Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• 1ef1c7ebba [CI] split tests to reduce CI time (#48219) Jiangyun Zhu 2026-07-12 04:00:14 +08:00
• 54503ecec0 fix(processor): route MiMo-V2-Omni media fetch through MediaConnector (#43117) 3 people 2026-07-11 08:52:52 -07:00
• 0067311536 fix(entrypoints): stop resolve_items leaking in-flight media fetch tasks on partial failure (#48333) ErenAta16 2026-07-11 18:42:08 +03:00
• 51878e5b6e [2/N][KV-Cache Layout Refactor] Pack K/V into the content dim across attention backends (#44455) 5 people 2026-07-11 11:11:16 -04:00
• 76fedaa2a5 [XPU][UT]Fix InternS1ProForConditionalGeneration AssertionError (#48232) Yejing Lai 2026-07-11 21:56:55 +08:00
• 19069bcbd5 FP32 router GEMV optimization (#48335) 4 people 2026-07-11 21:07:48 +08:00
• 1bd8f80a64 [CI] Point CI at Transformers release rather than release branch (#48328) Harry Mellor 2026-07-11 10:31:14 +01:00
• 0b6636cbcb [XPU]remove is_xxx from moe class and bump up kernels (#48079) Qiming ZhangandKunshang Ji 2026-07-11 02:27:13 -07:00
• 4a6440acef Bump Transformers version to 5.13.0 (#47867) Harry Mellor 2026-07-11 08:56:14 +01:00
• bec0a4ede6 [Revert] [Build] Update vllm ...builds FA3 with torch stable API (#48269) Lucas Wilkinson 2026-07-11 01:20:25 -04:00
• 3d99b0499a [Logs] DP Supervisor Log Improvement (#48278) Robert ShawandRobert Shaw 2026-07-11 00:07:00 -04:00
• 04d553f390 [Misc] Use meta tensor for KV cache stride calculation (#47316) Lucas Wilkinsonandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-07-10 23:24:59 -04:00
• 9c18e90f6c [BugFix] Fix packed HND KV cache reshape for FlashAttention (#47314) Lucas Wilkinsonandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-07-10 23:22:39 -04:00
• 092387963c [BugFix] weights processing peak memory reduction for nvfp4 MoE layers (#46276) Jimmy Lee 2026-07-10 19:05:35 -07:00
• 1bf3997eae [Quantization] Bound peak memory when repacking FP4 MoE weights for Marlin (#47851) Joe Rowellandmgoin 2026-07-11 03:46:13 +02:00
• 29fd688892 Add VLLM_FLASHINFER_AUTOTUNE_SKIP_OPS and skip CuTeDSL fp4_gemm autotuning by default (#48268) Michael GoinandClaude Fable 5 2026-07-10 21:13:08 -04:00
• ed908cf0a0 [Bugfix] Fix thinking_token_budget not enforced after natural </think> re-entry (#45984) Ashwin Giridharan 2026-07-10 15:47:51 -07:00
• 26ff616bbf [Bugfix][Test] Register Qwen/Qwen3.5-4B example model (#48276) Nick HillandClaude Opus 4.8 2026-07-10 22:01:23 +01:00
• f378f79b7c handle topk_ids padding in align sum kernel (#47785) gnovack 2026-07-10 13:33:28 -07:00
• 735def4fcf [Bugfix] Fix FlashMLA dense fp8 metadata crash (num_sm_parts clamp) (#48045) Matthew BonanniandClaude 2026-07-10 15:24:52 -04:00
• c227aaa3f8 [ROCm] Enable DeepSeek-V4 DSpark speculative decoding on AMD (MI350X / MI355X, gfx950) (#47419) larryli2-amdandAndreas Karatzas 2026-07-10 23:22:19 +08:00
• 08dfd68610 [Model] Add LongCat-Flash-Lite (n-gram embedding) (#47857) Michael Goin 2026-07-10 10:17:50 -04:00
• 978a6dfa3f [Build/CI] Build arm64 PR and postmerge image builds for Blackwell SM10x and SM110 (#48041) Tyler Michael Smith 2026-07-10 10:12:50 -04:00
• 85c09e9885 fix: correct load_weights track logic and enable weight integrity for… (#41811) 3 people 2026-07-10 22:08:20 +08:00
