Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• 89876b0c54 [GLM5] Implement op fusion for GLM5/DSV3.2 (#46876) Woosuk Kwon 2026-06-28 08:17:39 -07:00
• 5c91039c41 [GLM5.2 Perf] Replace MOE all-reduce with reduce-scatter, 3.1%~3.2 E2E Throughput improvement (#46635) Wentao Ye 2026-06-28 10:55:54 -04:00
• 5ecae3266c [ROCm][Perf][MLA] Add AITER FlashAttention MLA prefill backend (ROCM_AITER_FA) (#45033) xaguilar-amdandTJian 2026-06-28 16:52:00 +02:00
• 6eb63a1da6 [Bugfix][DSv3.2] Skip indexer weights for index-cache-skipped layers (#46600) frida-anderssonandAndreas Karatzas 2026-06-28 10:37:44 +02:00
• 09841ae705 [Render][Speculator] Add return_loss_mask to render endpoint for training data generation (#46846) RanranandBenjamin Chislett 2026-06-28 02:07:33 -05:00
• a2a92cbbaa [Hardware][AMD][CI] Tweak mirrored tests; improve CI base dependency change detection (#46930) Matt 2026-06-28 02:07:14 -05:00
• 35e6c86caa [Bugfix][MM][CG] Enable dual-path ViT CUDA graph for Step3-VL (#46034) Shanshan ShenandIsotr0py 2026-06-28 15:06:43 +08:00
• c7ca0bccae [ROCm][Perf] Add Fused Shared Expert (FSE) support for GLM-4.5/6/7 (#44313) 3 people 2026-06-28 10:04:08 +03:00
• c6741b2ad4 [Model] Support Unlimited OCR (#46564) 3 people 2026-06-28 14:09:18 +08:00
• a65f93fb2e [ROCm][CI] Add ci_base metadata for external cache orchestration (#46886) Andreas KaratzasandCodex 2026-06-27 23:51:19 -05:00
• 11a12305c0 [Model Runner V2][Spec Decode] Handle tuple hidden states from MTP draft models (#46786) Chauncey 2026-06-28 09:38:07 +08:00
• 798185d438 [KV-Offloading] Fix tensors_per_block stride (#46888) Varun Sundar RabindranathandVarun Sundar Rabindranath 2026-06-27 21:01:45 -04:00
• 9036c89ee4 [Hardware][AMD][CI] Patch Whisper multi LoRA test to use TRITON_ATTN for now (#46928) Matt 2026-06-27 17:30:49 -05:00
• b6caeb5a09 [Model Runner V2][Spec Decode] Use fp32 uniform threshold for acceptance (#46878) Giancarlo Delfin 2026-06-27 16:09:25 -05:00
• 8bf064f8d3 Fixed chunked embedding aggregation with request-id metadata (#46782) Taneem Ibrahim 2026-06-27 16:57:47 -04:00
• ea2ead1db3 [Misc] Fix incorrect layer type annotation in Fp8LinearMethod (#46818) jj shaoandshaojinjie.sjj 2026-06-28 04:23:59 +08:00
• 56aa067bf0 [CI Bug] Fix h100 AssertionError: Cold-start child failed (#46927) Wentao Ye 2026-06-27 16:17:33 -04:00
• 35e3850fa9 [Bugfix][Test] Fix test_flashinfer_cutlass_mxfp4_fused_moe on sm90 (stale weight/scale interleave) (#46915) xiaolinchen 2026-06-28 02:30:10 +08:00
• 51a99565c3 [ROCm][Perf] Fused shared expert for Minimax M3 (#46474) 3 people 2026-06-27 05:34:17 -07:00
• 867fd5e8ed [ROCm][Perf] Use flydsl moe with Minimax-M3 mxfp8 weights on gfx950 and implemented moe-backend selection (#46184) 5 people 2026-06-27 06:22:57 -04:00
• 9fd00ee006 [ROCm][CI] Move remaining mi250_2 tests out of the MI250 queue (#46905) Andreas KaratzasandCodex 2026-06-27 04:08:54 -05:00
• 091d13976c [ROCm][CI] Add TRITON_ATTN score absolute tolerance floor (#46891) peizhang56andClaude 2026-06-26 23:35:50 -07:00
• b588f66dc2 [GLM5.2 Perf] fused_indexer_q_rope_quant triton kernel, 1.9% ~ 3.3% E2E Throughput improvement. (#46862) Wentao Ye 2026-06-27 01:16:20 -04:00
