Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• a1448b4b69 [Kernels] Split up fused_moe/layer.py, isolate more modular kernel code (#28064) bnellnm 2025-11-11 09:29:02 -05:00
• fa1970201d [Docs] Fix grammar in CPU installation guide (#28461) Maryam Tahhan 2025-11-11 14:01:11 +00:00
• 3380543b20 Add request timeout override for multi-turn benchmarks (#28386) Ido Segev 2025-11-11 15:41:18 +02:00
• afffd3cc8a [Model] Pass mm_features directly into get_mrope_input_positions (#28399) Cyrus Leung 2025-11-11 21:14:48 +08:00
• 7dbe6d81d6 Fix Fused MoE LoRA Triton kernel bug (#28450) Chaojun Zhang 2025-11-11 20:46:47 +08:00
• b30dfa03c5 [Attention] Refactor CUDA attention backend selection logic (#24794) Matthew BonanniandLuka Govedič 2025-11-11 06:40:44 -06:00
• 2e78150d24 [CI] Add mergify rules for nvidia label (#28417) Michael Goin 2025-11-11 05:28:28 -07:00
• d381eb967f Multi turn benchmark progress bar for synthetic conversation generation (#28394) Ido Segev 2025-11-11 13:06:04 +02:00
• 9973e6e04a [Model][Qwen3VL] Slighly speedup fast_pos_embed_interpolate (#28434) Lukas Geiger 2025-11-11 10:35:10 +00:00
• c7991269dd [BugFix] 'DeepseekV2Config' object has no attribute 'use_mla'` (#28387) Fanli Lin 2025-11-11 16:45:38 +08:00
• f0359fffa4 [Bugfix] fix qwen3-next crash (#28202) Jiangyun Zhu 2025-11-11 16:24:28 +08:00
• 798c7bebca [EPLB] Refactor balance_packing to use numpy and optimize GPU-CPU transfers in EPLB (#28369) Sage Moore 2025-11-11 00:19:51 -08:00
• 4fd4b743a2 [Bugfix] Fix max image size for PaddleOCR-VL (#28442) Roger Wang 2025-11-11 00:07:24 -08:00
• cc079763c5 [BugFix] Avoid calling KV connector layer APIs when metadata is unset (#28253) 3 people 2025-11-11 09:39:36 +02:00
• a7adbc6c6b [Doc] Sleep mode documentation (#28357) 3 people 2025-11-11 13:44:35 +07:00
• e605e8e323 [Bugfix] Fix Stream Sync for Shared Expert Overlap (#28430) Robert ShawandVadim Gimpelson 2025-11-11 00:59:08 -05:00
• bca74e32b7 [Frontend] Add sagemaker_standards dynamic lora adapter and stateful session management decorators to vLLM OpenAI API server (#27892) 3 people 2025-11-10 20:57:01 -08:00
• 8d706cca90 [Misc] FlattenLogprobs -> FlatLogprobs (#28335) Zhuohan Li 2025-11-10 19:41:23 -08:00
• 57201a6a4c Fix rotary embedding benchmark script (#28323) Xin Yang 2025-11-10 18:57:12 -08:00
• f2d9ad0620 Only register rocm_aiter_ops if aiter is found (#28428) Michael Goin 2025-11-10 19:53:24 -07:00
• de540c0354 [Feature] Add env var VLLM_MOE_USE_DEEP_GEMM (#28422) Wentao Ye 2025-11-10 21:29:48 -05:00
• 39029d5192 [CI/Test Fix] Fix CP tests on Blackwell (#28404) Lucas Wilkinsonandgemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> 2025-11-10 20:36:29 -05:00
• 35d801f13f [Feature] Refactor batch invariant fp8 DeepGEMM (#27606) Wentao Ye 2025-11-10 19:08:40 -05:00
• 0bf29fadf5 [Test] Remove old non-varlen FA2 test (#28420) Matthew Bonanni 2025-11-10 17:57:41 -06:00
• a5a790eea6 [Bugfix] Ensure calculated KV scales are applied in attention. (#27232) Adrian Abeyta 2025-11-10 17:42:37 -06:00