• b12cca6a23 [Bugfix] Fix turboquant FP8 cast failure for BF16 models on Ampere GPUs (#39988) 3 people 2026-07-10 21:55:28 +08:00
• e257faf87d [Refactor] Remove unused rocm kernel combine_topk_swa_indices_ragged (#48158) Wentao Ye 2026-07-10 09:33:07 -04:00
• fabec87f63 [Model] Migrate MistralLarge3ForCausalLM to AutoWeightsLoader (#48153) FAN YUCHEN 2026-07-10 20:27:58 +08:00
• 7614b88ebd [Bugfix][Spec Decode] Fix DFlash draft/target layer-count mismatch (#48113) Nick HillandClaude Opus 4.8 2026-07-10 12:42:53 +01:00
• 68ea76e780 [Misc] Remove dead code in ViT functionality test (#48220) Isotr0py 2026-07-10 19:17:42 +08:00
• c241c7a2b0 [Rust Frontend] Add roundtrip fixtures for more chat parsers (#47883) ReidandBugen Zhao 2026-07-10 18:03:58 +08:00
• e23b19309b Deepstream video backend (#42424) 5 people 2026-07-10 14:53:30 +05:30
• f36284a8d2 [CI] Add TORCH_NIGHTLY=1 build mode (run full suite on torch nightly) (#47180) 3 people 2026-07-10 04:38:35 -04:00
• 424df4f65d [Model][CI/Build] Cosmos3: enable registry tests and register Cosmos3-Super (#48211) Mingfei Guo 2026-07-10 01:35:15 -07:00
• 074bdd0d99 [Rust Frontend] Integrate MM video support (#47959) Bugen Zhao 2026-07-10 16:15:33 +08:00
• 216ee58780 Add XPU nightly and release image publishing to DockerHub (#48126) 3 people 2026-07-10 15:56:00 +08:00
• 433f291195 [CI] Right-size test-area timeouts from nightly durations (#48186) Kevin H. LuuandClaude Opus 4.8 2026-07-10 00:53:16 -07:00
• 28eaf05d56 [XPU] Enable v1/sample tests on XPU CI (#44472) Chaojun Zhang 2026-07-10 15:40:51 +08:00
• 300e33797f [Perf] fuse more rmsnorm and all-reduce in qwen3.5 (#46998) Jiangyun Zhu 2026-07-10 15:37:51 +08:00
• 5715fde12c [Feature][Parser] Support include_reasoning param for non-Harmony models (#44301) albertoandChauncey 2026-07-10 08:34:02 +01:00
• e5588e49bc [Core][KV events] Report prefix-cache-reused blocks in full report mode (#45261) 3 people 2026-07-10 13:46:54 +08:00
• 95ed0feaa5 DCP supports hybrid attention (#40996) Yan XuandJingyi Yang 2026-07-10 12:34:45 +08:00
• 2d814a0082 [kv_offload] Emit tier-owned BlockStored events from FS/OBJ secondary tiers (#47923) 3 people 2026-07-09 20:17:23 -07:00
• 88e5e2c57b [CI/Build][AMD] Fix ROCm OOM in eagle_correctness_heavy by reserving CUDA graph memory (#47366) 3 people 2026-07-09 19:14:38 -07:00
• feb384ada2 [bugfix] bge-m3-sparse-plugin mismatch requests (#48112) Augusto Yao 2026-07-10 10:03:00 +08:00
• a0f6d767e4 [ROCm][CI] Move remaining engine/samplers AMD steps to mi325_1 (#48169) peizhang56andClaude 2026-07-09 17:20:15 -07:00
• f1a5adddb8 update marlin M size for EP (#48144) gnovack 2026-07-09 16:52:52 -07:00
• cac3e70cd4 Correct model layer aliasing for Bert style models (#43896) ap9272 2026-07-09 16:46:22 -07:00
• e12b91b032 [CI] Fix cargo-deny config flag ordering (#48170) Lucas Wilkinson 2026-07-09 17:43:58 -04:00
• 766469a4c4 [ROCm] Revert Part of [ROCm] Fix pooling startup workspace lock #47912 (#48154) Micah Williamson 2026-07-09 15:34:24 -05:00
• ea0fa34f49 [CI] Increase extract hidden states TP2 timeout (#48161) Lucas Wilkinson 2026-07-09 16:19:03 -04:00