• 455f25aa13 [CLI] Add flag to print TTFT and TPS in vllm chat (#46775) Benjamin Chislett 2026-06-27 01:15:10 -04:00
• d706dec904 fix: Correct reasoning-end detection for prompt history (#44551) 6 people 2026-06-27 06:15:06 +01:00
• 68ee8300a0 [ROCm][CI]Fix test_concat_and_cache_mla_rope_fused on ROCm (#46409) Divakar Verma 2026-06-27 00:38:13 -04:00
• ddd3855a28 [MoE Backend] add HPC-Ops MoE backend (#45924) 3 people 2026-06-27 11:18:07 +08:00
• 00e045b7c7 [ROCm][CI TG] refactor and fix deepep_moe test group (#46758) Divakar Verma 2026-06-26 22:45:23 -04:00
• 17a71d8702 [ROCm][CI] Relax fused layernorm quant test tolerances for one-ULP outliers (#46658) Divakar Verma 2026-06-26 22:44:29 -04:00
• 2e058851d3 fix(docker): eliminate race conditions in shared buildkit cache mounts (#44984) weizhoublue 2026-06-27 10:43:17 +08:00
• 1a92dfcce4 [Build] Show error message when using ROCm with LTO and different compilers (#35232) Dāvis 2026-06-27 05:43:00 +03:00
• d0f800811b [Build] Update vllm to point to vllm-project/flash-attention commit that builds FA3 with torch stable API. (#46644) Chris Leonard 2026-06-26 22:42:46 -04:00
• c6dd32a810 [ModelRunner V2] Support realtime embeddings (#46762) Nick Hill 2026-06-26 19:42:27 -07:00
• af16446bf3 Vram semaphore infra (#44465) Brandon PelfreyandRoger Wang 2026-06-26 17:32:51 -07:00
• 3f67477497 [CI] Don't try and download files that we already know don't exist (#46854) Harry Mellor 2026-06-27 00:56:39 +01:00
• 1d41009e81 [ModelRunner V2] Fix cross-attention block table sizing (#46753) Nick Hill 2026-06-26 16:34:21 -07:00
• b94f212e37 [ModelRunner V2] Deduplicate ModelState init logic (#46776) Nick Hill 2026-06-26 16:32:45 -07:00
• d8eb734d94 Fix Transformers backend FP8 MoE and remove some boilerplate (#46820) Harry Mellor 2026-06-27 00:16:05 +01:00
• 2ff76a5e85 [ROCm][Bugfix] Pass num_kv_splits to aiter mla_reduce_v1 (#46760) Rohan PotdarandClaude Opus 4.8 2026-06-26 16:58:40 -05:00
• 75fdcc82a5 [CI] Add @ivanium to CODEOWNERS for KV-cache/offload areas (#46873) Yifan Qiao 2026-06-26 14:48:53 -07:00
• 77f8796d16 [Frontend][Gpt-oss] Use process_eos() to flush Harmony Parser outputs. (#46437) yzong-rh 2026-06-26 17:18:47 -04:00
• c40d307731 [Core] Remove FlashAttention block size restriction for hybrid models (#36701) Thomas ParnellandClaude Opus 4.6 2026-06-26 23:16:39 +02:00
• 65e655d295 [GLM-5] Add DSV3.2/GLM5 to vllm/models/ (#46808) Woosuk Kwon 2026-06-26 14:09:05 -07:00
• 6e2fb02fe5 [ROCm][CI] Fix rlhf_nccl.py on ROCm (#46851) Charlie Fu 2026-06-26 15:41:49 -05:00
• 274325dd43 [ROCm][CI] Remove V1 Sample + Logits from mi250 Queue (#46867) Micah Williamson 2026-06-26 15:38:38 -05:00
• 95e6442a6b [Hardware][AMD][CI] Fix Kernels Quantization test timeout (#46859) Matt 2026-06-26 15:19:16 -05:00
• 701a23d99f [Bugfix][Model] Support tensor parallelism for DiffusionGemma (#45719) (#46177) 3 people 2026-06-26 16:05:04 -04:00
• dccb412e2c [Bugfix][Parser] Pass token IDs to parser.parse() in Responses API and batch serving (#46843) Ben Browning 2026-06-26 15:29:52 -04:00
• c6554f321c [CPU] Fix macOS/Apple Silicon hang by enabling OpenMP in the build (#46769) Michael GoinandClaude Opus 4.8 2026-06-26 12:32:21 -06:00