• b30372cbd0 [Perf] Move gc.freeze logic from EngineCoreProc to EngineCore for better coverage (#27896) Jialin Ouyang 2025-11-10 15:34:18 -08:00
• d17ecc6b19 [PERF] Allreduce fusion. Support torch native matching. Tuning of the thresholds (#24248) 4 people 2025-11-11 00:33:11 +01:00
• 021143561f [ROCm] Add missing gemm_a8w8_blockscale import (#28378) Yong Hoon Shin 2025-11-10 13:13:36 -10:00
• 30700b1cd7 [CI] Fix Plugin Tests Tests (#28413) Robert Shaw 2025-11-10 17:36:11 -05:00
• 4b94ed8f92 [Frontend][2/n] remove empty content from _parse_tool_calls_from_content (#28331) Andrew XiaandAndrew Xia 2025-11-10 14:07:49 -08:00
• 6dec9f6109 [BugFix] Fix DeepGEMM over-allocating workspace (#28254) Lucas Wilkinson 2025-11-10 17:01:17 -05:00
• bf6a3d0ff5 [Misc] Add more scoping for improved trace (#28329) Wei Wei 2025-11-10 13:03:21 -08:00
• 40d33264c6 [Bugfix][EPLB] Disabled shared expert overlap when EPLB is enabled (#28377) 3 people 2025-11-10 12:39:19 -08:00
• 9c84ca8293 [FA/Chore] Bump FA version for FP8 two-level accumulation (#27889) Jonas M. KüblerandLucas Wilkinson 2025-11-10 21:06:04 +01:00
• 6d54336ae5 [Bugfix] Fix llguidance backend, rollback when EOS was encountered (#25905) Rémi DelacourtandRussell Bryant 2025-11-10 20:53:32 +01:00
• 34553b9d27 [Performance] Support FP8 flashinfer TRTLLM MOE on Qwen3 and Qwen-3next (#27492) jiahanc 2025-11-10 09:34:57 -08:00
• b039bfda8f [Bugfix] Fix persistent_masked_m_silu_mul_quant tests (#28366) Varun Sundar RabindranathandVarun Sundar Rabindranath 2025-11-10 12:21:52 -05:00
• d0e186c16f [V0 Deprecation] Remove unused context_len and seq_len from M-RoPE (#28395) Cyrus Leung 2025-11-11 00:30:06 +08:00
• f080a83511 [RFC][ROCm][AITER] Keep all AITER kernels in _aiter_ops class like _custom_ops and _ipex_ops (#24490) vllmellmandLuka Govedič 2025-11-10 17:20:53 +01:00
• 40e2eeeb92 [Kernel] Optimization of the mm_k operator. (#28280) caozuobaandJee Jee Li 2025-11-11 00:03:46 +08:00
• b06b9470ca [Rocm][fused_moe][fp4] view weight to torch.float4_e2m1fn_x2 when running aiter fused moe for fp4 model (#27474) zejunchen-zejun 2025-11-10 23:38:56 +08:00
• 4673e465ff Add @tjtanaa to codeowner for ROCm and multi-modal (#28360) TJian 2025-11-10 05:39:17 -08:00
• 912744d066 [Fix] optimize visual token mask with caching and multi-token support (#28374) Ferreboandgemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> 2025-11-10 21:23:49 +08:00
• 15be507c86 [bugfix] fix siglip batch text output error (#28365) Yu Jiaqi 2025-11-10 21:21:15 +08:00
• 6f7de33bed [Metrics] Refactor LoRA state tracking (#26801) Mark McLoughlin 2025-11-10 08:34:36 +00:00
• a98cc35c34 Restore PlaMo2 unit test as pfnet/plamo-2-1b now supports transformers >=4.56 (#28019) Shinichi Hemmi 2025-11-10 15:50:02 +09:00
• e8697faf03 [V0 deprecation] Remove no longer used get_metadata_cls (#28370) Lucas Wilkinson 2025-11-10 01:32:09 -05:00
• 03fa4d3fb3 [Hardware][AMD][Model] Add Triton MoE tuning support and optimized configs for Qwen3 omni for MI308X (#28373) Xiake Sunandgemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> 2025-11-10 12:53:40 +08:00
