Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

main

Mono Color

• 17fa5ad327 [Refactor] Generalize attention graph variants in the decode runner (#38993) Liangsheng Yin 2026-09-11 00:45:58 -07:00
• f618022b73 [AMD][CI] Temporarily pause MI355X disaggregated nightly (#39044) YC Yen-Ching TsengandChen 2026-09-11 09:19:24 +02:00
• 69aa46b2fa [ROCm] Raise HiCache JIT block quota for mapped-host throughput (#39036) AMD-yanfeiwang 2026-09-11 15:12:24 +08:00
• a713349faf [HiCache] fix: preserve SWA host lock boundaries across splits (#38835) Shuwen Wang 2026-09-11 14:58:26 +08:00
• 76ef679cb1 [AMD] aiter: fail loudly on cross-layer KV sharing in target_verify (#38755) Vignesh Sethuraman 2026-09-10 23:37:03 -07:00
• 2465ee3948 [AMD] Fix DeepSeek block-FP8 loading on gfx94x (#38446) Bingxu Chen 2026-09-11 14:00:29 +08:00
• b9899b04c1 [AMD] Add gfx1151 (Strix Halo / Ryzen AI MAX+) Docker image (#33939) Alex NailsandHubert Lu 2026-09-10 22:46:33 -07:00
• 6182d45524 [Test] Fix Q8KV8 sparse-prefill pool fixture after page-size rename (#39019) Jialin Ouyang 2026-09-10 22:22:35 -07:00
• 2018c64e22 [PD] Add is_dummy truth-table wire tests for mooncake and nixl (#34977) Rishabh Sinha 2026-09-10 22:14:41 -07:00
• 40a84d6dfc [kv-shard 1/4] Logical-page placement with UnifiedRadixCache (#38356) Shunkangz 2026-09-11 13:06:42 +08:00
• 94ce940ff8 [Bug] Include DSA variant in exact-bucket graph admission (#38992) Liangsheng Yin 2026-09-10 21:03:03 -07:00
• b3dc0388ed test(lora): enable ROCm logprob accuracy coverage (#38791) Yikai Zhang 2026-09-10 20:29:06 -07:00
• 480b14edad [xpu] install xpu-kernel by released wheel (#37394) yuchengliu1andMa Mingfei 2026-09-11 10:47:27 +08:00
• 0fadad8933 DFLASH support added for XPU (#32798) ANSHUMAN TRIPATHYandMa Mingfei 2026-09-11 08:09:37 +05:30
• 690428b470 [XPU] Add xpu forward in Gemma3RMSNorm & Add test and benchmark for Gemma3RMSNorm (#36278) Miaomiao Jiang 2026-09-11 10:32:25 +08:00
• 7e3ae15f73 [AMD] aiter: honor per-layer softmax scale (#38754) Vignesh Sethuraman 2026-09-10 19:27:09 -07:00
• 3716e496ea Speculative Decoding support for intel_xpu attention backend on XPU target (#30548) ANSHUMAN TRIPATHY 2026-09-11 07:55:25 +05:30
• 2adb2e8485 [XPU] Adapt device agnostic API usage (#32093) ANSHUMAN TRIPATHYandmingfeima 2026-09-11 07:33:12 +05:30
• 67d3a2ea57 [XPU] Make checkpoint_engine worker device-agnostic (#32382) Siju Samuel 2026-09-11 07:26:39 +05:30
• ad7f57c9ea [Responses] Fix empty-prompt routing for token-first chat encoders (kimi_k3, inkling) (#35486) 3 people 2026-09-11 09:06:20 +08:00
• 9df72e8f5a Fix custom logit processor params when num_tokens_in_batch is used (#38730) jasonjk-park 2026-09-10 17:52:08 -07:00
• d0035da34e [Diffusion][Refactor] Refactor and unify RoPE execution for DiT models using RotaryEmbedding based on CustomOp (#33555) Alexandrandronnie_zheng 2026-09-11 03:31:05 +03:00
• 41da06adca [Refactor] Generalize DeepSeek V4 compressed pool management (#38954) Liangsheng Yin 2026-09-10 17:29:03 -07:00
• d006f40e24 [AMD][CI] Retire the ROCm 7.0 kernel wheel (#38767) 3 people 2026-09-11 08:25:55 +08:00
