Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• b3082d3044 [Rust Frontend] Remove the PyO3 tool-parser bridge (#59744) Bugen ZhaoandClaude Opus 5.5 2026-10-06 13:46:27 +08:00
• 5145e35e50 [Frontend] Port the MiniMax M3 tool parser to the parser engine (#59743) Bugen ZhaoandClaude Opus 5.5 2026-10-06 13:46:27 +08:00
• 6fddc9b21e [Profiler] Add per-session profiling controls to Python and Rust frontends (#57875) czhu-cohereandaoshen02 2026-10-05 22:21:19 -07:00
• 9a1f6fd4cb [Model][DeepSeek-V4.1] Permute mega-attention wq_b / wo_a while loading (#60064) aoshen02andClaude Opus 5.5 2026-10-06 13:20:09 +08:00
• 533d81b4b0 [ZenCPU][Test] Add zentorch/torch version pin consistency test (#52888) Chinmay-Kulkarni-AMDandCursor Agent 2026-10-06 10:42:19 +05:30
• eb2901b735 [Perf][DSv4.1] Run decoder replay layers in CUDA graphs and trim in PIECEWISE graphs (#59532) 3 people 2026-10-05 22:11:45 -07:00
• 21d93d0d8c [Attention][Model] FP8 KV cache for Triton DiffKV and MiMo-V2.6-Flash (#58128) 4 people 2026-10-06 06:18:29 +02:00
• 30ab428235 [Bugfix][Attention] Restrict FlashInfer sparse MLA FULL graphs to decode (#59751) Tianyu Guo 2026-10-06 12:13:23 +08:00
• a3e0243b1c [Bugfix][Parser] Mistral pre-v11: don't fail requests on unexpected tool call JSON (#54844) lzhan011andFlora Feng 2026-10-05 22:37:30 -05:00
• 8c417a2dae [Bugfix][Model] Fix Qwen3-Omni interleaved M-RoPE boundary positions (#59842) Tianyu GuoandOpenAI Codex 2026-10-06 11:35:53 +08:00
• 48eb44b719 [Bugfix][KVConnector][NIXL] Fix TP handshake test after per-region FA desc flags (#60157) Tianyu Guo 2026-10-06 11:19:41 +08:00
• 6fa67afed9 [Bugfix][Core] Retire stale Mamba spec blocks in prefill checkpoint steps (#59759) 3 people 2026-10-06 05:01:19 +02:00
• 2b32d9ac6a [Bugfix][Core] Drop stale block hashes when a session truncates tokens (#59103) 4 people 2026-10-05 19:51:03 -07:00
• b0b23a48d4 [Bugfix] Fix Qwen4Exp FP8 PLE pinned lookup kernel failing to compile below SM89 (#60142) 3 people 2026-10-06 05:56:44 +04:00
• 9465e1c783 [Bugfix][NIXL] Recover locally invalidated pull peer metadata (#55471) 5 people 2026-10-06 09:46:53 +08:00
• 9bf4e51030 [Bugfix][Frontend] Malformed $defs in a tool schema should be a 400, not a 500 (#54850) lzhan011andFlora Feng 2026-10-05 20:37:20 -05:00
• efd001756b [ROCm] Disable AITER attention query quantization on RDNA3 (#57750) amd-xavierwangandTJian 2026-10-05 21:37:18 -04:00
• 2537629eb2 [Bugfix][KVConnector][NIXL] Size per-region replicate flags by each region's block count (#60107) 3 people 2026-10-05 21:27:50 -04:00
• 0e8f8ecba8 [Bugfix][KVConnector][NIXL] Return NumPy descriptors from mixed-memory local registration (#60108) 3 people 2026-10-05 21:23:20 -04:00
• 4e198cef83 [TEST][XPU][CI] Fix ci uva kernel test instability (#60041) 3 people 2026-10-06 03:19:01 +02:00
• 97bd8c74eb [ROCm][DSv4.1][Perf] One-pass dequantize and a gather-sized grid for the compressed K cache (#56720) 3 people 2026-10-05 17:57:46 -07:00
• b2863c0de9 [Kimi-K3] Use the fused K/V pack kernel on the decode-context-parallel prefill path (#56881) 3 people 2026-10-05 17:13:55 -07:00
• 5cd8a2626b [Bugfix][Structured Output] Cap JSON schema nesting to prevent 500s and API-server stalls (#60036) Mai Hoàng ĐứcandFlora Feng 2026-10-06 07:06:59 +07:00
• fa423b614c [CI][ROCm] Wait for engine teardown between LM Eval models (#60100) Aarushi JainandAndreas Karatzas 2026-10-05 18:39:59 -05:00
