Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• 446c769482 [Distributed] Add opt-in FlashInfer PCIe IPC all-reduce backend (#53576) 3 people 2026-09-01 11:34:56 +08:00
• e16b5e518d [1/N][warmup][DSv4] Migrate generic MLA metadata and indexing kernels (#50175) Roberto L. CastroandCodex 2026-09-01 05:26:05 +02:00
• 188716ace7 [Bugfix][EC Connector] Fall back when MADV_POPULATE_WRITE is unsupported (#53190) xiaolinchenandClaude Opus 4.8 2026-09-01 11:17:45 +08:00
• 923949e6e3 [Feat] Add request-level preemption count histogram metric (#49984) Amy (Yu-Hsuan) Lin 2026-09-01 03:04:54 +00:00
• 882ca8d696 [Kernel] add Flashinfer cutedsl w4a16 linear (#53014) Lain 2026-08-31 19:49:14 -07:00
• 907b1a7f22 [CI][ROCm] Avoid redundant image pulls during smoke validation (#54408) Andreas KaratzasandOpenAI Codex 2026-08-31 19:48:25 -07:00
• c28feab989 [Core][MRV2] Freeze gc during V2 CG capture; skip per-descriptor cleanup (#54646) Nick Hill 2026-08-31 19:37:46 -07:00
• 225aec4809 [Rust Frontend] Migrate to new tekken crate (#53056) Jorge Menjivar 2026-08-31 19:35:21 -07:00
• 45aed9b0cd [CI] Broaden tool-calling issue auto-labeling (#54650) Flora Feng 2026-08-31 22:07:21 -04:00
• 22df3a34e0 [Perf][Rust Frontend] Count the tokenizer vocabulary once at construction (#54449) 3 people 2026-09-01 02:25:06 +01:00
• 58dace61fa [Kernel] Make prefix-prefill tiling independent of the KV page size (#54194) Zheng GongandCursor Agent 2026-09-01 09:19:55 +08:00
• 4c21d41733 [XPU] Route activation CustomOps to SYCL kernels (#53734) 3 people 2026-09-01 03:19:52 +02:00
• 8600db5dff [CI] Build CPU image against torch nightly for TORCH_NIGHTLY runs (#48750) 4 people 2026-08-31 20:25:57 -04:00
• e29af0a2af [XPU] bump up auto-round-lib to 0.15.0 (#54515) Zhenzhong Xu 2026-09-01 08:24:28 +08:00
• d4329ba53d [Bugfix][Rust Frontend] Fix adjacent DeepSeek V4 user content rendering (#53281) 3 people 2026-09-01 08:23:24 +08:00
• 89df6fcb80 [CI] Broaden structured-output issue auto-labeling (#54645) Flora Feng 2026-08-31 19:51:55 -04:00
• 6bafc049aa [Bugfix][PP] Never drop a decoding request from the sampled-token broadcast (#54436) AusarandClaude 2026-09-01 07:49:54 +08:00
• b05acd2ae0 [XPU] [CI] Add retry for v1/sample in Intel GPU CI (#53669) xiangdongandKunshang Ji 2026-09-01 07:34:57 +08:00
• 91752b7a3e [K3 Bug] Fix Kimi-K3 RecoverSSM startup failure 'MambaAttentionBackendEnum.GDN_ATTN declares 4 states, but provides 2 state copy funcs' (#54634) Wentao Ye 2026-08-31 19:22:49 -04:00
• 3a2ed6cbae [Kimi Bug] Fix gdn build_attn_metadata 'KimiK3KDAMetadataBuilder' object has no attribute 'layer_names' (#54636) Wentao Ye 2026-08-31 19:21:01 -04:00
• d6d6658543 [Kimi-K3][Perf] Make native CUDA AttnRes the SM100 default (#54261) Yongye Zhu 2026-08-31 16:18:12 -07:00
• 07ea9350ba [Kernel][Gemma4] Prune Triton sliding-window tiles for multimodal prefixes (#53147) mobicham 2026-09-01 00:21:22 +02:00
• d61b6e1878 [Bugfix][Spec Decode] Take the DFlash draft's RoPE layout from its own config (#54373) Zihan ZhangandClaude Opus 5 2026-08-31 15:13:20 -07:00
• 24d42f3553 [CI] Mark 1-GPU L4 test steps with device: l4 for EKS migration (#54549) Kevin H. Luu 2026-08-31 14:56:27 -07:00
