Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

main

Mono Color

• 677c1cbdc9 [Metrics] Propagate idle gaps across all scheduler loops (#40004) 3 people 2026-09-18 22:42:00 -07:00
• 8189e3896b [PD] Skip singleton transfer-status all-reduces (#40003) 3 people 2026-09-18 22:41:50 -07:00
• 5b42d10edf fix: restrict SafeUnpickler to explicit globals (#40259) 3 people 2026-09-18 22:38:11 -07:00
• 5e9342d16f [PP][DeepSeek V4] Overlap communication and optimize SM120 prefill (#38792) 3 people 2026-09-19 12:18:04 +08:00
• 929230a6f0 Fix GLM-OCR MTP multimodal embeddings and positions (#39088) 3 people 2026-09-19 12:10:04 +08:00
• 111b905bd1 fix(openai): recover logprobs token bytes from token_id (UTF-8 fragments) (#38604) Mind Lab 2026-09-19 11:30:04 +08:00
• f1fbbd17bb [diffusion] chore: update Cache-DiT to 1.5.1 for DMD Calibrator, SVDQuant DQ, etc (#40104) DefTruthandcopilot-swe-agent[bot] 2026-09-19 11:14:33 +08:00
• 090263eff6 [VLM] avoid CUDA placement on non-CUDA platforms (#38750) Vincent Liu 2026-09-18 20:13:22 -07:00
• c3aa09b0db [Kernel] Fuse hc_combine_norm for mid-size verify batches (9-96 rows) (#40208) Alexandkpham-sgl 2026-09-18 20:03:49 -07:00
• c475ac5eaf [diffusion] feat: out of tree platform support (#37547) hmalgewattaandMick Qian 2026-09-18 18:35:58 -07:00
• 986959e3c4 [Refactor] Deduplicate kernel helpers and remove unused code (#40197) Xiaoyu Zhang 2026-09-19 09:27:30 +08:00
• 10b0bcfd18 [PD] Allow decode radix cache and HiCache L1/L2 with DCP (#40263) Khoa PhamandCursor 2026-09-18 17:54:35 -07:00
• d507accadc [Test] Drop dead and strictly-subsumed CI test registrations (#40264) Liangsheng Yin 2026-09-18 17:49:13 -07:00
• fa7e83fd09 Name the two widths of the WORLD group (#40070) Cheng Wan 2026-09-18 17:47:02 -07:00
• 81421b91e9 One read path for every parallel name (#40069) Cheng Wan 2026-09-18 17:43:56 -07:00
• afe71f4b9e Read process groups through the runtime context (#40068) Cheng Wan 2026-09-18 17:40:32 -07:00
• 5931fd60ee Support unified memory page-envelope transfers in PD (#39477) 5 people 2026-09-18 17:39:50 -07:00
• d0730a0e8b Give the attention-DP width and rank one home (#40067) Cheng Wan 2026-09-18 17:34:33 -07:00
• 8ea0ee300d perf(sampling): avoid GPU syncs when applying custom logit processors (#39234) Lifan Shen 2026-09-18 17:09:30 -07:00
• b876213548 [Test] Add a ci-test-audit skill cataloging CI and test audit patterns (#40257) Liangsheng Yin 2026-09-18 16:35:49 -07:00
• 21e6c98ccb Fix corrupted chat prompts on mistral_common tokenizers (tool_choice auto never fires) (#39773) Alison ShaoandXinyuan Tong 2026-09-18 16:12:06 -07:00
• 2394b231c2 Fix Mistral3 retaining every vision-tower layer to read one (#39185) 3 people 2026-09-18 16:11:54 -07:00
• fa521e2758 [MM] Copy placeholder ids to CUDA asynchronously (#40010) 4 people 2026-09-18 16:10:37 -07:00
• ceb1d2e580 [PD] Enable optimistic prefill with buffer-only L3 write-through HiCache (#40043) Zhiqiang Xieandcctry 2026-09-18 16:04:15 -07:00
• 5e4b94b134 [MM] Skip VMM error gathers for text-only requests (#40005) 4 people 2026-09-18 16:00:43 -07:00
