Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• b6e2aa748b [CI] Initialize warmup registry in V2 QSA runner fixture (#56707) Lucas WilkinsonandOpenAI Codex 2026-09-13 13:34:09 -04:00
• dec0b5d63f [Bugfix] Skip Triton autotune inspection without Triton (#56676) Kevin H. Luu 2026-09-13 10:05:29 -07:00
• f09c52a587 [ROCm][Performance] Avoid blocking MiniMax M3 scalar upload (#56170) 3 people 2026-09-13 12:56:02 -04:00
• b7e0cdac5d [Bugfix] Detect unloaded NVFP4 weight scales with a NaN sentinel (#52501) 3 people 2026-09-13 07:58:09 -07:00
• 82a85dc1d2 [Attention] Add Triton/FlashInfer composite for multimodal prefix attention (#56305) 5 people 2026-09-13 10:29:41 -04:00
• 8cf9de9080 [Test][Determinism] Cover VLM batch invariance in default execution mode (#56528) Shenglei FuandClaude Sonnet 5 2026-09-13 07:22:03 -07:00
• dd4c841070 [LoRA] Add LoRA support for DeepSeek-V4 Flash Vision (#55897) linitra24 2026-09-13 22:10:54 +08:00
• de500290ff [XPU][CI] fix jit_warmup_triton_launcher (#56670) liuzhenwei 2026-09-13 21:44:24 +08:00
• 2e9f7bb454 [Pooling] Report actual input token usage for scoring APIs (#56573) Taneem Ibrahim 2026-09-13 08:20:21 -05:00
• c351fd3c64 [Core] Fix ValueError on KV load failure with a hybrid KV cache (#50388) Kebe 2026-09-13 22:03:05 +09:00
• fa1b3b1922 [Bugfix][EPD] Preserve explicit multimodal UUIDs with caches disabled (#56432) Tianyu Guo 2026-09-13 18:08:30 +08:00
• d2d649e674 [DS V4.1][Engram] Support async prefetch for offloaded engram lookups and engram DP sharding (#56512) 5 people 2026-09-13 02:44:23 -07:00
• e7a3963339 [Bugfix] Avoid repeated dummy initialization and random CPU Engram fills (#56682) Woosuk Kwon 2026-09-13 02:23:20 -07:00
• a987777755 [Bugfix][Gemma4] Keep image kwargs out of video preprocessing (#56652) YleanandCodex 2026-09-13 17:07:12 +08:00
• fa008bdccf [Perf][Kernel] Integrate DeepSelect TopK for the DSA sparse indexer (#56464) 3 people 2026-09-13 15:44:37 +08:00
• 1cfd972816 [CI] Fix NIXL transfer-rank geometry fixture (#56672) Lucas Wilkinson 2026-09-13 03:36:06 -04:00
• 410f6da5c4 [Bugfix][KV Offload] Submit CPU stores on no-forward steps (#56621) Lucas Wilkinson 2026-09-13 01:56:14 -04:00
• 0cf266a469 [Pooling] Support prompt embeddings in MRV2 decoder pooling (#56639) Taneem Ibrahim 2026-09-13 00:49:45 -05:00
• 93e051e290 [Bugfix][Benchmark] Make streaming TTFT/E2E latency accounting consistent across endpoints (#55508) SurajandClaude Opus 5 2026-09-13 01:20:54 -04:00
• 2671fedfc7 [MRV2] Revert explicit Triton JIT warmup migration (#56654) Woosuk KwonandOpenAI Codex 2026-09-12 21:07:56 -07:00
• c711f740b7 [ROCm] Auto-enable breakable CUDA graphs for DeepseekV41ForCausalLM (#56349) Markus HartikainenandJee Jee Li 2026-09-12 20:47:07 -07:00
• c2ea9f1a5f [Kernel][Perf][Quantization] Fix odd-row performance cliff in per-token-group quantization (#56478) Ayushman SinghandAyushman Singh 2026-09-12 23:27:08 -04:00
• 533dbe424b [Misc] Clean up fast loader daemon quant method verification (#55656) Isotr0py 2026-09-13 11:16:39 +08:00
• e52be1a62d [9/N][warmup][DSv4] Migrate MHC TileLang kernels (#50178) Roberto L. CastroandCodex 2026-09-13 02:25:23 +02:00
• aed894c190 [6/N][warmup][DSv4] Migrate sampling, and DFlash JIT kernels (#56323) 3 people 2026-09-13 02:24:24 +02:00
