Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• 8a4c517a60 [Docs] Restore the AIME25 label so GLM-5.3 FP8 and BF16 scores render again (#36950) Mohammad Miadh Angkad 2026-08-29 11:38:26 +08:00
• 505228823f [NPU] [DOC] udpate supported features on NPU (#36940) amote-i 2026-08-29 10:52:01 +08:00
• 92b0d447b0 [CPU] Fix wrongly causal-masked bidirectional attention (#35434) Chunyuan WU 2026-08-29 10:49:43 +08:00
• 51c18d9aa8 [NPU] [DOC] update npu best practice (#36476) amote-i 2026-08-29 09:38:52 +08:00
• 0df69849ba [Fix] Drop the duplicated DSpark draft sample_block call (#36934) Liangsheng Yin 2026-08-28 18:10:41 -07:00
• 7276a30c45 [Fix] Support LSE on the RadixAttention extra-kwargs graph path (#35453) Aurick Qiao 2026-08-28 18:04:00 -07:00
• a16872767f Update CUDA 13.4 image to flashinfer 0.6.18rc10, cutedsl 4.8. Fix sgl- wheel unpinning (#36929) Trevor Morris 2026-08-28 17:52:15 -07:00
• 89816a21a1 [AMD] Update v4 amd cookbook 0828 (#36828) Thomas Wang 2026-08-29 08:20:22 +08:00
• 36afd70f9c [ROCm][Bugfix] Use token-level KV indices in the aiter ASM context-prefill gather (#36852) Olga Miroshnichenko 2026-08-29 03:17:18 +03:00
• 4944e50e2c [AMD] Add Work-Centric (Lean) Attention: a persistent-CTA decode kernel for long-context serving (#33576) valechenandHAI 2026-08-28 17:00:44 -07:00
• db6f0a9d53 Refactor JIT kernel and expert-pack directory layout (#36704) Xiaoyu Zhang 2026-08-29 07:41:25 +08:00
• 50bc1a3767 [diffusion] Keep Cosmos3 Nano resident on 96 GB GPUs (#36641) Xiaoyu Zhang 2026-08-29 07:40:46 +08:00
• e1b3bba3cc [Kernel] Route large SM90 row/column-scaled FP8 GEMMs to Torch (#34318) Gregory Leleytner 2026-08-29 03:35:37 +04:00
• 96a4dcdde8 [CI] Slim JIT kernel unit tests (#36887) Xiaoyu Zhang 2026-08-29 07:26:49 +08:00
• d12b313b93 fix: KT's last MoE layer stops deferring experts again (#36921) Cheng Wan 2026-08-28 15:05:08 -07:00
• ca179b761b Use kernel build node for cu134 image (#36912) Trevor Morris 2026-08-28 14:56:23 -07:00
• 60f6d77a98 [mem_cache] Carry swa_evicted_seqlen into SWARadixCache.cache_unfinished_req (#36909) Liangsheng Yin 2026-08-28 14:29:27 -07:00
• 4d78d59e51 [MoE] Make simulated expert routing support DP>1, and fuse into one triton kernel (#29718) Jonny Kongandjonnykong 2026-08-28 14:22:45 -07:00
• f65b2b2b15 [Fix] Lazy-import aiter in DSv4 paged_decode to unbreak CPU CI (#36914) Liangsheng Yin 2026-08-28 14:16:53 -07:00
• 2a96ebf648 [AMD][DSV4] perf: retune decode split-K heuristic for MI355X (#36094) karverma-amd 2026-08-28 15:48:41 -05:00
• ade4f4ba8b fix: report the real backend for non-CUDA CI registrations in /rerun-test (#36778) Shuwen Wang 2026-08-29 04:45:59 +08:00
• 3254f9b47c [Blackwell] Reserve SMs for DeepGEMM MegaMoE grid barriers (#36657) weireweireandweireweire 2026-08-29 04:37:36 +08:00
• 70088aa5db Fix KeyError on batch requests whose state is freed before it is read (#36638) Mohammad Miadh Angkad 2026-08-29 03:57:07 +08:00
• c7879af887 [AMD] release rocm10 image for gfx1250 from amd_helios (#36892) YC Yen-Ching Tseng 2026-08-28 21:00:25 +02:00
• f611e0c073 [HiCache] Fence load-back behind the forward stream (#36738) Zhiqiang Xie 2026-08-28 11:12:51 -07:00