• bbb0f945ff [ROCm] Synchronize sparse MLA metadata before graph replay (#47404) ZihaoMu 2026-07-10 03:59:56 +08:00
• 2ded1b24e7 [KV Connector][Mooncake] Apply SWA lookup mask before hashing/key build (#47317) 4 people 2026-07-09 12:51:23 -07:00
• b0dec2a11b [ROCM][DSV32][Perf][MTP] Enable UNIFORM_BATCH CG mode in rocm_aiter_mla_sparse (#45149) 3 people 2026-07-09 22:35:10 +03:00
• ff8d3488f2 [Bugfix][MRV2] Reset num_accepted_tokens on add_request in all modes (#48132) Nick HillandClaude Opus 4.8 2026-07-09 19:17:11 +01:00
• 2285cfca46 [KVConnector] MultiConnector: give every sub-connector the request's real blocks in update_state_after_alloc (#46865) weishu 2026-07-09 11:10:39 -07:00
• e08a915146 [Bugfix] Preserve tensor causal metadata for grouped attention (#48135) Lucas WilkinsonandCodex 2026-07-09 13:57:53 -04:00
• 67e7ea8977 [ROCm][CI] Set all timeout_in_minutes to 180 (#48146) Charlie Fu 2026-07-09 12:52:26 -05:00
• 429f405748 [Bugfix] Guard CUDA-only rms_norm_per_block_quant in FUSED_OPS for non-CUDA builds (#47296) 3 people 2026-07-09 17:10:53 +03:00
• 753c5039f0 Pin PyNvVideoCodec to tested 2.0.4 wheel (#48056) Brandon Pelfrey 2026-07-09 07:07:50 -07:00
• 299d2b5655 [CI] Annotate built Docker image tags on the Buildkite build page (#48101) Kevin H. LuuandClaude Opus 4.8 2026-07-09 07:02:40 -07:00
• 85b3a7264b [Bugfix][Model Runner V2] Order uniform decodes first so spec decodes aren't misclassified as prefills (#47381) 3 people 2026-07-09 06:26:27 -07:00
• b83be00cdd Migrate Olmo and Olmo2 to the Transformers modeling backend (#48100) Harry Mellor 2026-07-09 13:00:23 +01:00
• 412414d8e0 Remove PersimmonForCausalLM and FuyuForCausalLM model architectures (#48096) 3 people 2026-07-09 19:59:08 +08:00
• ae6170f874 [P/D][Bugfix] Fix PD async KV load lookahead handling for MTP spec decode (#46694) ChaunceyandNick Hill 2026-07-09 18:01:22 +08:00
• e87521626f Sanitize server file paths from validation error responses (#46415) Muhammad FawazandMahad Durrani 2026-07-09 02:46:29 -07:00
• 1cd75b3dd4 [Bugfix] Fix race condition in KVBlockZeroer (#48085) Benjamin ChislettandWentao Ye 2026-07-09 18:18:19 +09:00
• 0206f10871 Add Intel XPU Docker release pipeline (#47880) 3 people 2026-07-09 16:13:51 +08:00
• ab7961a14a Remove TeleChatForCausalLM (#47989) 3 people 2026-07-09 15:34:54 +08:00
• a07765c6bd [Bugfix] Fix Qwen3-ASR transcription streaming postprocessing (#42478) JooHo LeeandJooHo Lee 2026-07-09 16:33:27 +09:00
• 1171467e91 [CPU] Fix Qwen-Next SSM type for AMX GDN (#48073) Li, Jiang 2026-07-09 15:09:31 +08:00
• 529af88842 [KV Offloading] Add free block iterator for CPU offload scheduling (#47849) Chauncey 2026-07-09 14:59:54 +08:00
• b8c7c86533 [XPU][LoRA] Fix torch.compile DEVICE_LOST by avoiding view-mutation in LoRA shrink (#47944) Chaojun Zhang 2026-07-09 14:08:30 +08:00
• 2c17d33f42 [Bugfix][ROCm] Change AttentionCGSuppoort in TritonMLA to UNIFORM_SINGLE_TOKEN_DECODE (#47144) music-dinoandAndreas Karatzas 2026-07-09 04:09:42 +02:00
• bc44f9feb7 [ROCm][CI][MoE] Fix double-transpose of fused w3 expert weights (#47874) 3 people 2026-07-09 01:59:28 +02:00
• 7802c20c4e [KVConnector][NIXL] Support pipeline-parallel prefill in push mode (#45880) qizixiandNick Hill 2026-07-08 16:49:23 -07:00
