Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• d940607629 [Core] Support min_tokens with speculative decoding (#32642) Yiliu Dongandqianlihuang 2026-02-27 01:31:28 +08:00
• 99c7892c5b [Perf] Optimize maxsim scores computation for pooling models, 13.9% E2E throughput improvement (#35330) Wentao Ye 2026-02-26 12:14:54 -05:00
• ec8f943db1 Add GlmOcrConfig for GLM-OCR model type recognition (#34982) hujia177 2026-02-26 09:04:42 -08:00
• f2ad952f40 [BugFix][kv_offload]: Fix kernel block size detection (#35125) Or Ozeri 2026-02-26 18:29:34 +02:00
• 9e2cabdf9c [ROCm] Update the torch version in rocm_build.txt to use the official 2.10 release (#34387) Sage Moore 2026-02-26 08:28:45 -08:00
• ec8ab9d254 [ROCm] Add dynamic mxfp4 quantization for DeepSeek V2 projection layers (#34157) 4 people 2026-02-26 10:00:49 -06:00
• 05972ea7e5 [Refactor] Remove dead or duplicate func utils or variables (#35318) Wentao Ye 2026-02-26 10:57:56 -05:00
• 111d869069 [Model] Add nvidia/llama-nemotron-embed-vl-1b-v2 multimodal embedding model (#35297) Jakub Zakrzewski 2026-02-26 15:17:17 +01:00
• 7fea7250a4 [Bug] Fix missing <think> tag after tool call in MiniMax 2.1 (#35352) stingoChenand冬马 2026-02-26 22:11:07 +08:00
• 845ee348ef [Misc] Standardize handling of mm_processor_kwargs.size (#35284) Cyrus Leung 2026-02-26 21:05:46 +08:00
• ec13e549d3 [Bugfix] Fix uint32 overflow in Mamba selective scan state pointer arithmetic (#35275) Asaf Gardin 2026-02-26 14:22:06 +02:00
• c6ca51598a [Bugfix] fix device_name for routing replay (#34336) Li-Yongwen 2026-02-26 20:18:38 +08:00
• c0615a296d [Bugfix] Fix Qwen2.5-Omni and Qwen3-Omni mixed-modality embed regression (#35368) Yueqian Lin 2026-02-26 06:58:23 -05:00
• 01914445b0 Remove bc-lint (#35274) Harry Mellor 2026-02-26 11:01:01 +00:00
• 5281713e11 [XPU] use fixed UMD version in dockerfile.xpu (#35392) Kunshang Ji 2026-02-26 18:54:55 +08:00
• 32693db8ce [Bugfix] [Qwen3.5]Fix Qwen3.5 FP8 quantization: tuple shard_id weight loading (#35289) HZYandIsotr0py 2026-02-26 18:26:15 +08:00
• e03ddcfbd4 [Hardware][Powerpc]Enable prefix caching and chunked prefill for ppc64le (#35081) Akash kaothalkarandAkash kaothalkar 2026-02-26 15:51:24 +05:30
• 02acd16861 [Benchmarks] Plot benchmark timeline and requests statistics (#35220) Sophie du CouédicandCyrus Leung 2026-02-26 11:17:43 +01:00
• ab87f85231 [Model] Ring 2.5 (#35102) Jiangyun Zhu 2026-02-26 18:17:11 +08:00
• 3827c8c55a [Test] Add tests for n parameter in chat completions API (#35283) Krish Gupta 2026-02-26 14:44:07 +05:30
• ade81f17fe [Bugfix][Hardware][AMD] Gate FP4 ops on gfx950 to prevent MI300X crash (#35250) Kevin McKay 2026-02-26 02:11:07 -06:00
• 6042e66cd5 [ROCm] Add extra step in config initialization to populate custom ops before compilation config init (#34848) Gregory Shtrasberg 2026-02-26 02:05:40 -06:00
• 9f9a675b23 [XPU][8/N] Fix kernel bugs in XPU LoRA and MOE LORA (#34115) Chaojun ZhangandKunshang Ji 2026-02-26 15:46:44 +08:00
• a07c4c5939 [BugFix][XPU] Fix speculative decoding on Intel XPU due to bug with IGC_ForceOCLSIMDWidth=16 (#35298) Ofir ZafrirandKunshang Ji 2026-02-26 09:15:16 +02:00
