Commit Graph
6055 Commits
Author SHA1 Message Date
desmond-intel 6dc7b3421b Inference Support Mamba 2 and 1 (#34556) 2026-09-12 20:51:18 +08:00
KunalandClaude Fable 5.1 fd32226706 Fix gpt-oss RunAI streamer weight ownership (#38908)
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-12 20:21:59 +08:00
Shuwen Wang bd45cd50ca [Unified Tree] Port SWA Branching-Point Caching to the Rust TreeCore (#37584) 2026-09-12 18:37:57 +08:00
Yanbin JiangandShuwen Wang 0b415fa573 [HiCache][LoRA] Isolate storage pages by extra key (#38577)
Co-authored-by: Shuwen Wang <47200617+alphabetc1@users.noreply.github.com>
2026-09-12 18:18:14 +08:00
Shuwen Wang b9cb96496d [Unified Tree] Preserve aux LRU recency when splitting nodes (#38482) 2026-09-12 18:17:38 +08:00
Qiaolin Yu 6953dae005 [Qwen 3.8 Next] Remove unused tokenwise QSA implementation and tests (#38960) 2026-09-12 01:32:01 -07:00
Tuan Nguyen Gia 7c195b9151 [AMD] Fix Quark load of MiniMax-M3 MXFP4 index_qkv_proj (#37254) 2026-09-12 00:50:41 -07:00
Xiaoyu ZhangandCursor 0a57403468 [Perf] Optimize Qwen3-VL unique-image serving on H100 (#36411)
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-12 15:16:59 +08:00
Khoa PhamandClaude Fable 5.1 6ba96d329f [DCP] Resolve --dcp-comm-backend to fi_a2a/a2a by default for every model (#39165)
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-11 23:48:42 -07:00
YAMY 55e5e21c88 [Qwen3.8-Next] Add PD state transfer for Flash Next (#36651) 2026-09-11 22:31:45 -07:00
Aurick Qiao c389dd0863 Fix RunAI object-storage checkpoint index filtering (#38988) 2026-09-12 12:25:57 +08:00
iridiumine a984c78330 [NPU] Support DFlash speculative decoding for MiMo-V2.5-Pro (mxfp4) (#37565) 2026-09-12 11:59:23 +08:00
ff1ce11348 [diffusion] model: support VDN-H3 with a hybrid_window_attn_h3 backend (#37903)
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: Haocheng Xi <xihc@berkeley.edu>
Co-authored-by: Mick <mickjagger19@icloud.com>
2026-09-12 11:36:32 +08:00
FrancisandQiaolin-Yu e91c948057 feat: support TP>1 Domino rollout for DFlash V2 (#37069)
Co-authored-by: Qiaolin-Yu <liin1211@outlook.com>
2026-09-11 18:51:11 -07:00
a207786205 [PD] Transfer the DCP-replicated DSPARK draft KV in DCP1->DCP-N relayouts (#37709)
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-11 17:15:18 -07:00
Xinyu Kang 6671cfc775 [AMD] Fix weight checking for AITER-shuffled block FP8 weights (#34330) 2026-09-11 15:49:56 -07:00
YAMY f963d7a27c fix(qsa): make the paged sparse-decode gather memory-safe (zero-fill scratch, int64 offsets, dequant FP8 on gather) (#38851) 2026-09-11 15:47:25 -07:00
Yanbin Jiang ec30f19e4a [LoRA] Support MoE in full and breakable prefill CUDA graphs (#38578) 2026-09-11 14:40:02 -07:00
Hanming Lu 45715e7f20 [Fix] Disable NCCL graph buffer registration for the TP LM-head all-to-all (pure-DP decode hang under request bursts) (#38936) 2026-09-11 13:10:48 -07:00
Bingxu Chen a338a9a01c [AMD][CI] Skip failing Wave test and relax multi-LoRA output check (#38585) 2026-09-11 12:25:20 -07:00
Jeremy Zhang f69d6fc28a [OpenAI] Propagate PD routing metadata through /v1/responses (#35503) 2026-09-12 01:33:38 +08:00
Vignesh Sethuraman ddf02a4f58 [AMD] aiter: resolve SWA KV pool for draft workers + guard paged decode (#38756) 2026-09-11 10:13:48 -07:00
Mohammad Miadh AngkadandMohammad Angkad 4309c7ce19 Fix stale DSV4 indexer metadata names in the TopK v2 dispatch test (#39101)
Co-authored-by: Mohammad Angkad <mohammad.angkad@radixark.ai>
2026-09-11 08:31:42 -07:00
Xiaoyu Zhang d6b5dca90c [Diffusion] Optimize SANA-WM convolution post-processing and streaming GDN (#38529) 2026-09-11 23:30:17 +08:00
Shangming Cai 2c10f87991 [PD] Read nixl TransferInfo.is_dummy as a field in unit tests (#39100) 2026-09-11 23:20:02 +08:00
2a46cf2ca0 [PD] Share the prefill->decode failure notification across backends (#36612)
Co-authored-by: inkcherry <mingzhi.liu@amd.com>
Co-authored-by: Shangming Cai <csmthu@gmail.com>