• 95d6d6f4bb [Bugfix] Use int8 workspace for FlashInfer MLA decode (#48046) Nick HillandClaude Opus 4.8 2026-07-09 00:39:40 +01:00
• 56da398dac Fix embed scaling + CUDA graphs in Transformers modelling backend (#48010) Harry Mellor 2026-07-09 00:14:33 +01:00
• 26831949b4 [ROCm] Fix pooling startup workspace lock (#47912) Andreas Karatzas 2026-07-08 17:59:50 -05:00
• 6cf7b26bd4 [docs] Fix the docs build (#48008) Martin HickeyandHarry Mellor 2026-07-08 23:47:22 +01:00
• 5f85975624 [Feat] Add runtime monitor for post-warmup TileLang compilation (#46718) Roberto L. Castro 2026-07-09 00:11:28 +02:00
• dcdd756d75 [CI] GSM8K eval integration test for KV offloading (#46893) 6 people 2026-07-08 17:59:48 -04:00
• 0d2f4e7c9c Allow FlashInfer A2A backends for TRTLLM FP8 MoE Modular (#46661) Thien Tran 2026-07-09 05:58:39 +08:00
• 49abadaedb [ROCm][Bugfix] Fix empty-tensor .max() crash in AITER FA (#47894) djramic 2026-07-08 23:58:36 +02:00
• 089e412878 [Perf] Integrate TRTLLM BF16 MoE Modular Kernel (#45182) Kaihang Jiang 2026-07-08 17:36:14 -04:00
• a5d19cbb95 [Core] Move MRV1 late_interaction_runner.py out of MRV2 subtree (#48014) Nick Hill 2026-07-08 19:30:11 +01:00
• 8347c6e6e1 updated flash_attn GIT_TAG to point to torch Stable ABI FA3 commit (#47995) Chris Leonard 2026-07-08 13:56:29 -04:00
• b2cf70ea3a [CI] BugFix Eval Small Models Distributed test for DiffusionGemma (#47980) Ilya MarkovandMarkov Ilya 2026-07-08 19:00:56 +02:00
• d1f1d86797 [Bugfix] Re-enable benchmarking of librispeech dataset. (#47033) almayne 2026-07-08 17:19:26 +01:00
• f05603fa28 [Bugfix][DCP] Cast LSE to fp32 in a2a combine to fix bf16 bitcast crash (#47801) shawn 2026-07-08 08:41:26 -07:00
• c2ecd0f888 Fix FlashAttention MLA prefill V unpadding (#42642) Martin VitandMatthew Bonanni 2026-07-08 11:22:20 -04:00
• 0d12618e98 [Spec Decode] Support hybrid (SWA + full attention) DFlash drafters (#47914) Michael GoinandClaude Opus 4.8 2026-07-08 11:12:45 -04:00
• 68b4a1d582 Fix NVML capability lookup for visible devices (#47892) Tyler Michael Smith 2026-07-08 09:07:44 -04:00
• 572b25b03e [Bug] Fix Batched DeepGEMM (#47884) 3 people 2026-07-08 09:05:03 -04:00
• 9f2b3b093c Improvement of Docker image build for IBM Power using prebuilt wheels from IBM published devpi index (#46017) 4 people 2026-07-08 18:31:14 +05:30
• cd0de48d08 [Bugfix][V1] Free out-of-window blocks on the processed-token basis under async scheduling (#47728) 4 people 2026-07-08 20:34:19 +08:00
• 934eeaecfb [CI/Build][BugFix][The Rock] Fix get_ssm_device_name to return sanitized, usable filename (#47781) rasmith 2026-07-08 07:12:54 -05:00
• 2cae98dfa5 [Rust Frontend] Handle continue_final_message with renderer sentinel (#47844) Bugen Zhao 2026-07-08 19:57:04 +08:00
• db39d60010 Add tuned selective_state_update float32 config for AMD Instinct MI355 (#47943) vanshbhatia-amdandvanshbhatia-amd 2026-07-08 07:41:26 -04:00
• a1ab51afb6 [Bugfix] Allocate HY V3 expert_bias in float32 to prevent silent downcasting (#47797) 4 people 2026-07-08 19:25:53 +08:00
• e7b3853bac Remove router weight upcast for DSv2-related models (#47970) Thien Tran 2026-07-08 19:19:10 +08:00