• d3a51da92a [Benchmark] Simplify SLA scan (#35306) Cyrus Leung 2026-02-26 14:35:41 +08:00
• 186ea22efe [Misc][Harmony] Move Responses API only harmony utils to responses/harmony.py (#35339) Flora Feng 2026-02-26 01:35:16 -05:00
• 4a9c07a0a2 [BugFix] anthropic/serving_messages: fix tool call arguments streaming (#34887) DanieleandNicolò Lucchesi 2026-02-26 06:39:48 +01:00
• 9d37941017 [torch.compile] Sequence Parallelism threshold compile ranges (#28672) 3 people 2026-02-25 21:00:12 -08:00
• 4171ff6dd9 [CPU][Feat] Enable KleidiAI INT8_W4A8 for all input dtypes (#34890) Fadi ArafehandTyler Michael Smith 2026-02-26 05:00:10 +00:00
• 13025e71e8 [Model Runner V2] Add coding style guide (#35325) Woosuk Kwon 2026-02-25 20:42:40 -08:00
• 71dfce6aa6 [Kernel] Refactor FlashInfer allreduce for mnnvl backend (#34109) 3 people 2026-02-25 19:17:20 -08:00
• 2aa4140402 openpangu-vl support video input (#34134) 3 people 2026-02-26 11:08:09 +08:00
• 86c3b5a808 [BugFix] Fix fp4 quant kernel on CUDA 12.8 (#35210) Roberto L. Castro 2026-02-26 03:32:50 +01:00
• 160424a937 [Bugfix] Fix CUDA compatibility path setting for both datacenter and consumer NVIDIA GPUs (#33992) 3 people 2026-02-26 11:15:51 +09:00
• 9511a3f8ee [Bugfix] Fix AttributeError in SMControlContextManager (#35338) Lucas Wilkinson 2026-02-25 21:01:10 -05:00
• de527e1cec [UX] Add --moe-backend arg for explicit kernel selection (#33807) Michael GoinandRobert Shaw 2026-02-25 20:44:44 -05:00
• 1976356ee6 [MoE Refactor] MXFP4 Cutlass Experts to MK (#34542) Yongye Zhu 2026-02-25 17:32:39 -08:00
• cbf8f7028c [UX] Add --performance-mode {balanced,interactivity,throughput} (#34936) Michael Goin 2026-02-25 20:28:31 -05:00
• 6831650c40 [offloader] v2: Hide weight onloading latency via prefetching (#29941) Ming YangandMichael Goin 2026-02-25 17:20:59 -08:00
• ed42507f6d [ROCm][CI] Amending deletion of AMD mirror (#35322) Andreas Karatzas 2026-02-25 16:17:56 -06:00
• 9571e99945 [ROCm][CI] Extending attention backend coverage for Eagle spec decode tests (#35265) Andreas Karatzas 2026-02-25 16:16:18 -06:00
• c97234c08b fix(mxfp4): Disable monolithic path for TRITON backend with EP (#34270) 3 people 2026-02-25 15:33:42 -06:00
• b188bab441 [CI][AMD][BugFix] Add torch.cuda.set_device to test_punica_ops so punica kernels execute on same device as tensor (#34985) rasmith 2026-02-25 13:18:00 -06:00
• 15d76f74e2 Revert "[Misc] Enable weights loading tracking for quantized models" (#35309) Lucas Wilkinson 2026-02-25 12:20:15 -05:00
• 8fd6975479 [ROCm][CI] Disable skinny GEMMs in multimodal tests to fix non-deterministic results (#35049) Andreas Karatzas 2026-02-25 10:48:37 -06:00
• 5d18bf8b32 [Bugfix] Fix Harmony preamble visibility in Responses API (#32114) pushkar 2026-02-25 21:38:16 +05:30
• 0788ff0a15 [Bugfix] Gracefully disable AllReduceFusionPass on GPUs without multicast support (#35085) haosdent 2026-02-25 23:31:45 +08:00
• d72b0be33c [XPU]Fix for Qwen-OMNI crash (#35249) Chendi.Xue 2026-02-25 09:31:07 -06:00
• 42489e43c2 [Misc][LoRA] Increase max vocab size limit to 258048 in logits processor (#34773) Bhoomit 2026-02-25 07:30:55 -08:00