• 6b2b9fd934 [CI] lora/test_mixtral.py : Add additional expected outputs due to flakiness (#28322) Varun Sundar RabindranathandVarun Sundar Rabindranath 2025-11-09 21:45:29 -05:00
• c5f685b3ae [ROCm][Platform] Add RX7900XTX device id in _ROCM_DEVICE_ID_NAME_MAP (#28279) JartX 2025-11-10 00:09:36 +01:00
• c4768dcf47 [Kernel] Fix fused_gdn_gating (#28343) Jiangyun Zhu 2025-11-10 05:26:35 +08:00
• a65a934ebe [CI/Build] Temporary fix to LM Eval Small Models (#28324) Zhewen Li 2025-11-09 13:08:38 -08:00
• 4a8d6bd168 Fix cu_num_generated_tokens slicing logic in LogprobsLists.slice() method (#28214) usberkeley 2025-11-10 03:11:46 +08:00
• 636efd10a5 [Core] Separate out attention metadata building logic from prepare inputs (#26764) Lucas Wilkinson 2025-11-09 13:51:43 -05:00
• 289eb6c537 [Core] Simplify async KV output aggregation (#28327) Nick Hill 2025-11-09 09:44:13 -08:00
• 19d91ece4b [CI] Fix flaky test_eagle_correctness test (#28364) Nicolò Lucchesi 2025-11-09 17:04:59 +01:00
• 7ae5a5fb11 [Misc] Add some comments in qwen3-next (#28267) Jiangyun Zhu 2025-11-09 15:59:24 +08:00
• de2b78305f [ROCm] Add env to enable/disable aiter triton gemm (#28321) Yong Hoon Shin 2025-11-08 20:27:00 -10:00
• e5e9067e61 [Misc] fix typo and add detailed log (#28178) Ning Xie 2025-11-09 13:33:46 +08:00
• 3a7d580343 fix: close issue 28338 by fixed python version (#28339) yihong 2025-11-09 13:07:26 +08:00
• 05f8d69077 [chore] Move some wikimedia images to S3 (#28351) Kevin H. Luu 2025-11-08 17:58:26 -08:00
• 404d7a9d14 [Performance][gpt-oss] Revert gpt-oss max cudagraph size to 1024 (#28345) Mohammad Miadh Angkad 2025-11-09 06:50:10 +08:00
• 171133f929 [Bugfix] Fix test fused quant layernorm tests (#27865) 3 people 2025-11-08 23:31:33 +01:00
• 32787d0644 Remove setuptools upper bound constraint (<80) (#28337) Cole Murray 2025-11-08 14:30:18 -08:00
• 975676d174 [Feat] Drop-in Torch CUDA Profiler (#27841) Benjamin Chislett 2025-11-08 17:07:37 -05:00
• 77d702a22b Enhance run_cluster.sh for multi-NIC support (#28328) Ev Lacey 2025-11-08 14:04:16 -08:00
• 2108a571d7 [DCP] Support dcp kv_cache interleave size > 1 (#26696) zhangsicheng5andQiuChunshuo 2025-11-09 03:45:27 +08:00
• 47604137a2 [Bugfix] Spec decode + structured output + spec model max len edge case (#28298) Andy Lo 2025-11-08 19:44:25 +00:00
• 26990d25dc [Bugfix] Update device name for H200 detection (#28349) Robert Shaw 2025-11-08 14:01:11 -05:00
• d9ab1ad9d1 reasoning_content -> reasoning (#27752) Harry Mellor 2025-11-08 04:15:08 -08:00
• 608bb14462 [Attention] Remove max cudagraph size limit of 992 (#27840) 22quinn 2025-11-07 22:33:27 -08:00
• 4a36681f85 [flashinfer][fix] do not check nvcc availability when using pre-downloaded cubins (#27990) Xiaozhu MengandLu Fang 2025-11-07 22:25:21 -08:00
• d15afc1fd0 Refactor CPU/GPU extension targets for CMake build (#28026) Abolfazl Shahbazi 2025-11-07 22:17:35 -08:00
• 934a9c3b79 [Model] Consolidate Deepseek-MoE implementation with DeepSeek-v2 (#28101) Isotr0pyandKunshang Ji 2025-11-08 13:01:27 +08:00
• 70af44fd10 [bugfix] support eagle with lora cudagraph specialization (#28318) gnovack 2025-11-07 19:25:45 -08:00