• 6dd813e0b7 [Bugfix][MoE] Allow MiniMax2 routing in TRTLLM BF16 monolithic backend (#59571) MasonandYongye Zhu 2026-10-06 07:24:14 +08:00
• 0eac152707 [Bugfix] Route every speculation-capable row through the speculative path so recurrent state stays consistent (#56531) 3 people 2026-10-06 03:03:44 +04:00
• 4a705c7b54 [Bugfix][Frontend] Preserve logprobs when top_logprobs is null (#47838) Ting SUNandFlora Feng 2026-10-06 00:43:31 +02:00
• 89a00995c9 [CI] Read each profiler round's trace as it stops in test_gpu_profiler (#59769) Kevin H. LuuandClaude Opus 5.5 2026-10-05 15:25:11 -07:00
• 2e06395fd3 [Bugfix][Frontend] Build tool-call grammars from the tools the prompt renders (#59879) RomeJayandFlora Feng 2026-10-06 06:01:03 +08:00
• 3fb1cd796b [Frontend] Extract parse and assembly out of batch chat derender (#60073) Martin HickeyandFlora Feng 2026-10-05 22:57:45 +01:00
• 2e3154aa18 [HiSparse] Log steady-state max concurrency and expose host-tier utilization gauges (#58949) 3 people 2026-10-05 17:20:32 -04:00
• 8542c3eff5 [CI] Fix test_mixed_warmup_gate after #57053 (#60116) Matthew BonanniandClaude Opus 5.5 2026-10-05 17:17:00 -04:00
• 18c8a65edd [BugFix][Frontend] Pass reasoning_ended through render → generate (#60059) (#60062) Yida WengandFlora Feng 2026-10-06 04:39:25 +08:00
• 74c5cbcd7b [Bugfix][MRV2][Spec Decode] Honor dynamic K in autoregressive speculators, 1.2~1.3x kernel perf improvement (#57053) Wentao Ye 2026-10-05 14:54:11 -04:00
• c32495d58f [watermark] compatibility validation (#56801) Raphaël RiallandandSimon Veitner 2026-10-05 20:37:13 +02:00
• 88f3c601b3 [Docs] Add reviewer area of interest for TheEpicDolphin (#60092) Giancarlo Delfin 2026-10-05 11:35:19 -07:00
• d09ff77637 [Bugfix] Give each data-parallel engine its own global RNG streams (#59788) aoshen02andClaude Opus 5.5 2026-10-06 01:38:18 +08:00
• 877ddcc456 [Bugfix][Spec Decode] Reserve the bonus KV slot for fill-in DSpark (#59105) 3 people 2026-10-05 19:35:51 +02:00
• 182247257f [ROCm][Perf] Use a stacked hipBLASLt GEMM for the bf16x3 router (#52668) akii96 2026-10-05 20:28:07 +03:00
• 54d93af9fb [Bugfix][HiSparse] Reject cudagraph_mode=FULL at startup (#59688) Matthew BonanniandClaude Opus 5.5 2026-10-05 12:42:27 -04:00
• 64cb683842 [Model] Add Nemotron 3.5 ASR transcription support (#59827) Sohaib Ahmed 2026-10-05 21:28:09 +05:00
• 87954c5b04 [CI/Build] Declare the video modality on the CohereCompass reference model (#60063) stefankoncarevicandAndreas Karatzas 2026-10-05 18:25:42 +02:00
• d23af12cb1 [ROCm][Perf] Reach the fused QSA pre-indexer from the AMD path (#57947) Mikko TukiainenandCursor 2026-10-05 19:22:29 +03:00
• 528772a4bf [Bugfix][Core] Retain encoder cache references for repeated multimodal inputs (#59942) bzsuniandcjackal 2026-10-05 23:52:46 +08:00
• edde9d2ce4 Remove redundant dependency requirement for TPU (#59977) charlesli640 2026-10-05 08:45:37 -07:00
• f2d8fbf459 [CI/Build] Run the prefill token scoring test under batch invariance (#60056) stefankoncarevic 2026-10-05 17:32:33 +02:00
• 9e931a0c6c [CI] Add CODEOWNERS for HiSparse (#60060) Matthew BonanniandClaude Opus 5.5 2026-10-05 11:18:29 -04:00
• 60932a6401 [Bugfix][Structured Output] Preserve literal values in Guidance disable_additional_properties (#58709) 3 people 2026-10-05 15:12:40 +00:00
• 30d4032363 [Model][GLM-5.3-Flash] Support DCP for the kpool sparse indexer (#59211) Nicolò Lucchesi 2026-10-05 16:44:14 +02:00