• 00143e9c23 Fix dataclasses.asdict on the msgspec ServerArgs (#38958) Alison ShaoandClaude Opus 5 2026-09-10 17:22:54 -07:00
• 52c191da52 [Deps] Retire the CUDA 12 lane (#38404) Mohammad Miadh AngkadandMohammad Angkad 2026-09-11 07:58:09 +08:00
• fae8cd84cb Support NemotronH_Omni_Reasoning_V3 in SGLang (#35599) 3 people 2026-09-10 19:57:22 -04:00
• 203d7e812c Fix KV-canary workspace accounting after graph capture (#38596) cctry 2026-09-10 16:10:12 -07:00
• 42bbaac259 [metrics] Report logical prefill token counts (#38566) cctry 2026-09-10 16:09:21 -07:00
• dc5f59c3a2 [Refactor] Clarify DeepSeek V4 metadata names for V4.1 (#38947) Liangsheng Yin 2026-09-10 15:42:52 -07:00
• d076eec427 [SM120] Use exact query-head widths for DeepSeek-V4 sparse MLA decode (#36655) Pengyun Lin 2026-09-11 06:40:17 +08:00
• bb15be6d79 [Spec] Stage Inkling MTP draft metadata before verify (#38169) paulzhang-tmandQiaolin-Yu 2026-09-10 18:32:19 -04:00
• a63efd9056 [Spec] Support large MTP batches in short-convolution metadata (#38558) paulzhang-tm 2026-09-10 18:13:17 -04:00
• fd7743e0e1 [Sampling] Support sampling masks with overlap scheduling (#36631) 4 people 2026-09-10 14:52:44 -07:00
• 55b45cb45a fix(qsa): dequantize FP8 cached prefixes in the sparse prefill kernels (#38855) YAMY 2026-09-10 14:07:39 -07:00
• 06dfe05d65 [Deps] Raise smg-grpc-servicer floor to >=0.9.0 to unbreak SMG E2E CI (#38801) 3 people 2026-09-10 13:55:36 -07:00
• a26273d668 [AMD] Quantize the bf16 MTP draft experts online to MXFP4 for Qwen3.5 (#38748) 3 people 2026-09-10 13:00:41 -07:00
• 887c401e15 docs: sync LMSYS SGLang blog cards (#36773) sglang-botandsglang-bot 2026-09-10 11:37:14 -07:00
• 4b7331fb77 Make the remaining DeepSeek-V4.1 NVIDIA cells start (#38861) zijiexiaandClaude Opus 5 2026-09-10 11:04:53 -07:00
• 209654c420 [NPU][Hicache] Optimize HiCache L2 IO with Memfabric acc_offload (#38826) James 2026-09-11 00:23:36 +08:00
• 19b83b4b9d [Radix Cache] Fix PureSWA tail release without insertion (#38349) Shuwen Wang 2026-09-10 23:55:48 +08:00
• cc7e43ad22 [HiCache] Account for newly pinned ancestors in load-back quota (#38481) Shuwen Wang 2026-09-10 23:54:08 +08:00
• 92dffebe16 [NPU] Set DEEPEP_HYBRID_DEPLOYMENT for new DeepEP tests; switch glm5_2 to w8a8; tune nightly timeouts (#38775) pllimax 2026-09-10 23:32:54 +08:00
• 12771786f2 [AMD] Restore AITER verify runtime sizing reverted by #34647 (#38575) Chengze Fan 2026-09-10 05:57:47 -07:00
• 613002281e [CI] Drop the GPTQ dynamic-config test for the deleted non-Marlin kernel (#38881) Mohammad Miadh AngkadandMohammad Angkad 2026-09-10 20:55:06 +08:00
• 908226fea2 [Rust TreeCore] Support external cache linker (#37306) Jialin Ouyang 2026-09-10 04:22:24 -07:00
• 334e94d8ac [NPU] fix pp 2 hang on npu (#38249) liupeng374 2026-09-10 19:21:37 +08:00
• a37ded1693 [Cookbook] DeepSeek-V4.1: add the HiCache L2 knob to the Playground (#38844) Yuhao YangandClaude Code 2026-09-10 18:20:34 +08:00
• dc2157dcd6 [JIT] Port the expert-pack MXFP4 kernels to load_jit and fix their launch limits (#38830) Mohammad Miadh AngkadandMohammad Angkad 2026-09-10 17:50:51 +08:00