• c9bba091f8 [HiCache] Key storage prefetch by the request namespace (#36382) Zhiqiang Xie 2026-08-28 11:12:10 -07:00
• ef20fab38a config: the forwarding slots go; the dispatcher calls the family directly (#36792) Cheng WanandClaude Opus 5 2026-08-28 10:26:29 -07:00
• 7bc3204117 config: three cache and pool readers take the bags (#36791) Cheng WanandClaude Opus 5 2026-08-28 10:21:32 -07:00
• 43c63a22ff config: the derived parallel widths are computed from the leaves (#36790) Cheng WanandClaude Opus 5 2026-08-28 10:18:46 -07:00
• c2928e86d7 config: the resolution pipeline moves out of the record (#36789) Cheng WanandClaude Opus 5 2026-08-28 10:17:24 -07:00
• 726665e08e [CI] Read subprocess stdout on a background thread to avoid EOF deadlock (#36673) pllimax 2026-08-29 01:11:02 +08:00
• 8e04f66a70 HiCache: avoid unnecessary all-reduce in check_prefetch_progress (#36425) Chao Shi 2026-08-29 00:52:53 +08:00
• 23cb11ae01 Fix KV cache pool sized far too small when weight-loading memory is still referenced (#36583) Alison Shao 2026-08-28 09:47:23 -07:00
• fcb4ff1ee3 [NPU] Update sgl-kernel-npu version (#36433) 黄孝君 2026-08-29 00:42:28 +08:00
• 6ff2fe6a6a CI: split JIT kernel unit tests into two partitions (#36775) Po-Han Huang (NVIDIA) 2026-08-29 00:32:58 +08:00
• 9579bff860 [AMD] Add ROCm 10 (gfx942 / gfx950) release images (#36434) Bingxu Chen 2026-08-28 23:03:39 +08:00
• 395c2258c3 [Docs] Add GLM-5.3 cookbook (#36827) 4 people 2026-08-28 07:56:06 -07:00
• b644771e07 [Fix] Route the Mooncake MoE A2A backend through Kimi K3's EP-A2A / SP-MoE fast path (#36862) 匡胤鑫 2026-08-28 21:46:35 +08:00
• ecbadf0b4b [NPU] [Diffusion] support distributed inference pipeline for GLM-Image (#31320) Артем СавкинandXiaoyu Zhang 2026-08-28 15:39:05 +03:00
• 803b4fb31c [diffusion] refactor: scope model-specific API parameters (#35613) Mick 2026-08-28 19:08:30 +08:00
• d56706459c bugfix for index_fill_ on NPU (#36759) ming_wang 2026-08-28 17:41:49 +08:00
• 74df026877 fix(cpu): skip GPU JIT MoE top-k on CPU (#35677) Xinguo ZhuandMa Mingfei 2026-08-28 17:31:26 +08:00
• 9cee0a31d1 [XPU] Lazily import tvm_ffi-dependent all_reduce kernel in minimax_m2 (#35290) K Chiranjeevi 2026-08-28 14:48:03 +05:30
• 69a49fede8 fix(kimi-k3): preserve dense ModelSlim MLA weights (#36603) McZyWu 2026-08-28 17:15:52 +08:00
• 1e6d041f78 [BugFix][VLM] keep Qwen3-VL MoE inference deepstack order (#34690) Zheng Wengang 2026-08-28 17:02:16 +08:00
• 3785b2d20f [HiCache] Reject load-back specs that claim nodes pinned by an in-flight load-back (#35931) Zhiqiang Xie 2026-08-28 01:59:06 -07:00
• eebb99c049 [diffusion][kernel] avoid 4D scale-shift autotuning (#36521) Xiaoyu Zhang 2026-08-28 16:58:19 +08:00
• 45424d8434 [diffusion][kernel] support transposed residual-gate add (#36504) Xiaoyu Zhang 2026-08-28 16:54:37 +08:00
• 1061e34785 [HiCache] Retry L3 storage prefetch after a missed attempt (#36227) Zhiqiang Xie 2026-08-28 01:35:21 -07:00
• 3f1031d697 [HiCache] Heal the storage existence cache on a hybrid prefetch discard (#36386) Zhiqiang Xie 2026-08-28 01:33:40 -07:00
• 989e51ba9c [Docs] Rename Tencent cookbook page titles to "Hy4 preview" / "Hy3 preview" (#36823) Haoguang Cai 2026-08-28 01:31:52 -07:00