• f5c3cc240b [Perf][Kernel] Tune cooperative topk for medium batch-sizes (#53382) 3 people 2026-08-31 22:44:38 +02:00
• 85c1365bd9 [Bugfix] NemotronHMTP: add hf_to_vllm_mapper so quant exclusions reach the MTP draft (#53790) Juhi MittalandClaude Opus 5 2026-08-31 12:10:24 -07:00
• a9dc631429 [Bugfix] Reject empty bad-word tokenizations (#53433) Andreas Karatzas 2026-08-31 12:07:17 -07:00
• f9c7c6e090 [Rust Frontend][CI] Remove TCP port races from mock-engine tests (#54481) Andreas KaratzasandOpenAI Codex 2026-08-31 12:02:55 -07:00
• f5e441de10 [Bugfix][Test] Fix off-by-one error in sampled token rank causing flaky logprobs test (#53976) Qiming Zhang 2026-08-31 11:49:43 -07:00
• 65ce85fcdc Add Laguna-XS-2.1-INT4 to nightly CI (#52961) Joe RowellandMisha Goin 2026-08-31 20:22:00 +02:00
• 2ba984a5d0 [ROCm][DSpark][DCP] Serve prefix cache hits under DCP for Kimi-K3 (#53598) YukioZzzandandyluo7 2026-09-01 02:18:10 +08:00
• dbb7fffddb [ROCm][MLA][DCP] Support causal multi-token verification (#51705) 3 people 2026-09-01 01:42:28 +08:00
• 39e276eaeb [Structured Output] Let terminal grammars stop under min_tokens (II) (#54218) Artem Perevedentsev 2026-08-31 19:39:27 +03:00
• e9dd6d4834 [CI] Exclude kv_transfer changes from broad spec-decode/kernels/multimodal triggers (#54365) Nicolò Lucchesi 2026-08-31 18:17:12 +02:00
• 3593c964de [ROCm] Add TheRock preview docker updates, Keep Python 3.12 and Ubuntu 22.04 (#49925) rasmithandAndreas Karatzas 2026-08-31 11:09:46 -05:00
• 82936c409d [Tests][XPU] Limit Qwen2-VL generation length to avoid flaky numerical divergence (#54172) Fanli LinandKunshang Ji 2026-08-31 22:59:57 +08:00
• bd575a0d0b [AutoRound] Support AutoRound Format Block-Wise FP8 in vLLM (#47434) Zhenzhong XuandKunshang Ji 2026-08-31 22:56:20 +08:00
• dafbef15a1 [Core] Add max_num_queued_reqs and max_num_queued_tokens for queue size management (#49445) Nicolò Lucchesi 2026-08-31 16:26:49 +02:00
• 76ff0cdff2 [Bugfix][ROCm] Preserve AITER unified-attention metadata during graph replay (#53821) andyluo7andOpenAI Codex 2026-08-31 06:53:12 -07:00
• c5d840ff6a [KV Connector][Offloading] Certify attention-only hybrids in the canonical portability gate (#51689) 3 people 2026-08-31 15:53:19 +03:00
• 4c58a0c398 [Bugfix][KV Offload] Unlink /dev/shm region after all workers map it (barrier variant of #51317) (#52596) 3 people 2026-08-31 15:53:00 +03:00
• e0d27040dd [Bugfix][KV Offload][P2P] Preserve aborted loads until abort completion (#52571) Uttam KumarandOr Ozeri 2026-08-31 04:34:48 -07:00
• e2c8eeac40 [kernel] Fused embedding kernel (#53677) Jee Jee Li 2026-08-31 19:34:14 +08:00
• bed3280f50 [KV offload] Order CPU->GPU loads against the compute stream (#50696) 3 people 2026-08-31 14:31:18 +03:00
• 810bc3250c [Frontend][Performance] Resolve async media across modalities concurrently (#54537) waizuichougou 2026-08-31 18:50:17 +08:00
• 9debcd5990 [Test][Qwen3-VL] Cover compiled DeepStack input contract (#53529) maithilijoshi20 2026-08-31 19:46:50 +09:00
• 9acbc5360a [KV Offload] Preserve KV event metadata until final residency removal (#52068) Moein Khazraee 2026-08-31 03:28:30 -07:00
• eeb549a74d [Frontend] Move engine/protocol.py out openai folder (#54492) wang.yuqiandCyrus Leung 2026-08-31 18:13:50 +08:00