• e19a3e172e [5/N] Share HiSparse host cache across TP ranks (reopens #52760) (#56629) 6 people 2026-09-12 19:48:25 -04:00
• ebe1dec2da [PCP][DCP] Enable PCP+DCP on sparse-MLA models (#56157) 6 people 2026-09-13 01:24:44 +02:00
• 72d4d83157 [ROCm][Bugfix] Fix elastic EP scaling deadlock (#56610) MattandClaude Opus 4.8 2026-09-12 16:09:19 -07:00
• 7ee8a6dd01 [watermarking] Dual-key gumbel-max watermarking for speculative decoding support (#56122) 4 people 2026-09-12 20:42:01 +02:00
• 1ee4be4dbc [CI] Update DeepSeek V4.1 MegaMoE routing test (#56599) Kevin H. LuuandOpenAI Codex 2026-09-12 11:31:13 -07:00
• fd7cd4b883 [CI] Fix Qwen3 Omni DSpark load test config (#56600) Kevin H. LuuandOpenAI Codex 2026-09-12 11:29:58 -07:00
• 7bcca16729 [Bugfix] Fix DeepGEMM FP8 warmup coverage (#56452) Ayushman Singhandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-09-12 14:20:38 -04:00
• 29332cf936 [4/N] Expose HiSparse cache metrics via KV connector stats (#56061) 3 people 2026-09-12 19:47:55 +02:00
• dff76bc3e8 [Bugfix][CI] skip conftest for NPU compatibility test (#55799) Zeyu Yang 2026-09-13 01:03:28 +08:00
• 3bb23e2def [CI/Build] Add DSv4.1 auto-label rules and narrow DSv4 (#56332) jcotant-inferactandClaude 2026-09-12 09:50:16 -07:00
• 9d3e991ece Add @arpera to CODEOWNERS of Structured Output (#56501) Artem Perevedentsev 2026-09-12 19:32:12 +03:00
• c377114636 [Bugfix][CI] Update CuTeDSL indexer Q sentinel test for migrated wrappers (#56594) Roberto L. Castro 2026-09-12 18:12:58 +02:00
• 22f6e4eccb [Frontend][Rust] Reject empty structured-output values (#54821) Clinton ThomasandLucas Bourtoule 2026-09-12 11:12:25 -04:00
• 658c8131c7 [Elastic EP] Reuse CUDA graphs across reconfiguration (#54985) Itay Alroy 2026-09-12 17:26:39 +03:00
• 13e221f830 [Perf] Fuse DSV4.1 input metadata preparation with Triton (#56562) Woosuk Kwon 2026-09-12 07:20:04 -07:00
• 2f59050eda [XPU][CI] Skip ROCm test on non-ROCm platforms (#56555) liuzhenwei 2026-09-12 21:26:55 +08:00
• 986e2f870a [Refactor][ROCm] Migrate the RDNA3 W4A16 MoE to the oracle/experts pa… (#55522) JartX 2026-09-12 15:20:26 +02:00
• 1b29c508bc [Bugfix] Initialize data parser in Nano-Nemotron audio test (#56401) Taneem Ibrahim 2026-09-12 08:07:20 -05:00
• a0914ab7d0 [Nano-Nemotron] Fix Nano-Nemotron precomputed multimodal embeddings (#56398) Taneem Ibrahim 2026-09-12 08:07:10 -05:00
• bde4feb061 [XPU][CI] Remove pip install dependency in test yaml files (#55171) Kunshang Ji 2026-09-12 20:03:12 +08:00
• d86257e283 [CI/Build] Give the Torch ABI audit time to start (#56596) Kevin H. LuuandOpenAI Codex 2026-09-12 03:29:22 -07:00
• 6b153463a8 [Build] Define _USE_MATH_DEFINES for FlashMLA targets (#54007) Srijan KeshriandShengqi Chen 2026-09-12 15:15:32 +05:30
• 33fa95a0cf [CI] Sample GPU utilization and memory alongside test timelines (#56541) Kevin H. LuuandCodex 2026-09-12 02:41:54 -07:00
• 756794a9a7 [Frontend] create unified Cohere parser (#56392) Walter Beller-MoralesandFlora Feng 2026-09-12 04:04:46 -04:00
• d406f09eb8 [ROCm][CI] Stage F gating (#55252) Andreas KaratzasandOpenAI Codex 2026-09-12 02:40:14 -05:00
• 30118ba27d [DSV4.1] Remove compressor-aware image sentinel token padding (#56554) Isotr0py 2026-09-12 15:07:36 +08:00