• d3b972cbf0 fix(lora): build the MoE LoRA align JIT kernel on ROCm (#36379) Yikai Zhangandamdpilot-upstream-sync 2026-08-28 01:18:42 -07:00
• 0c7d017dbb [AMD][Fix] Qwen3.5: make empty-batch guard tuple-aware on fused AR+quant path (#35341) Bingxu Chen 2026-08-28 16:11:53 +08:00
• 2a7fb511c9 [AMD][CI] Limit HiCache MGSM eval concurrency on ROCm (#36308) Bingxu Chen 2026-08-28 16:05:59 +08:00
• 2960d69622 [Cookbook] Hy4-Preview follow-ups: runtime-accurate recipes + released-model info (#36808) zijiexiaandClaude Fable 5 2026-08-28 00:19:07 -07:00
• acc918b3ec [AMD] Qwen3.5 ASM FMHA chunked-prefill context attention (#36758) 3 people 2026-08-27 23:37:45 -07:00
• 1948b61ad4 [Cookbook] Add the Hy4-Preview model page (Tencent) (#36804) zijiexiaandClaude Fable 5 2026-08-27 23:26:13 -07:00
• 9fc60a8afd [PD] Fix MORI-IO ABORT bootstrap message handling (#29133) 5 people 2026-08-28 14:02:35 +08:00
• 137fab2f48 fix(deps): pin compressed-tensors to 0.18.0 (#36794) YAMY 2026-08-27 22:52:47 -07:00
• baf09f3954 [AMD] Enable deepseek-v4 topk_transform v2 kernel (#36684) Thomas Wang 2026-08-28 13:36:11 +08:00
• cce0b53466 [AMD] Increase gfx950 DSA model indexer topk_transform kernel occupancy (#35628) wufann 2026-08-28 13:20:52 +08:00
• 43b5a57dbb [Docs] Feature GLM-5.3-Flash in the popular-models banner (#36784) zijiexiaandClaude Fable 5 2026-08-27 21:56:10 -07:00
• 0665102ce5 [Fix] Resolve tool argument types through top-level anyOf/oneOf/allOf (#36626) Xinyuan Tong 2026-08-28 12:50:30 +08:00
• 7088f21922 [diffusion] fix: make tail_attn_meta cuda-graph capturable (#36658) Yihao WangandClaude Fable 5 2026-08-27 21:22:07 -07:00
• ad5a105a4a [Diffusion] Fix the five unit tests failing on main (#36726) Kevin MiandClaude Fable 5 2026-08-27 21:18:32 -07:00
• 2380121e9b [HiCache] Stop populating host-pool mmaps twice (-13% allocation time) (#36705) 3 people 2026-08-28 11:54:11 +08:00
• 2b209711d8 [AMD][DSV4] perf: MXFP8 MoRI dispatch to match the w4a8 MoE input format (#36119) karverma-amd 2026-08-27 22:45:40 -05:00
• aa0a0aa3c3 [AMD][DSV4] perf: bound the MoRI receive buffer during decode (#36130) karverma-amd 2026-08-27 22:27:52 -05:00
• ce6e1f46b4 Limit concurrent build jobs for cu134 container (#36756) Trevor Morris 2026-08-27 20:24:35 -07:00
• 6480cce9bc perf: skip redundant scheduler metadata gather for DP1 (#36568) YAMY 2026-08-27 20:04:54 -07:00
• daf6317196 [mem_cache] Add free_full to release the full side of a tombstoned SWA node (#36637) Liangsheng Yin 2026-08-27 19:34:48 -07:00
• b7686e17d6 [k3] declare packed_modules_mapping on KimiK3ForConditionalGeneration (#36211) Kurt Shuster 2026-08-27 22:22:36 -04:00
• 84a6f51bd6 Fix DeepSeek V4 multistream QKV buffer lifetime (#36547) Aurick Qiao 2026-08-27 19:10:36 -07:00
• cbd0271574 [diffusion] model: add cosmos3 transfer capability (#34747) 3 people 2026-08-28 03:37:22 +02:00
• de2fb50120 [AMD] Enable aiter mla asm path through padding attn heads for Kimi K3 (#36356) billishyahao 2026-08-28 09:30:29 +08:00
• 22f37d414b [NPU] Chain PR test jobs and disable two DeepSeek-V4-Flash perf tests (#36672) pllimax 2026-08-28 09:16:09 +08:00