• d8de4ae322 [Bugfix][KVOffload] P2P tier declares REQUEST_LEVEL on the producer leg (#52912) liranschour 2026-08-31 13:07:20 +03:00
• 699e180df4 [Bugfix][SM120] DSv4: pass contiguous C128A decode topk indices on SM120 (#53574) 3 people 2026-08-31 17:44:42 +08:00
• 28bf75c9a9 [Bugfix][Frontend] Truncate prompt_is_token_ids with the prompt (#54509) Hotragn Pettugani 2026-08-31 05:29:01 -04:00
• 399247cc88 [Bugfix][MM] Fix MiniCPM-o image processor reuse on Transformers v5 (#54501) Andreas Karatzas 2026-08-31 02:07:52 -07:00
• 5bfd76372d [Renderer] Shutdown the renderer properly. (#52124) wang.yuqi 2026-08-31 16:40:37 +08:00
• 1b9539d37c [Quantization][Autoround][XPU] Support AutoRound MXFP8 MoE models (#51248) jl9876andKunshang Ji 2026-08-31 16:37:01 +08:00
• c6c33f2b1f [CPU] Support FP16/BF16 persisted GDN state on AMX (#52191) Tianmu LiandOpenAI Codex 2026-08-31 01:32:20 -07:00
• 2a61f060d3 [XPU] Ensure unquantized linear weight is N-contiguous (#53536) zofiaandopencode 2026-08-31 16:01:24 +08:00
• fdbf2ddbd2 [Bugfix][CPU] Fix several bugs (#54042) Li, Jiang 2026-08-31 15:33:06 +08:00
• 2cf82bcdd1 [Bugfix][DCP] Fix NVIDIA DeepSeek-V3.2 / GLM-5.2 fused attention (#50005) 3 people 2026-08-31 15:30:45 +08:00
• f9d666f917 [KV Offload] Forward ownership in KV cache events (#52067) 3 people 2026-08-31 00:28:41 -07:00
• 8e92248f79 [Kernel] Retire the DSv3 router GEMM CUDA kernel (#54040) Jee Jee Li 2026-08-31 15:15:31 +08:00
• da0b2d8b17 [Performance] Optimize Dots3 NOTE runtime (#53517) 3 people 2026-08-31 15:05:18 +08:00
• 7292ee2791 [Bugfix][Multimodal] Honor modality-scoped mm_processor_kwargs (#53808) Prudhvi Vuda 2026-08-31 02:37:12 -04:00
• 4ae172231c [Frontend] Forward cache salt for content parts (#54315) eligottsandOpenAI Codex 2026-08-30 23:29:53 -07:00
• 8fd9eb85d5 [Bugfix][Frontend] Truncate prompt_token_offsets with the prompt (#54407) Hotragn Pettugani 2026-08-31 02:13:37 -04:00
• e126687a9a [Model] Support Qwen3.8-Flash-Next (#53896) +515 people 2026-08-31 13:57:56 +08:00
• 687db59744 [Bugfix][Frontend] Truncate pooling prompts before padding them (#54364) Hotragn Pettugani 2026-08-31 01:12:03 -04:00
• 44fe2a392b [CPU] add CPU support for Voxtral (#53921) LironKesem 2026-08-31 00:20:37 -04:00
• 648b7468b8 [CI/Build] Fix Kimi K3 Eagle3 test fixture (#54482) Jee Jee Li 2026-08-31 11:29:45 +08:00
• 555ea65e8c [Frontend] Add video embeds input support (#54242) Isotr0py 2026-08-31 11:20:47 +08:00
• 5707355209 [Bugfix][MLA] Fix BLHNC addressing for FlashInfer sparse MLA (#54465) Lucas Wilkinson 2026-08-30 21:13:42 -04:00
• c92b29a1d4 [Bugfix] Make metadata send non-blocking in GroupCoordinator.isend_tensor_dict (#49274) AI-Infra Rookie 2026-08-31 07:51:25 +08:00
• 7ab2923489 [Flashinfer] Upgrade Flashinfer version to 0.6.18 (#54313) 4 people 2026-08-30 19:01:05 -04:00
• 7a100bb617 [CI] Restore gpu_1_queue routing for torch-abi audit (#54468) Kevin H. Luu 2026-08-30 14:59:36 -07:00
• b2dc864bb6 [Bugfix][Spec Decode] Keep default CUDA graph sizes memory-safe (#54418) Kevin H. Luu 2026-08-30 11:02:31 -07:00
• 9d0fe9bac8 [Bugfix][Quantization][MoE] Route weight only NVFP4 checkpoints through W4A16 (#54427) Aditya JhaandClaude 2026-08-30 22:32:15 +05:30
• e79961c946 [codeowners] Add jperezdealgaba to security file ownership (#54358) Juan Pérez de Algaba 2026-08-30 18:24:21 +02:00