2026-09-11 23:16:17 +08:00
593c7a900d Add granite_thinking_parser reasoning parser for Granite 4.2 (#38693)
Signed-off-by: Yousaf Shah <yousaf.shah@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
2026-09-11 07:18:35 -07:00
335f6aab27 [DSV4] Support raw-index output in TopK v2 (#33672)
Co-authored-by: weireweire <20922698+weireweire@users.noreply.github.com>
Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>
Co-authored-by: Po-Han Huang (NVIDIA) <53919306+nvpohanh@users.noreply.github.com>
2026-09-11 07:17:36 -07:00
Xinyuan Tong e8a36d339c Auto-detect GLM-5.3 chat templates as glm45/glm47 parsers (#38297) 2026-09-11 09:02:44 -04:00
MickandMick Qian 358c163250 [diffusion] fix: recover ipc jit initialization after interrupted builds (#39034)
Co-authored-by: Mick Qian <mickqian@users.noreply.github.com>
2026-09-11 19:34:09 +08:00
822e73ccdd [Unified Cache][AMD] Support DeepSeek-V4 unified KV in direct external linkers (#38269)
Co-authored-by: amd-danli103 <danli103@amd.com>
Co-authored-by: Duyi-Wang <duyi.wang@amd.com>
Co-authored-by: TianDi101 <tiandi920722@gmail.com>
2026-09-11 01:49:29 -07:00
Mohammad Miadh AngkadandMohammad Angkad ad5af539cd [CI] Trim DSV4 trtllm B200 tests (#39013)
Co-authored-by: Mohammad Angkad <mohammad.angkad@radixark.ai>
2026-09-11 01:04:16 -07:00
Vignesh Sethuraman 76ef679cb1 [AMD] aiter: fail loudly on cross-layer KV sharing in target_verify (#38755) 2026-09-10 23:37:03 -07:00
Jialin Ouyang 6182d45524 [Test] Fix Q8KV8 sparse-prefill pool fixture after page-size rename (#39019) 2026-09-10 22:22:35 -07:00
Rishabh Sinha 2018c64e22 [PD] Add is_dummy truth-table wire tests for mooncake and nixl (#34977) 2026-09-11 13:14:41 +08:00
Shunkangz 40a84d6dfc [kv-shard 1/4] Logical-page placement with UnifiedRadixCache (#38356) 2026-09-10 22:06:42 -07:00
Yikai Zhang b3dc0388ed test(lora): enable ROCm logprob accuracy coverage (#38791) 2026-09-10 20:29:06 -07:00
ANSHUMAN TRIPATHYandMa Mingfei 0fadad8933 DFLASH support added for XPU (#32798)
Co-authored-by: Ma Mingfei <mingfei.ma@intel.com>
2026-09-11 10:39:37 +08:00
Miaomiao Jiang 690428b470 [XPU] Add xpu forward in Gemma3RMSNorm & Add test and benchmark for Gemma3RMSNorm (#36278) 2026-09-11 10:32:25 +08:00
ANSHUMAN TRIPATHY 3716e496ea Speculative Decoding support for intel_xpu attention backend on XPU target (#30548) 2026-09-11 10:25:25 +08:00
ANSHUMAN TRIPATHYandmingfeima 2adb2e8485 [XPU] Adapt device agnostic API usage (#32093)
Co-authored-by: mingfeima <mingfei.ma@intel.com>
2026-09-11 10:03:12 +08:00
Siju Samuel 67d3a2ea57 [XPU] Make checkpoint_engine worker device-agnostic (#32382) 2026-09-11 09:56:39 +08:00
ad7f57c9ea [Responses] Fix empty-prompt routing for token-first chat encoders (kimi_k3, inkling) (#35486)
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
2026-09-11 01:06:20 +00:00
jasonjk-park 9df72e8f5a Fix custom logit processor params when num_tokens_in_batch is used (#38730) 2026-09-10 17:52:08 -07:00
Liangsheng Yin 41da06adca [Refactor] Generalize DeepSeek V4 compressed pool management (#38954) 2026-09-10 17:29:03 -07:00
Mohammad Miadh AngkadandMohammad Angkad 52c191da52 [Deps] Retire the CUDA 12 lane (#38404)
Co-authored-by: Mohammad Angkad <mohammad.angkad@radixark.ai>
2026-09-10 16:58:09 -07:00
fae8cd84cb Support NemotronH_Omni_Reasoning_V3 in SGLang (#35599)
Signed-off-by: Ryan Stewart <rystewart@nvidia.com>
Signed-off-by: rystewart-nvidia <rystewart@nvidia.com>
Co-authored-by: elvischenv <219235043+elvischenv@users.noreply.github.com>
Co-authored-by: Po-Han Huang (NVIDIA) <53919306+nvpohanh@users.noreply.github.com>
2026-09-10 16:57:22 -07:00
cctry 203d7e812c Fix KV-canary workspace accounting after graph capture (#38596) 2026-09-10 16:10:12 -07:00
cctry 42bbaac259 [metrics] Report logical prefill token counts (#38566) 2026-09-10 16:09:21 -07:00
Liangsheng Yin dc5f59c3a2 [Refactor] Clarify DeepSeek V4 metadata names for V4.1 (#38947) 2026-09-10 15:42:52 -07:00