• 96b31770f9 [diffusion] fuse Helios paired transposed RoPE (#36502) Xiaoyu Zhang 2026-08-28 08:57:37 +08:00
• 26fd7fdaa2 [Feature] Support PP in full prefill CUDA graphs (#35451) Aurick QiaoandYuwei An 2026-08-27 17:32:00 -07:00
• 7cbe564829 [Fix][XPU/ROCm/NPU] Defer sgl_kernel.quantization import in expert_pack (#36529) ashwini rathi 2026-08-28 04:47:24 +05:30
• 76217f6603 Revert "[NPU] [bugfix] Fix import of ggml_moe_a8_vec and Fix NPU MLA HiCache backup accessing missing data_ptrs" (#36747) Baizhou Zhang 2026-08-27 16:16:30 -07:00
• 5640e53cab [NPU] [bugfix] Fix import of ggml_moe_a8_vec and Fix NPU MLA HiCache backup accessing missing data_ptrs (#36640) chx96642264andsglang-npu-bot 2026-08-28 07:14:48 +08:00
• 5d52f02f22 [misc] Fold the allocator free-group flag into free_group (#36739) Liangsheng Yin 2026-08-27 16:12:47 -07:00
• 6ccfeb59bc cookbook: add a Speculative card to the GLM-5.3-Flash playground (#36740) zijiexiaandClaude Opus 5 2026-08-27 16:03:59 -07:00
• dc10483592 [Kernel] Add H200 MoE configs for Qwen3.5 and Qwen3.6 (#35374) Vedant V JhaveriandCopilot 2026-08-27 16:01:22 -07:00
• e283c9f8ea [AMD][CI] Merge the four MI35x DeepSeek-V3.2 nightly jobs into two to save runtime (#36736) 3 people 2026-08-27 15:40:27 -07:00
• 0132848349 Revert "[Fix] Disable --enable-symm-mem under CUDA graphs on Kimi hybrid models" (#34842) skyandClaude 2026-08-28 05:19:27 +08:00
• ff5578eb4e [1/N][Mix] Mixed Chunk Prefill Base (#36288) Yuwei AnandClaude Fable 5 2026-08-27 12:59:56 -07:00
• 6ff2a20ccf config: the record is not an object that gets passed around (#36622) Cheng WanandClaude Opus 5 2026-08-27 12:57:10 -07:00
• fd40a331bf config: a parallel size has one spelling; a patched scope declares its own (#36621) Cheng WanandClaude Opus 5 2026-08-27 12:56:42 -07:00
• ca1d7ed8e6 config: a parallel leaf with no live counterpart is read bare (#36620) Cheng WanandClaude Opus 5 2026-08-27 12:56:11 -07:00
• 7c3b5a6732 config: every handler declares its cuda-graph decisions (#36725) Cheng WanandClaude Opus 5 2026-08-27 12:55:34 -07:00
• bd4bb1781a config: resolution declares, and nothing writes a field (#36618) Cheng WanandClaude Opus 5 2026-08-27 12:53:17 -07:00
• d1f14431fd GLM-5.3-Flash cookbook: HiCache for LL, fusion-flag drop, EAGLE, default-cell numbers, DCP4 overlay (#36544) Xinyuan Tong 2026-08-28 03:18:25 +08:00
• 46a544e0a0 [Docs] GLM-5.3-Flash: point at compute-mamba-ratio for the KDA/KV pool split (#36719) zijiexiaandClaude Opus 5 2026-08-27 12:16:35 -07:00
• a0a2295271 Refactor server_args constants and layout (#36676) Lianmin Zheng 2026-08-27 12:16:16 -07:00
• 11de5e2281 docs(cookbook): add GB10 (DGX Spark) MXFP4 cells for Ling-3.0-flash (#36364) Xinyuan Tong 2026-08-28 03:15:17 +08:00
• 20a491d1d3 [Fix] Re-encode multimodal embeddings after cache mismatch (#36595) Mickandjohn.xh 2026-08-27 23:24:30 +08:00
• 024a7a1031 [diffusion] Fix native LingBot-Video text encoding (#36542) Xiaoyu Zhang 2026-08-27 22:58:22 +08:00
• 8ad76415e2 [PD][mori] Align prefill transfer control plane for unified control plane (#36160) inkcherry 2026-08-27 21:43:13 +08:00
• a7e3f590ca [Diffusion] Fuse LongCat residual gate updates (#36577) Xiaoyu Zhang 2026-08-27 21:07:04 +08:00