• 781f5ebf52 Bump arctic-inference requirement (#28174) 3 people 2025-11-07 18:31:18 -08:00
• 0852527647 [Perf][DeepSeek] Add sigmoid+bias fusion to fused_grouped_topk from TRTLLM (#28124) 3 people 2025-11-08 10:20:55 +08:00
• 61d25dc44b Update gpu.rocm.inc.md to add support for AMD Ryzen AI MAX / AI 300 Series (gfx1151, gfx1150) (#28308) Hamid Mukhtar 2025-11-07 21:09:21 -05:00
• d0c7792004 [Bugfix][LoRA][Spec Decode] Support LoRA with speculative decoding (#21068) 5 people 2025-11-07 20:58:22 -05:00
• b158df2813 remove resolve_op_overloads and use splitting_ops directly (#28081) Boyuan Feng 2025-11-07 17:13:13 -08:00
• 1aaecda078 [XPU] Enable Expert parallel for MoE models (#28263) Kunshang Ji 2025-11-08 08:33:11 +08:00
• 811df41ee9 Update Flashinfer from v0.4.1 to v0.5.2 (#27952) Harry Mellor 2025-11-07 16:24:42 -08:00
• 67a2da890e [PerfFix] Avoid separate thread for MP executor shm spin (take 2) (#28319) Nick Hill 2025-11-07 14:11:03 -08:00
• da786e339e [Core] Rework handling of async scheduling config (#28250) Nick Hill 2025-11-07 12:01:23 -08:00
• 18903216f5 [Bugfix] Fix and add tests for GptOss reasoning parser (#28000) Benjamin Chislett 2025-11-07 14:28:04 -05:00
• d0ceb38ae8 [Build] Fix release pipeline failing annotation (#28272) Simon Moandgemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> 2025-11-07 10:06:45 -08:00
• 155ad56d7b [doc] add guide about the provided PTX was compiled with an unsupported toolchain (#28305) youkaichao 2025-11-08 00:26:34 +08:00
• 5fb4137c99 [README] Add Arm CPUs to the list of supported targets (#28290) Fadi Arafeh 2025-11-07 15:41:47 +00:00
• 68a72a5cc1 Revert "[PerfFix] Avoid separate thread for MP executor shm spin (#28012)" (#28289) Nicolò Lucchesi 2025-11-07 16:07:01 +01:00
• 0f872b7977 [Log] update shm wait time msg (#28255) Boyuan Feng 2025-11-07 06:43:30 -08:00
• 4b1ff13221 [Feature] Default ignore_eos True for random dataset (#28227) Wentao Ye 2025-11-07 07:35:33 -05:00
• e0d6b4a867 [CLI] add --max-tokens to vllm complete (#28109) Iceber Gu 2025-11-07 20:21:40 +08:00
• 72b1c2ae2c [Bugfix] Use latency MOE backend as default for Flashinfer and other misc fixes (#27439) Pavani Majety 2025-11-07 04:18:39 -08:00
• e0919f331d [Core][MM] Add mechanism to configure multimodal fields which should stay on CPU (#28168) Lukas Geiger 2025-11-07 12:14:29 +00:00
• 8e19d470af [fix] Revert "fixing mm placeholder replacement issue with gemma3" (#28285) Kevin H. Luu 2025-11-07 04:09:09 -08:00
• 1958bda9b4 [Misc][Model][Refactor] Pass the prefix into Linear layers (#28259) Mengqing Cao 2025-11-07 19:38:38 +08:00
• 7bdb42b2f2 [CPU]Avoid repeated random sample compile (#28260) Zhang Xiangze 2025-11-07 19:03:57 +08:00
• 315068eb4a [FixBug]Aeala/ShareGPT_Vicuna_unfiltered marked as multimodal benchmark (#28265) 汪志鹏 2025-11-07 17:35:22 +08:00
• ccd98b59c1 [Perf] Introduce FlattenLogprobs to store logprobs results to reduce GC overhead (#28171) Jialin Ouyang 2025-11-07 00:27:12 -08:00
• 21b82f4ea2 [Kernel] LoRA triton kernels support PDL (#27402) Jee Jee Li 2025-11-07 16:05:48 +08:00