• 803f0c93d2 Fix DSA partial DP-TP mode log to use derived attn_tp_size (#39871) hunhokimandHun-ho Kim 2026-09-19 07:54:45 +09:00
• f3851486cb [Perf] Fuse SWA page lookup and mapping clear (#38948) Jialin Ouyang 2026-09-18 15:51:07 -07:00
• 81a199f56a [HiCache] Read the in-flight buffer backup's node id from its snapshot in sanity_check (#40013) 3 people 2026-09-18 15:47:56 -07:00
• 0e5347db82 Support MXFP8 and deferred route weighting in DeepEP v2 (#40030) 4 people 2026-09-18 15:40:38 -07:00
• 6cc9090d1f [mem_cache] Release up to owned_kv_len on radix cache insert (#40075) Liangsheng Yin 2026-09-18 15:37:52 -07:00
• a0534f8cca [HiCache] Stop arming a prefetch retry for a too-short storage span (#40042) Zhiqiang Xie 2026-09-18 15:33:58 -07:00
• d346b214fb feat(kv-cache): support SM100 NVFP4 GenMHA and speculative decoding (#36340) Sam (Kesen Li) 2026-09-19 05:50:46 +08:00
• f5a1434700 [HiCache] Document transfer arguments (#40239) Lianmin ZhengandJialin Ouyang 2026-09-18 14:08:19 -07:00
• cd4dd81c22 [AMD][DSV4] fix: drop shadowing local get_exec import that breaks model startup on ROCm (#40186) amd-danli103 2026-09-19 04:28:38 +08:00
• 6a9c7001d3 [Logprob] Borrow graph-pool memory for input logprob logits construction (#40007) metamergebotandcctry 2026-09-18 13:18:01 -07:00
• da2f434951 [Spec] Add explicit prefill shared-read capability for plugins (#39502) Jialin Ouyang 2026-09-18 13:11:01 -07:00
• aed3fb1cdd [Router] Log every request at one site; derive its outcome from the final status (3/3) (#39465) 4 people 2026-09-18 12:11:00 -07:00
• 248c202b46 Use runtime token widths for Triton speculative verification (#39859) Lianmin Zhengandraghotham 2026-09-18 11:07:55 -07:00
• 6bd1a0af1d Add registration for external model configurations (#39452) Lianmin Zheng 2026-09-18 10:06:24 -07:00
• 50a7de47d5 [Benchmark] Add agentic rollout simulator and offline explorer (#40034) Byron Hsu 2026-09-18 10:06:13 -07:00
• 4e0b56c811 [Router] Treat an upstream 503/429 as backpressure, not a breaker fault (2/3) (#39464) 3 people 2026-09-18 09:46:29 -07:00
• 191172fa74 [Unified Tree] fix: exempt host-locked aux nodes from the sanity_check host-LRU check (#39980) Shuwen Wang 2026-09-18 23:29:29 +08:00
• 45938a24ae [AMD] GLM-5.2 MI355X MXFP4: bump image to 20260916, use HIP Top-K (#40148) ChangLiu0709 2026-09-18 16:21:58 +01:00
• 7714b182f2 [Bugfix] Fix top-1 MoE routing with non-unit scaling (#40187) Xiaoyu Zhang 2026-09-18 22:53:13 +08:00
• 81363bf8cb [kernel] Share the warp vectorized copy and enforce its alignment (#36176) 3 people 2026-09-18 22:40:54 +08:00
• 9784d5f979 [diffusion] doc: sync CFG and tracing documentation (#39883) MickandMick Qian 2026-09-18 18:03:40 +08:00
• a6cf05817f dsv4.1: remaining model and runtime integration (#38798) 9 people 2026-09-18 02:55:30 -07:00
• 1b200ffaaa [Quant] Serve 32-wide-K ue8m0 block-FP8 linears through the FlashInfer MXFP8 GEMMs (#40039) Liangsheng Yin 2026-09-18 02:51:12 -07:00
• 6c7c5e78de [NPU] Run arch35 block-FP8 dense linears on the native MXFP8 GEMM (#39823) iridiumine 2026-09-18 17:19:31 +08:00
• 7e6d5cbfac [NPU] Gate DFlash replay metadata refresh behind spec_algorithm check (#39879) iridiumine 2026-09-18 17:07:07 +08:00