• f9c9e8ac24 [LoRA] Remove tensorizer (#60024) Jee Jee Li 2026-10-05 22:01:56 +08:00
• d3547f9d03 [Test] Re-enable Voxtral HF reference test on Transformers v5 (#59771) Abhay ShuklaandClaude Opus 5.5 2026-10-05 19:13:49 +05:30
• 0e468adb43 [Model] Extend device-side mm normalization to Kimi K2.5 / K3 (#59278) cjackal 2026-10-05 22:28:13 +09:00
• ff53f32409 [Attention][MLA] Support fp8_ds_mla KV cache for NoPE-512 models on SM90 (#59246) Nicolò LucchesiandLeoyzen 2026-10-05 15:12:00 +02:00
• d0d6e5f3a2 [Bugfix] Fix Mamba page size AssertionError with spec decoding on GraniteMoeHybrid, FalconH1 and Zamba2 (#59975) Vadim GimpelsonandClaude 2026-10-05 16:49:47 +04:00
• 51eeb0c58f [Bugfix] Load stacked expert weights for non-gated MoE (#59031) roikoren755andmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-10-05 15:26:55 +03:00
• 4ff028d77e [Bugfix][Qwen4Exp] Honor --kv-cache-dtype-skip-layers in QSA attention (#60023) Stefano CastagnettaandThien Tran 2026-10-05 14:13:30 +02:00
• 55b80221ed [Perf][Qwen4Exp] Keep the HC up projection on the skinny GEMM path (#60027) Stefano CastagnettaandThien Tran 2026-10-05 14:09:36 +02:00
• 710ac56e69 [Security] Restrict Pillow image formats on untrusted media paths (#60022) Juan Pérez de Algaba 2026-10-05 12:40:15 +02:00
• edca360f13 [Bugfix][Qwen4Exp] Load PLE tables unquantized under Quark checkpoints (#59443) 4 people 2026-10-05 13:40:02 +03:00
• c04c79e50d [Performance] Add SM121 TP=2 skinny-GEMM plans (#59632) 3 people 2026-10-05 15:57:51 +05:30
• b1401e0aa7 [Bugfix][NIXL] Count heartbeats as remote engine activity (#59873) Maroon Ayoub 2026-10-05 12:36:13 +03:00
• ae53b06898 [Bugfix][GLM-5.3-Flash] Keep batch x heads out of gridDim.z in the GLM-5.3-Flash fused recurrent KDA kernel (#56974) 3 people 2026-10-05 17:31:45 +08:00
• b1f229fb75 [Model] Declare SupportsEagle3 on Qwen3ASRForConditionalGeneration (#52824) Ian Eavesandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-10-05 04:23:47 -05:00
• e3af5bf3ce [LoRA] Code cleanup (#60017) Jee Jee Li 2026-10-05 17:15:53 +08:00
• 4a30c4cad0 [Model][DeepSeek-V4] Make MegaMoE shared-expert finalize independent of linear post-load order (#59927) aoshen02andClaude Opus 5.5 2026-10-05 16:39:49 +08:00
• 4a3ebc6b5b Merge branch 'main' into patches MOVZX 2026-10-05 08:21:46 +00:00
• 34051ad714 feat: Add video support for the Transformers backend (#57441) 3 people 2026-10-05 11:51:15 +04:00
• 042ab0305c [Perf][Qwen4Exp] Merge QSA QKVG and indexer QK projections (#59533) 3 people 2026-10-05 15:03:29 +08:00
• 92044241a0 [Bugfix][ROCm] Preserve config during GPU memory profiling (#58014) Thiago Mota MartinsandFangzhou Ai 2026-10-05 02:28:33 -03:00
• e8a53a55e6 [Bugfix][Attention] Support int4_per_token_head on non-power-of-two head sizes (#56198) 4 people 2026-10-05 12:52:39 +08:00
• ac8c2130a8 [ROCm] Bump AITER to 0.1.24.post1 (#59794) Micah WilliamsonandFangzhou Ai 2026-10-04 23:19:19 -05:00
• 0c16eee3f1 [Bugfix][KV Offload] Reset SimpleCPU eager-store placement state on MRV2 resume (#57816) Kam BasraandOpenAI Codex 2026-10-05 09:00:07 +07:00
• f83ecfac16 Merge remote-tracking branch 'upstream/main' into patches MOVZX 2026-10-05 08:52:31 +07:00
• 4f52fa35ef [Bugfix] Fix Qwen4Exp PLE embedding rejecting INC (AutoRound) checkpoints (#59990) Vadim GimpelsonandClaude 2026-10-05 05:52:06 +04:00