• 372477d2d9 [Bugfix][Examples] Launch prefill and decode concurrently in NixlPushConnector demo (#55088) junuxyz 2026-09-12 15:22:34 +09:00
• dca96bf97b [Bugfix][Spec Decode] Avoid fastsafetensors deadlock for PP draft models (#54416) Canlin GuoandCodex 2026-09-12 14:22:30 +08:00
• eed1f3d0c6 [Rust Frontend] Support generation blocks in HF chat template (#56378) Bugen Zhao 2026-09-12 12:56:55 +08:00
• d43bb2f37f [3/N] HiSparse: host-resident sparse-MLA decode hot-buffering (#53781) 8 people 2026-09-12 00:36:47 -04:00
• 120ec4ebd2 [5/N][warmup][DSv4] Migrate NVIDIA CuTeDSL attention kernels (#53566) Roberto L. CastroandCodex 2026-09-12 06:05:59 +02:00
• 485421b1c3 [Rust Frontend] Honor HF revisions, offline mode, and cache directory (#56386) Bugen ZhaoandCodex 2026-09-12 11:39:17 +08:00
• 5f3e4b7447 Revert "[Agents] Link Triton skill to JIT kernel warmup guide" (#56546) Woosuk Kwon 2026-09-11 20:15:02 -07:00
• ec6b0494fe [CI] Bump CUTLASS DSL to 4.7 (#54927) Jiangyun ZhuandCodex 2026-09-12 11:11:29 +08:00
• 31759ccb6d [Rust Frontend][Multimodal] Accept preprocessed multimodal gRPC features (#55047) 3 people 2026-09-11 19:29:49 -07:00
• a0844fa6c6 [XPU][CI] skip DeepSeek-V4.1-Flash in test_tensor_schema.py (#56463) liuzhenwei 2026-09-12 09:24:30 +08:00
• 06e57f622c [ROCm][Kimi-K3] Fix non-contiguous state_indices crash and GPU-sync assert in fused KDA/MLA prefill (#56526) Divakar Verma 2026-09-11 21:17:39 -04:00
• 80650454dc [CI] Give every Buildkite step an explicit key, and a hook to enforce it (#56329) Tahsin Tunan 2026-09-12 07:08:05 +06:00
• 56d7270138 [Bugfix][NIXL] Don't evict a remote engine a transfer is still reading from (#54689) Yizheng Jiao 2026-09-11 18:06:09 -07:00
• 46d2b23ac5 [ROCm][DSV4.1][Perf] Use AITER mHC for the delayed pre block (#56503) Fangzhou AiandFangzhou Ai 2026-09-11 20:35:48 -04:00
• 9d88ceb026 [KDA] Update flashKDA to support bf16 checkpoint state (#56485) Wei Zhao 2026-09-11 19:53:24 -04:00
• 988d9b6777 [Rust Frontend][gRPC] Surface engine generation errors (#56405) AlecandCodex 2026-09-11 15:31:55 -07:00
• bbaa1b9340 [ROCm][CI] Extend timeout for Basic Models (other) (#56522) Micah Williamson 2026-09-11 17:11:59 -05:00
• 2d75e586fc [ROCm][Kernel][DSV4] Remove tl.constexpr to avoid cold-compile churn in indexer gather kernel (#56153) amd-xavierwang 2026-09-11 17:34:17 -04:00
• 9dcf6bf344 [BugFix] Fix DP token padding in dflash attention metadata (#56181) Giancarlo Delfin 2026-09-11 14:33:53 -07:00
• c1b69aa0d4 [Bugfix][KV Connector] Only enforce disk block alignment for O_DIRECT (#55424) mevinceandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-09-11 13:12:59 -07:00
• cc5dd0a857 [MRV1] Scope breakable cudagraphs to the piecewise path only (#56312) Nick HillandAndreas Karatzas 2026-09-11 12:55:34 -07:00
• dffb863c97 Upgrade tpu-inference to v0.29.0 (#56388) Lyle Lai 2026-09-12 03:32:47 +08:00
• 0c1e89ceb9 [Bugfix][Kimi-K3] Fix KDA projection overlap on Hopper (#55426) chengchengpeiandTyler Michael Smith 2026-09-11 12:07:30 -07:00
• 00e5cda926 [Agents] Link Triton skill to JIT kernel warmup guide (#56499) Misha Goin 2026-09-11 15:05:41 -04:00