• bbfcda48ce [CI] Add a unified-memory rerun test group (#40147) Cheng Wan 2026-09-18 01:59:07 -07:00
• d6090f92bf [NPU] Fuse MXFP4 W4A8 MoE gmm1 + swiglu + requant into one kernel (#39881) iridiumine 2026-09-18 16:50:49 +08:00
• 2dee23a876 [ROCm][diffusion] Enable fused qk norm and rope on ROCm (#35573) 3 people 2026-09-18 04:40:32 -04:00
• 1e8699fda3 [NVIDIA][comm] Merge EP+MoE-TP post-experts all-reduces into one _TP reduction (#32963) Shu Wang 2026-09-18 03:35:10 -05:00
• 8ac39c66d8 [NPU] support kimi k3 on A5 and improve performance (#39589) zhaozx-cn 2026-09-18 16:33:54 +08:00
• 6de4666e43 [Router] Derive error status from a failure class; preserve the worker's status (1/3) (#39463) 3 people 2026-09-18 01:32:45 -07:00
• 1fdd6c8921 [Runtime] Let out-of-tree platforms provide full graph backends (#37969) Rumit DesaiandXiaoyu Zhang 2026-09-18 00:42:46 -07:00
• c46bf5e990 [MoE] Disable FlashInfer fused finalize by default for numerical accuracy (#40105) Ziang Li 2026-09-18 00:15:57 -07:00
• 956d414dad Update ci permission (#40121) Ke Bao 2026-09-18 15:12:36 +08:00
• 0be8a0af0e [DSV4] fix: keep the TileLang JIT cache under SGLANG_CACHE_DIR (#39364) Shuwen Wang 2026-09-18 15:06:35 +08:00
• 3bf243d6a9 [Router] Shard the cache-aware KV tree by chain root (#39167) 4 people 2026-09-18 00:00:17 -07:00
• f86f60081d [NPU] Adapt hicache for K3 hybrid models (#39415) iridiumine 2026-09-18 14:48:20 +08:00
• 6952538980 [NPU] [DOC] delete unsupported api --disable-hybrid-swa-memory for npu (#40058) amote-i 2026-09-18 14:31:05 +08:00
• 4dbba37965 Verify the Ling-3.0-flash-VL FP4 lane on H200 and disable shared-expert fusion in quant recipes (#39419) Xinyuan Tong 2026-09-18 14:20:18 +08:00
• 3ce3b4969f [NPU] Avoid repeated BF16 wo_a weight transposes in DeepSeek-V4 decode (#39919) Jensen 2026-09-18 14:10:43 +08:00
• d337b865af Update NPU tag to post4 version and related scripts (#39953) 黄孝君 2026-09-18 14:03:07 +08:00
• 3d1b9e7549 [Bugfix] Include SM121 in DeepGEMM packed-scale selection (#39482) Bojiang LiandXiaoyu Zhang 2026-09-17 22:57:14 -07:00
• 0dad91d50f Fix int4 MoE tuner config filename (#35260) maithilijoshi20 2026-09-18 14:56:38 +09:00
• 6e1d9c1464 [profiler] Ignore PREBUILT batches in profile-by-stage (#40098) Hanming Luandhanminglu 2026-09-17 22:33:55 -07:00
• 6215aecd51 [diffusion] feat: add metrics support (#19084) Chi McIsaacandMick Qian 2026-09-17 22:22:03 -07:00
• 65ef55e2a8 [Scheduler] Add shortest-prefill-first scheduling (#40024) Yuwei An 2026-09-17 21:18:58 -07:00
• c055dc6ff6 [CI] Check B200 NUMA mapping against sysfs numa_node (#40055) Mohammad Miadh AngkadandMohammad Angkad 2026-09-18 12:14:47 +08:00
• 826d5170ae [Bug] Guard FlashInfer CUTLASS MoE against 0-token inputs (#38780) 3 people 2026-09-18 11:03:27 +08:00
• f447bb7080 [Kernel] Coalesce the KDA CuTe DSL decode state transpose: ~3x faster, bit-identical (#39680) Zixin HuangandClaude Fable 5.1 2026-09-17 19:53:13 -07:00
• db39b7f961 [Fix] Guard conditional top-logprob keys in the completions echo path (#34776) 4 people 2026-09-17 19:47:26 -07:00