• 49e0f47978 [Bugfix][DSv4.1] Keep the compressor ring out of the null block (#58560) 5 people 2026-10-05 03:49:41 +02:00
• 1388100560 [Kernel][LoRA] Add deterministic split-K=8 LoRA shrink for batch invariance (#59377) 3 people 2026-10-04 15:46:16 -07:00
• 7867d6c52d [Frontend] Name the served models in the model-not-found 404 (#59889) TyroneandHarry Mellor 2026-10-04 22:45:51 +02:00
• 155488d853 [Bugfix][Determinism] Preserve output dtype in batch-invariant mean (#59106) 3 people 2026-10-04 13:07:46 -04:00
• b0eb87fe49 [Bugfix][Frontend] Return streaming errors before the first token (#40986) Stefan WangandRobert Shaw 2026-10-04 09:42:49 -07:00
• 89439db727 [CI][Docs] Fix decode/prefill consistency test prefix construction; add GLM-4-9B and Phi-4 to batch-invariant tested models (#53692) 5 people 2026-10-04 23:55:18 +08:00
• bd42276e1e [Misc][Platform] Check aligned KV block sizes against every attention backend (#58457) 3 people 2026-10-04 08:13:20 -07:00
• bdd31c31de [CI] Stop requiring both API servers to record weight sync metrics (#59521) Aarushi JainandAndreas Karatzas 2026-10-04 09:14:27 -05:00
• f98d1fc493 [ROCm][Bugfix] Fix ROCM_ATTN sliding-window boundary (#59550) 00CC 2026-10-04 21:48:46 +08:00
• d64f6cd08a [CI] Auto-label pooling PRs and issues (#59913) Taneem IbrahimandLucas Wilkinson 2026-10-04 08:26:10 -05:00
• d61081dc3d [CI] Bump peft to satisfy transformers 5.18 minimum (#59932) Harry MellorandClaude Opus 5.5 2026-10-04 11:53:07 +01:00
• 155d23cb00 [Bugfix][XPU] Make DeepSeek V4 FP8 sparse decode graph-capturable (#59159) Yongqi WangandKunshang Ji 2026-10-04 18:31:15 +08:00
• 0872ddf4b9 [Bugfix][Multimodal] Normalize single-channel audio to 1D (#56691) 3 people 2026-10-04 14:11:41 +08:00
• 18f8f96025 [Bugfix][KV Offload] Release pending CPU lookup pins on cache reset (#59862) Zupeng Wang 2026-10-04 12:30:17 +08:00
• 6f75cc7d5d [Bugfix][Responses API] Reuse streamed item ids in final harmony response (#59859) Yufeng HeandFlora Feng 2026-10-04 11:01:04 +08:00
• 50b404e71e [Bugfix][Frontend] Preserve abort finish_reason for scale-out token streams (#47933) Ting SUNandFlora Feng 2026-10-04 04:57:26 +02:00
• ce49174247 [Bugfix] Fix FlashInfer all_reduce backend selection (#56891) Elvir Crnčević 2026-10-04 04:18:58 +02:00
• 5e56e9af2a [Feature] Release the CUDA graph pool on sleep (#59160) aoshen02andClaude Opus 5.5 2026-10-04 09:13:10 +08:00
• 9e0b6fcf77 [CI] Bump Transformers version to 5.18.0 (#59621) Harry Mellor 2026-10-04 01:42:35 +01:00
• 84bcbc6264 [DCP] Enable TokenSpeed MLA with block-interleaved DCP (#59462) Wei Zhaoandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-10-03 13:26:07 -04:00
• 4ac0d0eac2 [ROCm][RDNA3] Fix W4A16 split-K accuracy and determinism (#54706) 3 people 2026-10-03 23:44:17 +08:00
• 85c31d47ad Merge remote-tracking branch 'upstream/main' into patches MOVZX 2026-10-03 22:06:10 +07:00
• 2877121b2d Merge remote-tracking branch 'upstream/main' into patches MOVZX 2026-10-03 21:59:03 +07:00
• f03026a548 [Bugfix][MoRIIO] Keep discovery heartbeats running while workers hold the GIL (#59441) Hexiang WangandCodex 2026-10-03 22:55:18 +08:00
• e319f86f15 [Feature] Per-row candidate IDs for prefill token scoring (M2 of #56860) (#56984) 3 people 2026-10-03 22:17:09 +08:00
• bc21cba967 [Frontend] Add output_mode to /inference/v1/generate (RFC #56851 Phase 1) (#58588) Martin Hickey 2026-10-03 12:38:25 +01:00