• 3d3b96488f Migrate Voxtral to mistral-common 1.11.5 audio API (#46705) Julien Denize 2026-06-26 20:06:31 +02:00
• 658b54efe4 [ModelRunner V2] Update scheduler tests to cover MRV2 paths (#46771) Nick Hill 2026-06-26 09:36:31 -07:00
• abc71548ef [CI/Build][CPU] Add test image cache clean-up (#46831) Li, Jiang 2026-06-26 23:28:49 +08:00
• 4e07ca2c92 [Core] Add VLLM_GPU_SYNC_CHECK env var (#44800) Nick Hill 2026-06-26 08:24:33 -07:00
• e71bc6da85 [Rust Frontend] Use oss-harmony for Harmony output processing (#46799) Bugen Zhao 2026-06-26 23:24:13 +08:00
• 37ce34922f [CI] Fix failing CUDA graph capture in Triton MOE (#46735) fxmarty-amd 2026-06-26 16:21:20 +02:00
• c2507fb293 [ROCm] [MoE] [Perf] Shared-expert fusion for bias-routed MoE; enable on MiniMax-M3 mxfp8 model (#46545) Hongxia YangandClaude Opus 4.8 2026-06-26 10:05:20 -04:00
• 8921c4be88 [ROCm] [Performance] Optimize aiter moe for DeepSeekV4 (#46122) TJian 2026-06-26 21:43:27 +08:00
• 8e394244a5 [ROCm]Enable AITER MoE backend for MiniMax-M3-MXFP4 (#46419) qli88andTJian 2026-06-26 08:35:35 -05:00
• 302954e5f6 [ROCm] [CI] fix transcription flakiness AMD: Entrypoints Integration (API Server OpenAI - Part 1) (mi325_1) (#46823) TJian 2026-06-26 21:33:35 +08:00
• 950ee4c2e4 [API] Add token offsets to render endpoints (/v1/.../render) (#44226) Hyunkyun Moon 2026-06-26 21:02:52 +09:00
• d980a3cc6e [ROCm] Fix AITER_UNIFIED_ATTN Dispatching After AITER Bump (#46780) Micah WilliamsonandRohan138 2026-06-26 04:09:56 -05:00
• bf292b5f6b [Docs] Remove BambaForCausalLM from supported hybrid models list (#46071) AgenticSparkandCopilot 2026-06-26 01:02:50 -07:00
• 5e3dad04b1 [Misc] Move the legacy api_server.py to the examples directory. (#46783) wang.yuqi 2026-06-26 15:43:29 +08:00
• 63e161f296 [Bugfix][Tool Parser] PoolsideV1: fix string whitespace and required named tool choice (#46486) Joe Rowell 2026-06-26 08:05:16 +02:00
• c7645bce04 Remove grok model arch from vllm (#46706) Tiezhen WANG 2026-06-26 14:02:10 +08:00
• 35a49fcfc2 [CI][Bugfix] Spawn engine in mm cache sleep test to fix ROCm HIP error (#46749) peizhang56andClaude 2026-06-25 22:38:26 -07:00
• 915e99ec67 [ROCm][Bugfix] Fix HIP fork re-init in multimodal offline examples (#46741) peizhang56 2026-06-25 22:37:47 -07:00
• 5b33041746 [ModelRunner V2] Fix whisper test (#46773) Nick Hill 2026-06-25 22:10:36 -07:00
• 1a4984520e [Hardware][AMD][CI] Fix AMD CI image build (#46792) Matt 2026-06-26 00:05:12 -05:00
• e312c5cb25 [Rust Frontend] Make Granite4 string argument scanning incremental (#46507) Reid 2026-06-26 11:54:03 +08:00
• 1502cf6274 Fix relative allowed local media paths (#45263) Matti4 2026-06-26 05:45:20 +02:00
• d350fa8ddd [Bugfix][Rust Frontend] Reject min_tokens above max_tokens (#46733) ReidandBugen Zhao 2026-06-26 11:41:33 +08:00
• dbc49b6b99 [CI][NIXL] Fix NIXL EP import canary for the nixl 1.3.0 wheel and pin nixl==1.3.0 (#45166) ovidiusmandWentao Ye 2026-06-26 04:33:42 +02:00
• 552a9dbe59 [NVFP4][Emulation] Fuse NVFP4 weight dequantization with compute in triton kernel for w13/w2 MOE MLP linears (#44667) fxmarty-amd 2026-06-26 04:33:00 +02:00
• 02a1f23711 [DFlash] Fuse precompute kv per-layer rmsnorms (#46761) Giancarlo Delfinandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-06-25 21:32:07 -05:00