• af5e6afa0a [Bugfix] Fix step3p5 reasoning with interleaved thinking (#34211) Mario HongandChauncey 2026-02-25 23:13:01 +08:00
• ee59a7c615 [Tests] Add GSM8k check to SpecDec E2E tests (#34772) Benjamin Chislett 2026-02-25 07:51:14 -05:00
• 709eadbb0b Doc link typo (#35281) Joao Ganteandgemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> 2026-02-25 11:00:31 +00:00
• 90fc7f9109 Fix custom processors that use deleted behaviour for Transformers v5 (#35107) Harry Mellor 2026-02-25 10:36:21 +00:00
• 675ec59aa9 [Bugfix][CPU] Fix basic unit tests failing in CPU platforms (#34677) Yanwen LinandHarry Mellor 2026-02-25 00:36:15 -08:00
• 80e60a6133 [Doc] Suggest "--managed-python" flag when installing python using uv (#33069) Yanwen Lin 2026-02-25 00:19:43 -08:00
• 26e722f906 [DOC][BugFix] Specfiy build dependency installation (#34513) jonoillarandJon OILLARBURU 2026-02-25 09:04:06 +01:00
• 2c619e5e3f [Docs]Fix documentation formatting in architecture overview (#34679) lichuangandHarry Mellor 2026-02-25 16:00:15 +08:00
• 8a685be8d9 docs: document committer proposal process in governance (#35225) Simon MoandCursor 2026-02-24 23:58:48 -08:00
• 2465071510 [Perf] Add opt-in SM100 Oink RMSNorm custom-op path (#31828) Laura WangandLu Fang 2026-02-24 23:01:53 -08:00
• cd43673668 [Perf] Optimize FP8 gemm of sm120. (#34424) wenshuai 2026-02-25 14:25:24 +08:00
• 35d44b4557 [XPU]Support CUDAGraph on XPU Platform (#34482) 4 people 2026-02-25 14:22:52 +08:00
• 8ad54a991b [Platform] Add current_platform.num_compute_units interface (#35042) Kunshang Ji 2026-02-25 14:22:49 +08:00
• 92510edc32 remove cuda check in top_k_top_p_triton kernel (#35011) Kunshang Ji 2026-02-25 14:22:31 +08:00
• a6c137521c [Misc] Add shard_id validation for MergedColumnLinear (#35055) Isotr0py 2026-02-25 14:12:28 +08:00
• 4572a06afe [Misc] Enable weights loading tracking for quantized models (#35074) Isotr0py 2026-02-25 14:11:03 +08:00
• 5cc29cfb8b [compile] Improve error message during artifacts load failure. (#35115) Zhengxu Chen 2026-02-25 01:01:09 -05:00
• 8fae54faff [Linear Attention] fix bug for linear attention + prefix caching + reset_prefix_cache (#35157) Chen Zhang 2026-02-24 22:00:19 -08:00
• f7967577f5 Remove requirement to use --hf-overrides for DeepseekVLV2ForCausalLM (#35203) Harry Mellor 2026-02-25 06:00:06 +00:00
• af770b8e7b [Bugfix] Fix AttributeError when passing StructuredOutputsParams to CompletionRequest (#35237) pks 2026-02-25 07:00:03 +01:00
• 2ff3e436ad [Responses][CI] Filter negative token IDs in schema fuzz test to avoid 500 errors (#35231) Andreas Karatzas 2026-02-24 23:52:44 -06:00
• c2c4c4611a [FIX] fused moe with lora shared expert dual stream (1.07x otps) (#34933) Jhao-Ting ChenandMichael Goin 2026-02-24 20:40:45 -08:00
• f38f8c9742 [ROCm]: Enable customop and rope+kvcache fusion for AITER RoPE (#35180) Rohan Potdar 2026-02-24 22:36:40 -06:00
• ec1d30c0f6 [Responses] Decouple SSE event helpers from Harmony context (#35148) Flora Feng 2026-02-24 23:05:25 -05:00
• e3b2324ec4 [Frontend] Use init_app_state and FrontendArgs in run_batch (#32967) Pooya DavoodiandCyrus Leung 2026-02-24 19:40:39 -08:00