• 8a7f98c8c3 [Kernel][MoE] Optimize batched_moe_align_block_size with cooperative writes (#53280) Misha Goin 2026-09-11 15:05:22 -04:00
• dc07f1638f [Bugfix][MLA] Read sparse model settings from text config (#56160) Andy Lo 2026-09-11 19:02:13 +01:00
• 6fe67cbbf3 [Spec][V2] Support MTP speculative decoding under pipeline parallelism (#46994) 3 people 2026-09-11 13:55:21 -04:00
• 9a35c081e8 [Docs] Correct API key authentication scope (/inference does NOT bypass the API key) (#56269) Sören Dréano 2026-09-11 18:54:35 +02:00
• 9dd969da09 [Model][ROCm] Enable DeepSeek V4 Vision (#55107) Andreas KaratzasandOpenAI Codex 2026-09-11 11:17:30 -05:00
• 127143e27f Revert "[Rocm][Kimi-k3] Fix pipeline_parallel support for the kimik3 DCP mode (#53664)" (#56429) Shanshan Shen 2026-09-12 00:17:26 +08:00
• b87339888d [httpx migration] Import httpx from huggingface_hub (#56460) LucainandClaude Opus 5 2026-09-11 17:29:31 +02:00
• 5fe77aecfc [Deprecation] Deprecate items scheduled for 0.29 (#55353) Wentao Ye 2026-09-11 11:17:02 -04:00
• 1e1060f998 [Kimi Perf] Group fp8 mla cahche insertion, 4~6x kernel level performance improvement for small batch (#55356) Wentao Ye 2026-09-11 11:16:57 -04:00
• 22258a26bc [Multimodal] Use GPU NVDEC for EPD encoder-only instance video media IO (#53675) Isotr0py 2026-09-11 22:20:11 +08:00
• 5392fbca2a [ROCm][Docker] Pin AINIC apt repo to snapshot 1.117.5-a-77 (#56459) djramic 2026-09-11 16:14:16 +02:00
• d5a9d0f59b [Metrics] Consolidate Prometheus histogram bucket defaults into a single module (#48866) Guy StoneandClaude Fable 5 2026-09-11 15:12:36 +01:00
• c191787a68 [Bugfix] Align vLLM YaRN with Transformers and stop re-scaling max_model_len (#56446) Harry MellorandClaude Opus 5 2026-09-11 14:01:38 +01:00
• a5b714eee4 [Bugfix] Fix Qwen3-VL and Cosmos3-Edge text architectures for CPU and pipeline parallelism (#53699) labAxiaoming 2026-09-11 20:58:15 +08:00
• 912dfb3758 [Bugfix][Scoring] Warn when serving original Qwen3 reranker without chat template (#56017) 김재억andClaude Opus 5 2026-09-11 21:52:45 +09:00
• 89dbb26445 [Bugfix] Fix GLM-OCR MTP position masking during CUDA graph capture (#56447) Yuxuan Zhang 2026-09-11 20:31:10 +08:00
• 295ac4e52e [Bugfix][Multimodal] Parse decoded video frame lists as a single video (#55326) 4 people 2026-09-11 20:16:05 +08:00
• 7ef4d9bfed [Bugfix][Responses] Fix browser.find action type (#55305) Nicole LiHui 🥜andChauncey 2026-09-11 20:07:04 +08:00
• b4da4d17ae [ROCm][Bugfix] Fix AITER preshuffled FP8 block-scale kernel (#56433) Matt 2026-09-11 04:44:57 -07:00
• a2bc2ffb2c [Bugfix][Pooling] Restore token limits for offline Jina scoring (#56415) shaohuaxiandCodex 2026-09-11 03:26:44 -07:00
• 79c137c6ad [Bugfix][Rust Frontend][Renderer] Align DeepSeek tool-call arguments with deepseek-recipe (#56260) 4 people 2026-09-11 18:19:55 +08:00
• 8c1d1c2974 [Misc] Log FlashInfer allreduce workspace init failure as error (#55127) Nicolò LucchesiandHarry Mellor 2026-09-11 11:35:56 +02:00
• eb7e894432 [ROCm][CI] Add HY-V4 generation coverage (#55667) Andreas Karatzas 2026-09-11 04:24:28 -05:00
• e77daef89e [Model] Support DeepSeek-V4.1-Flash (#56214) 6 people 2026-09-11 02:11:19 -07:00
• e6821ceac9 [XPU][CI] Add decord to test requirements (#56355) Tony Lin 2026-09-11 16:58:27 +08:00