• 652d962bc9 [Model Runner V2][Spec Decode] Reduce TP communication for draft token generation (#46448) yihengandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-06-26 10:30:07 +08:00
• 5314665bad [Model Runner V2][DFlash] Enable dflash attention backend selection (#46770) Giancarlo Delfin 2026-06-25 21:29:25 -05:00
• 3daea7ceb9 [Bugfix][MRV2] Forward seq_lens_cpu_upper_bound for mamba hybrid models (#46759) Michael Goin 2026-06-25 20:03:09 -06:00
• cc7981599e [Refactor] Remove dead kernel code (#46405) Wentao Ye 2026-06-25 21:09:56 -04:00
• 32bb3195f0 [ModelRunner V2] Bound memory for large logprobs requests (#46746) Nick Hill 2026-06-25 18:04:06 -07:00
• ad28d605e6 [Bugfix] Default tie_weights to sharing the weight (fix tied quantized embeddings, e.g. ModelOpt Gemma4) (#45544) Mike GandMichael Goin 2026-06-25 17:46:28 -07:00
• ae7c8ec223 [Rust Frontend] Switch rustls to native-tls/OpenSSL (#46696) Bugen Zhao 2026-06-26 08:19:44 +08:00
• 1d3f4cb3a4 [Rust Frontend] Extract renderer fixture test utilities (#46719) Bugen Zhao 2026-06-26 08:12:38 +08:00
• f9e684499f [Rust Frontend] Migrate gemma4 to unified parser (#46602) Bugen Zhao 2026-06-26 07:59:57 +08:00
• c53994e134 [Model Runner V2][Spec Decode] Use log1p to compute residual during rejection sampling (#46665) Giancarlo Delfin 2026-06-25 18:46:10 -05:00
• 27da2a2ac4 [Hardware][AMD][CI] Use Triton-based AITER MHA for LM Eval Qwen-3.5 Models Tests (#46691) Matt 2026-06-25 17:08:04 -05:00
• a2e8ec3d52 [CI] Depend GPQA Eval DGX Spark job on arm64 image build (#46736) Michael Goin 2026-06-25 15:07:04 -06:00
• e8c24a7695 [Kernel] Vectorized fp32 moe_sum reduction and support any topk (#46643) Michael GoinandClaude 2026-06-25 15:02:28 -06:00
• 2a6f8f0c05 [ROCm][CI] Fine-tuning queues and test names (#39238) Andreas Karatzas 2026-06-25 15:24:09 -05:00
• c5e3c40877 Fix P/D with DP Supervisor (#46628) Robert Shaw 2026-06-25 16:13:08 -04:00
• 8b4d93ba2b [Perf] Remove redundant clone for GLM, Deepseek etc (#46651) Wentao Ye 2026-06-25 16:09:00 -04:00
• e8e7b592d1 [Kernel][MoE] Tune block-FP8 fused MoE for low-batch decode (#46642) Michael Goin 2026-06-25 13:38:28 -06:00
• e53a17232c [ROCm]: Bump aiter to 0.1.16.post2 (#46692) Rohan Potdar 2026-06-25 13:53:26 -05:00
• 96eb8ddc41 [CI] Re-enable skipped glm and seedoss parser tests (#46671) Flora Feng 2026-06-25 13:11:41 -04:00
• 8fa36fbbeb [Bugfix] FLASHINFER_MLA_SPARSE_SM120 compatibility with GLM-5 NVFP4 (#46506) Gabriel Wu 2026-06-26 00:12:00 +08:00
• e45b279928 [Bugfix] Fix NVFP4+MTP crash: force unquantized mtp.fc for Qwen3Next (#46316) Ranran 2026-06-25 11:05:04 -05:00
• d490b98162 [Core] Avoid mixed length specdec batches via padding (#45237) 5 people 2026-06-25 23:34:44 +08:00
• 1744adc256 [ROCM] [Communication] Add INT3 quantization method for quickreduce (#45666) haoyangli0109 2026-06-25 23:14:15 +08:00
• cdfa2fd7e9 [ROCm][CI] rm duplicate Distributed Torchrun ci test (#46729) Divakar Verma 2026-06-25 10:58:17 -04:00
• 6f3da461d1 [Pooling] Fix Cohere embed billed image token accounting for mixed-content inputs (#46093) Taneem Ibrahimandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-06-25 10:44:29 -04:00