• dbf0da817a [Core] Cleanup engine pause/sleep logic (#34528) Nick Hill 2026-02-24 19:33:34 -08:00
• 3bbb2046ff [Bugfix] Fix expert_ids padding values in moe_align_block_size kernel (#35161) Xin Yang 2026-02-24 17:14:24 -08:00
• 576fe50333 Adding Nemotron fp8 Triton MoE Config (#34674) yugong333andMichael Goin 2026-02-24 15:56:38 -08:00
• a0e50a4260 Convert wvSplitKQ to 16x16 MFMA in prep for mi4xx. (#34100) Hashem Hashemi 2026-02-24 15:35:21 -08:00
• 9fa5b25a23 [Bug][DSV3.2] Always prepare metadata for DeepGEMM Sparse Attention (#35075) Benjamin Chislett 2026-02-24 17:55:22 -05:00
• ea97750414 [CI] Fix Distributed Tests (#35236) Robert Shaw 2026-02-24 17:31:56 -05:00
• 067c5d9ad1 [ROCm][CI] Added MI325 mirrors (#34923) Andreas Karatzas 2026-02-24 15:37:15 -06:00
• f5972a872f [Model][Spec Decode] Nemotron-H MTP and Mamba Speculative Decoding Support (#33726) 4 people 2026-02-24 12:49:56 -05:00
• a9e15e040d Add @MatthewBonanni to CODEOWNERS (#35207) Matthew Bonanni 2026-02-24 12:45:10 -05:00
• 542ca66357 Revert "[CI/Build] Remove redundant OpenTelemetry pip install from CI configs" (#35211) Lucas Wilkinson 2026-02-24 12:26:42 -05:00
• fc8456c336 [CI/Build] Fix kernels test location (#35205) Cyrus Leung 2026-02-25 01:20:34 +08:00
• 9ce8fad2a9 [Perf] Optimize Python Slice for Structured Output using islice instead of [:] (#33593) Wentao Yeandgemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> 2026-02-24 12:02:36 -05:00
• c38b8d5a31 Remove padding_index from models that don't use it for better Transformers v5 compatibility (#35189) Harry Mellor 2026-02-24 16:04:46 +00:00
• 60da0e1544 [CI] Remove Duplicated Tests (#35199) Robert ShawandRobert Shaw 2026-02-24 10:53:30 -05:00
• 9609b1f18d Integrate flashinfer mm_mxfp8 in ModelOpt MXFP8 (#35053) danisereb 2026-02-24 17:45:13 +02:00
• a0c7081695 Fix fallback to default tactic (flashinfer autotuner) with trtllm_fp4_block_scale_moe (#35088) danisereb 2026-02-24 17:25:44 +02:00
• 34ce0ffd1f [CPU][Perf] Accelerate Attention head for s390x using vector intrinsics (#34434) R3hankhanandLi, Jiang <jiang1.li@intel.com> 2026-02-24 20:55:39 +05:30
• 0de5333989 Fix GLM4 parser tests (#34905) Robin NabelandChauncey 2026-02-24 14:27:42 +00:00
• a87cc50859 [Attn,KV-cache] Use per-head scales in the attention selector (#34281) 3 people 2026-02-24 15:02:43 +01:00
• 761e63e541 [Frontend] Always pass supported_tasks to validation (#35186) Cyrus Leung 2026-02-24 20:16:33 +08:00
• d12d201409 [Bugfix] Fix failing FunASR processor test (#35111) Isotr0py 2026-02-24 20:13:45 +08:00
• b3ad37c5db [glm-asr] change defaults dummy audio size (#35108) eustlb 2026-02-24 13:13:33 +01:00
• 14561fabfd [Perf] Optimize pooling model redundant copy, 1.8% throughput improvement (#35127) Wentao Ye 2026-02-24 07:13:11 -05:00
• c77f3e1207 [compile] Save aot compile artifacts atomically. (#35117) Zhengxu Chen 2026-02-24 07:11:01 -05:00
• 012dee9233 [Feature] Add LoRA tower/connector support for Llama 4 Vision (mllama4) (#35147) Dor HuriandJee Jee Li 2026-02-24 14:10:32 +02:00
• f1c664545b Make voxtral compile friendly (#33959) Tugsbayasgalan ManlaibaatarandNicolò Lucchesi 2026-02-24 16:33:35 +08:00