• f6895a5fcb [Bugfix][Multimodal] Avoid caching full prompts in fallback (#54439) waizuichougouandOpenAI Codex 2026-08-31 00:19:38 +08:00
• 56058fd572 [Bugfix][Kernel] Keep packed GDN decode beta in FP32 (#53877) 4 people 2026-08-30 22:15:36 +08:00
• 87b9b5b8d9 [Test][VLM] Add batch-invariance tests for Qwen3-VL (#53531) macheroandClaude 2026-08-30 21:49:41 +08:00
• fe6db3ed5e [Bugfix] Validate scale-out transfer params (#54324) Taneem Ibrahim 2026-08-30 08:46:12 -05:00
• 488b6da105 [Doc] Fix griffe warnings in HYV4 tool parser (#54412) Harry MellorandClaude Opus 5 2026-08-30 06:21:22 -07:00
• 5e71a11eb2 [CI] Mark L4 GPU test steps with device: l4 for EKS migration (#54326) Kevin H. Luu 2026-08-30 04:40:41 -07:00
• 78fa18910e ci: add MIG slice size to H200 job labels (#54420) Kevin H. Luu 2026-08-30 04:35:01 -07:00
• 8c51b92654 [Bugfix] Avoid global config lookup in sparse indexer forward (#54400) Jiangyun ZhuandOpenAI Codex 2026-08-30 18:31:47 +08:00
• 2c7d7dd64a [Perf][ROCm] Dual-stream decode with hipgraphs (#52033) 3 people 2026-08-30 10:24:20 +02:00
• 4f78a8fdd0 [Model] Honor cap_pixels_per_frame in Qwen3-VL memory profiling (#54380) Devon KrismanandClaude Fable 5 2026-08-30 04:07:29 -04:00
• b383e16396 [Bugfix] Reset cached Mamba align metadata on profiling teardown (#54044) Canlin Guoandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-08-30 15:43:20 +08:00
• 8fa4c6cdb3 [CI] Add explicit step keys to 18 hardware test steps (#54330) Kevin H. LuuandKimi Code 2026-08-30 00:08:25 -07:00
• 1dc464d426 [Bugfix] Bound cache_salt length to prevent DoS via scheduler CPU exhaustion (#54353) Juan Pérez de Algaba 2026-08-30 06:28:16 +02:00
• 1ebff996a1 [Fix] Improve ROCm detection in WSL environments (#38434) Yizhou 2026-08-30 12:23:10 +08:00
• 7a67941c1d [Rust Frontend][gRPC] Add audio and video media inputs (#53760) Connor Carpenter 2026-08-29 23:31:37 -04:00
• b016ed8ea3 [Bugfix][Multimodal] Release Qwen2.5-VL and Qwen3-VL RoPE caches with the model (#54346) 3 people 2026-08-30 11:10:29 +08:00
• dbf662c9e8 [ROCm][MLA] Reach FULL cudagraphs for AITER MLA speculative decoding (#51171) 4 people 2026-08-29 21:49:55 -05:00
• b5707bf994 [Bugfix][MoE] Enable cuBLAS out_dtype router GEMM on all CUDA archs (fixes family-120/GB10) (#54048) hcl 2026-08-30 10:18:42 +08:00
• 680e2177e4 [CI][Ray] Fix flaky multi-node assignment test after placement-group teardown (#53621) yzong-rh 2026-08-29 19:48:22 -04:00
• 6cddad414e [CI][Test] Deflake test_mem.py sleep-mode asserts via allocator bookeeping (#54312) Oxana KorzhandClaude Opus 5 2026-08-29 13:05:17 -06:00
• fe755c8899 [Bugfix][Model Runner V2][Spec Decode] Decouple the draft's gumbel noise stream from the target's (#54282) Giancarlo Delfin 2026-08-29 09:58:29 -07:00
• 7fbfca2670 [Bugfix][Frontend] Only echo the assistant turn in batched chat completions (#52529) Kaif KohariandCyrus Leung 2026-08-29 17:44:17 +01:00
• 5b0e5b69ac [Bugfix][Frontend] Validate stop_token_ids against vocab size (#54196) jackandQwertyJack 2026-08-30 00:25:35 +08:00
• 129087ddab [Perf] Reuse topk SparseMatrix routing metadata in GPT-OSS MoE forward (#45457) Shenglei FuandClaude Fable 5 2026-08-29 09:14:39 -07:00