• 8022505705 Revert "[CI] Temporarily disable GB300 tests" (#38842) Mohammad Miadh Angkad 2026-09-10 17:41:15 +08:00
• c9c26d56b2 [diffusion] docs: sync snapshot and minimax-h3 subblock features (#38784) MickandMick Qian 2026-09-10 17:23:43 +08:00
• 8a6ab89bf0 [AMD] Enable Fast Triton Sparse MLA backend (#30575) 3 people 2026-09-10 03:58:00 -05:00
• 5caafd2118 Fix the DeepSeek-V4.1 reasoning example and mark the B300 cells verified (#38839) zijiexiaandClaude Opus 5 2026-09-10 01:49:44 -07:00
• fa6e657b93 [misc] Update CI permission (#38834) Baizhou Zhang 2026-09-10 01:38:58 -07:00
• b23db06769 pyproject(xpu): drop human-eval git dep to unblock image build (#38824) ashwini rathi 2026-09-10 13:54:43 +05:30
• 55b4f4f195 [Test] Add offline Transformers loader compatibility checks (#38336) 3 people 2026-09-10 13:37:43 +05:30
• 9a2f17f41d Add INT4 and FP4 lanes to the Ling-3.0-flash-VL cookbook (#38527) Xinyuan Tong 2026-09-10 15:23:50 +08:00
• 1b77f498a0 [NVIDIA] Support flashinfer Mega Moe (#31470) 5 people 2026-09-10 02:22:47 -05:00
• c0b790cf7f Delete cutlass_mla, non-Marlin GPTQ, AWQ AOT kernel, and Dual Chunk Flash Attention (#32114) Brayden ZhongandBrayden Zhong 2026-09-10 03:12:01 -04:00
• 7152c14384 Fix DeepSeek-V4 routing: sqrtsoftplus underflow and unfloored renorm (#34459) Brayden ZhongandBrayden Zhong 2026-09-10 03:10:50 -04:00
• 6b2e13bcb0 [AMD] Fix the diffusion perf fixture lookup (#38686) YC Yen-Ching Tseng 2026-09-10 09:07:08 +02:00
• bd922bc39b [AMD][CI] Fix UMBP test buffer after MoRI upgrade (#38672) Bingxu Chen 2026-09-10 15:02:00 +08:00
• 13fb796c81 [AMD][CI] Drop the dead miles ROCm 7.0 nightly image build (#38694) 3 people 2026-09-10 15:00:28 +08:00
• 026e61bd94 [AMD] Restore AMD CI registrations dropped by #37436 (#38581) Michael 2026-09-10 14:59:10 +08:00
• f3152439d0 [AMD][CI] Publish ROCm 10 release images and kernel wheel (#38763) Michael 2026-09-10 14:57:00 +08:00
• a1ec35a330 docker(xpu): install libssl-dev so JIT hicache_hash_cpp builds (#38796) ashwini rathi 2026-09-10 12:07:20 +05:30
• 058cca0192 [HiCache] Fix side pools to use resolved host allocator (#38350) Shuwen Wang 2026-09-10 14:28:20 +08:00
• 69777c4d36 Add DeepSeek-V4.1 Flash cookbook (#38802) zijiexiaandClaude Opus 5 2026-09-09 23:15:30 -07:00
• 03e4c06589 Fix Qwen3.5 GDN multi-item scoring (#33922) DAI0818andPo-Han Huang 2026-09-10 13:30:39 +08:00
• 6f481ad0e3 [router] Raise chat body cap to 32 MB for multimodal payloads (#38735) Kan WuandClaude Fable 5 2026-09-09 22:14:49 -07:00
• 01ad27ecd9 [CPU] Fix native KV hash compilation in Xeon image (#34767) Sage 2026-09-10 07:51:25 +03:00
• 6ab6df54b9 Add missing test dependencies to pyproject.toml variants (#36098) ashwini rathiandarathi-hlab 2026-09-10 10:19:16 +05:30
• c415f977b8 [AMD] Update v4 args for agentic workload (#38677) Thomas Wang 2026-09-10 12:41:05 +08:00
• 2f7393f0d2 [Simulator] Give the OFFLINE/BLOCKING comparison tolerances real headroom (#38732) Alison Shao 2026-09-09 20:10:38 -07:00
• f1a512c51c [Config] msgspec.Struct for the config tier (#38753) Cheng Wan 2026-09-09 19:41:19 -07:00
