Bugfix] Fix allocation & free logic of SingleWriterShmRingBuffer (#27117) 3 people 2025-10-28 23:01:24 +08:00
• e88bdd60d9 [FLA] Introduce Kimi Delta Attention(KDA) to VLLM (#27654) Zhiyuan Li 2025-10-28 22:56:28 +08:00
• 05e034f085 [nit]: Fix import for the lmcache integration (#27600) Samuel ShenandSamuel Shen 2025-10-28 07:40:55 -07:00
• 936643a868 [BugFix] Also consider RAY_EXPERIMENTAL_NOSET_* when storing compilation cache (#27294) ℍ𝕠𝕝𝕝𝕠𝕨 𝕄𝕒𝕟 2025-10-28 16:22:28 +02:00
• b186149e8e [Bugfix][Frontend] validate arg priority in frontend LLM class before add request (#27596) Junpu Fan 2025-10-28 07:02:43 -07:00
• 2abbd351ef [Core] Enable async scheduling for external_launcher mode (#27394) 22quinnandZhuohan Li 2025-10-28 06:52:47 -07:00
• 446912d1cb fix: allow HuggingFace standard chat template params via **kwargs (#27622) 3 people 2025-10-28 21:12:34 +08:00
• a00d6254e9 [compile] Disable dynamo guards check for AOT compilation. (#27288) Zhengxu ChenandCyrus Leung 2025-10-28 08:58:12 -04:00
• 05181cc57f [Hybrid] Add mamba_block_size to Engine Args (#27289) Asaf Joseph Gardin 2025-10-28 14:54:24 +02:00
• 259504e147 [compile] Add enable_prompt_embeds to compile hash. (#27285) Zhengxu ChenandCyrus Leung 2025-10-28 08:46:03 -04:00
• 0484b64248 [Bug] Fix shape issue for eplb expert weights (#27589) Wentao YeandCyrus Leung 2025-10-28 08:44:05 -04:00
• f58d9b6404 [Misc] Separate out utils.counter and move utils.Device to engine (#27588) Cyrus Leung 2025-10-28 20:20:46 +08:00
• 44b5ce956d [Bugfix] In LongRoPE, decide short vs long based on max_model_len (#27431) Matthew Bonanni 2025-10-28 08:00:56 -04:00
• 7a865f2325 [V0 Deprecation] Remove vestigial V0 logits_processors.py file (#27601) Nick Hill 2025-10-28 04:17:45 -07:00
• 2fa90bda27 Fix a robust parsing issue in KimiK2ToolParser that causes IndexError (#27565) wangln19andwangln19 2025-10-28 19:11:50 +08:00
• 0291fbf65c [CI/Build] Fix amd model executor test (#27612) Zhewen Li 2025-10-28 01:58:11 -07:00
• b46e4a06f1 [Core][Bookkeeping Optimization] Update against numpy view of is_token_ids tensor (#27618) Jialin Ouyang 2025-10-28 01:13:10 -07:00
• d34f5fe939 [Bugfix][CPU] Fallback oneDNN linear to torch linear to fix half gemm support on legecy platforms (#27526) Li, JiangandIsotr0py 2025-10-28 14:25:44 +08:00
• bdb01a38fe [Hardware][AMD][Model] Triton MoE tuning configs for GLM-4.6 for MI300X (#27323) Eric Yue 2025-10-28 13:58:06 +08:00
• 5b3c35a68e [ROCm] [Doc] Update ROCm installation docs (#27327) vllmellm 2025-10-28 13:00:50 +08:00
• 61fbfe5274 [Bugfix] fixed inconsistent finish_reason handling between V0 and V1 engines (#27555) Chauncey 2025-10-28 10:18:08 +08:00
• 255e34ca50 [Stability fix] turn off HMA allocator when connector is set (#27592) Kuntai Du 2025-10-27 18:32:23 -07:00
• a8d2e326ec [Bugfix][CI] Fix config resolving logic with remote models (#27610) Roger Wang 2025-10-27 17:48:32 -07:00
• 53a56e658b [gpt-oss][2/N] Support input_messages in responsesRequest (#26962) Andrew XiaandAndrew Xia 2025-10-27 16:15:49 -07:00
• 69f064062b Code quality improvements: version update, type annotation enhancement, and enum usage simplification (#27581) usberkeley 2025-10-28 01:50:22 +08:00
• 921e78f4bb [ROCm] Update AITER branch for ROCm base docker (#27586) Micah Williamson 2025-10-27 12:22:33 -05:00
• 6ebffafbb6 [Misc] Clean up more utils (#27567) Cyrus Leung 2025-10-27 23:30:38 +08:00
• 3b96f85c36 [Chore]: Stream tokens vs characters in tool call parser tests (#26513) Ben Browning 2025-10-27 11:06:25 -04:00
• 23ad820553 fixing mm placeholder replacement issue with gemma3 (#27538) tingtinggithub 2025-10-27 07:34:01 -07:00
• 5d3be3ba4c [Bugfix][LoRA][FusedMoE] Select MxFP4 Backend based on LoRA Enablement (#27487) Varun Sundar RabindranathandVarun Sundar Rabindranath 2025-10-27 10:32:50 -04:00
• 4f882be4a0 [Model] Siglip2 Model Support (#27566) Yu Jiaqi 2025-10-27 21:57:37 +08:00
• 9273754222 [Hybrid] Added supports_mamba_prefix_caching Protocol (#27339) Asaf Joseph Gardin 2025-10-27 15:05:20 +02:00
• f4e8154076 [Kernel] Enable moe LoRA kernel support FP16 (#27468) Jee Jee Li 2025-10-27 19:48:37 +08:00
• a663f6ae64 [cpu][perf] Fix low CPU utilization with VLLM_CPU_OMP_THREADS_BIND on AArch64 (#27415) Fadi Arafeh 2025-10-27 11:14:55 +00:00
• a4fc21895e [Bugfix] Fixed when return_token_ids=False, the first event still contains prompt_token_ids. (#27561) Chauncey 2025-10-27 19:06:43 +08:00
• a3e8611da5 [Bugfix] Limit the default value of max_model_len when it is not specified by users (#27556) Shanshan Shen 2025-10-27 18:16:20 +08:00
• 7c2bdb83dc [Misc] Clean up utils (#27552) Cyrus Leung 2025-10-27 17:05:40 +08:00
• 9932ed6a83 [Kernel] Adding split_K implementation for fused_moe_lora (#27291) 3 people 2025-10-27 02:05:24 -07:00
• 2d631d28c6 [Doc] Slight improvement to M2 and beyond (#27554) Jee Jee LiandRoger Wang 2025-10-27 17:02:10 +08:00
• b368382964 [Model] Deprecate merge_by_field_config=False (#27551) Cyrus Leung 2025-10-27 16:43:00 +08:00
• a806c14cc7 [Performance][LoRA] add context varying params to 'do_not_specialize' in fused moe lora (#27445) gnovack 2025-10-26 23:31:55 -07:00
• 181bf5bbde [Docs] reemove the incorrect enable_reasoning parameter (#27550) yyzxw 2025-10-27 14:17:19 +08:00