• 740f57a02c [Spec] Fix CDF boundary handling in TreeSpeculativeSamplingTargetOnly (#35798) Nan Jiang 2026-09-17 19:41:52 -07:00
• 0214954f26 [gRPC] Stream engine state changes (#39915) William Arnoldandishandhanani 2026-09-18 11:17:25 +09:00
• a407915c17 [CPU] Avoid prefill CP predicates during decode graph capture (#39690) YanbingJiangandMa Mingfei 2026-09-18 09:21:05 +08:00
• f65c70bb7d [Kernel] Move CUDA and ROCm speculative kernels to JIT (#40033) Liangsheng Yin 2026-09-17 17:27:26 -07:00
• 20518d8518 [Scheduler] Align RadixCache no-insert cleanup with kv_len_to_handle (#35204) Nan Jiang 2026-09-17 17:00:59 -07:00
• 7bc9152447 [Test] Consolidate kernel tests under plural kernels tree (#39966) Xiaoyu Zhang 2026-09-18 07:37:48 +08:00
• b98a2d1096 [Docs] GLM-5.3-Flash cookbook: temporarily remove the DCP option (#40036) Khoa Pham 2026-09-17 16:17:25 -07:00
• 1f0c73e9bd [DSV4] Generalize attention metadata, sparse prefill, and KV pool over compress ratios (#39921) Liangsheng Yin 2026-09-17 15:55:19 -07:00
• 4f52a27563 [AMD] Fix DSV4 FP4 dequant path for AITER on ROCm (#35123) Ren YuzhouandRen Yuzhou 2026-09-18 04:51:22 +08:00
• 126c2f1bc1 [CI] Add metamergebot to CI permissions (#40017) Lianmin Zheng 2026-09-17 13:39:00 -07:00
• 72d9419bef [Router] Sample k random candidates for the min-load fallback (--min-load-choices) (#39170) 4 people 2026-09-17 12:09:09 -07:00
• 6c73368c32 [AMD][DSV4] Enable hicache on deepseek-v4 fp8 unified attn (#37778) Thomas Wang 2026-09-18 02:29:23 +08:00
• 1f60ddef5d [PD] Introduce runtime role switching between prefill and decode (#28403) 4 people 2026-09-18 01:45:12 +08:00
• a98d921658 [DP Attn] Fix crash for no token all-gather case (#39899) Byron HsuandByron Hsu 2026-09-17 09:21:09 -07:00
• 25c9f724d4 fix(multimodal): handle tensor images in exact-token preprocessing (#30368) 3 people 2026-09-17 12:20:21 -04:00
• e4cbb28ea1 [router] Improve SGLang chat render parity (#39133) 3 people 2026-09-17 08:53:40 -07:00
• 7ccbf5fd04 [NPU][CI] Increase e2e multi-node test timeout to 210 minutes (#39952) pllimax 2026-09-17 20:50:33 +08:00
• a9fb1c3238 dsv4(npu): support prefill context parallelism with interleave and zigzag (#39427) vstone-w 2026-09-17 19:59:41 +08:00
• 11c35b8433 [AMD] Load fused shared experts for Qwen4-Exp and Qwen3.5 MTP (#38878) jacky.cheng 2026-09-17 19:56:31 +08:00
• a1b4ec02ae [NPU][Diffusion] FA MXFP8 and modelslim w4a4f8 and w8a8f8 support for Wan2.2 and FLUX (#39438) Arseniy Mironov 2026-09-17 13:13:05 +03:00
• 6ca866ea29 [HiCache][Perf] fix: batch HiCache D2H submits per step for hybrid pools (#39050) Shuwen Wang 2026-09-17 17:53:58 +08:00
• 1a90ae6727 Add Agentic-Aware Tail-Optimized LRU eviction to the unified radix cache (#34012) yl3469andShuwen Wang 2026-09-17 02:52:59 -07:00
• 575759d90a [NPU] Support nccl backend for --remote-instance-weight-loader (#39413) silencejade 2026-09-17 17:28:15 +08:00
• e970453b43 [NPU]Refactor weight processing and add NPUSwigluLimit activation (#38420) 3 people 2026-09-17 16:40:32 +08:00
• 2d08cc5ede [AMD][Spec] Enable GDN ReplaySSM target-verify on ROCm (#38184) chuyehandchuyeh 2026-09-17 16:29:14 +08:00