• 106cc561f9 [AMD] ci: move the miles nightlies from rocm700 to rocm10 (#37495) Xinyu JiangandZhiyao Jiang 2026-09-09 22:37:34 -04:00
• 53dc77ff4e [Config] One writer for the declaration stash; no exception to the write seal (#38752) Cheng Wan 2026-09-09 19:22:06 -07:00
• 9a1b1d2d5e docker(xpu): drop redundant setvars.sh from torch_memory_saver RUN (#38665) ashwini rathiandarathi-hlab 2026-09-10 07:44:50 +05:30
• fd596a474c [XPU][Fix] Pack device-pointer tables as uint64 to avoid 64-bit address overflow (#35051) 3 people 2026-09-10 07:42:18 +05:30
• 3700c4ee26 [CI] Temporarily disable GB300 tests (#38770) Alison Shao 2026-09-09 18:31:00 -07:00
• 7f0285093e [diffusion] feat: spill large tensors over shared memory like numpy arrays (#38656) Mick 2026-09-10 09:14:06 +08:00
• ce555ed82a [diffusion] refactor: refactor utility ownership and document helper placement (#38699) MickandMick Qian 2026-09-10 09:11:01 +08:00
• 3ff226ba8f [NPU]Support GLM5.2 and FP8 DSA&Indexer kvcache for 950 (#38250) LinyuanLi 2026-09-10 09:10:10 +08:00
• a7e00b7576 [CI] Answer unrecognized slash commands instead of skipping silently (#38736) Alison Shao 2026-09-09 18:01:18 -07:00
• 92d831d3d7 [AMD] Parallelize aiter spec-decode KV index building over token blocks (#37659) 3 people 2026-09-09 17:53:53 -07:00
• 880d6fa64d [DSv4] Integrate TRT-LLM DSv4 Attention for SM100/103 (#30805) 3 people 2026-09-09 17:40:07 -07:00
• 0084030179 Add Opt-In for GLM-5.3 Flash breakable prefill CUDA graphs (#38522) William Hu 2026-09-09 20:02:08 -04:00
• 2092f6df05 [CI] Install helion 1.4.0 for the KDA Helion kernel tests (#38688) 3 people 2026-09-10 07:00:17 +08:00
• 51c8581a26 [Rust] Gate health on startup warmup completion (#37994) Po-Han Huang (NVIDIA) 2026-09-10 06:55:13 +08:00
• beaf3d9252 [HiCache] Replace skip_lock_node_ids with a segment lock protocol (#36848) Zhiqiang Xie 2026-09-09 14:55:51 -07:00
• a84ffd1326 feat: add optimized Domino rollout to DFlash V2 (#36899) FrancisandQiaolin-Yu 2026-09-10 05:12:15 +08:00
• 2948a62a6f [CI] Add /run-full-ci and /run-extra-ci slash commands (#38734) Alison Shao 2026-09-09 13:56:38 -07:00
• 96d91ef926 [Model] Support GLM-5.3 Flash NVFP4 loading (#38621) Po-Han Huang (NVIDIA)andYAMY 2026-09-10 04:51:08 +08:00
• bede776c2a fix(unified-memory): evict Full KV for Mamba byte shortfalls (#36713) 3 people 2026-09-10 05:46:51 +09:00
• 6c1d0b1b29 Revert "[Spec] Publish the final multi-layer EAGLE shared-read event" (#38041) paulzhang-tm 2026-09-09 16:01:16 -04:00
• 95a88bfd69 Relax GSM8K thresholds for the GLM-5.2 DSA-MTP variants (#38725) Alison ShaoandClaude Opus 5 2026-09-09 12:46:10 -07:00
• 354ee46c26 Keep VMM capability votes on CPU (#38722) cctry 2026-09-09 11:56:30 -07:00
• 8733da8cf4 [rust-server] Use node-local HTTP ports for DP attention (#34430) Sage 2026-09-09 21:22:44 +03:00
• 7b791c9534 [Bugfix] Keep a shared MAX_LEN prefill CUDA graph bucket when the graph captures a DP gather (MegaMoE sparse-DP hang) (#37933) 3 people 2026-09-09 10:52:16 -07:00
• 0027af2eac [diffusion] [NPU] Optimize LTX-2/2.3 inference performance for NPU (#34722) 6 people 2026-09-09 19:36:18 +03:00