Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• 55b1c09e73 [core] Consolidate compiled-kernel caches under SGLANG_CACHE_DIR (#32434) Shu Wang 2026-08-05 13:54:27 -07:00
• 717a559f02 [Scheduler] Align WAR fences with CUDA graph metadata reads (#33587) Jialin Ouyang 2026-08-05 13:52:20 -07:00
• ea65f8ddc9 Feat/spectrum (#31491) 6 people 2026-08-05 23:23:17 +03:00
• 5c4f72f92a [Build] Add srt_empty extra group for device-agnostic install (#31300) 3 people 2026-08-06 04:17:46 +08:00
• de34dd11e9 [CI] Fold duplicate-server suites and prune the retract matrix on 1-gpu-5090 (#33745) Liangsheng Yin 2026-08-05 12:41:51 -07:00
• 36853b8ffc [Spec] Support logprobs with DFlash (#33459) Jason Mancuso 2026-08-05 15:37:12 -04:00
• 1a045669e4 [CI] Merge tokenizer worker tests and drop redundant triton attention e2e (#33641) Liangsheng Yin 2026-08-05 11:55:11 -07:00
• 5f79cf3511 [DCP] Match the replicated draft KV pool's page granularity to its allocator (#33348) Khoa Pham 2026-08-05 11:40:28 -07:00
• b1bd871df5 [Unified Radix Cache] Complete the tree-core interface boundary (#33580) Jialin Ouyang 2026-08-05 11:39:28 -07:00
• 96c89863a3 Measure prefill busy time between launches (#33595) cctry 2026-08-05 11:26:27 -07:00
• acaab22d09 [diffusion] feat: add SageAttention packed varlen path for minimax-h3 (#33703) WenhaoZhang 2026-08-06 01:19:58 +08:00
• b3cdd016ba Add Ling-3.0-flash cookbook (#33556) Xinyuan TongandZijie Xia 2026-08-05 22:53:34 +08:00
• 4e7209caa8 [NPU] Add causal conv1d (#28267) zhaozx-cn 2026-08-05 22:22:49 +08:00
• 3425c93666 [diffusion] Wan VAE RMSNorm+SiLU fusion behind quality=high (H200 FastWan2.2 e2e 9.611 -> 9.125 s) (#33546) Xiaoyu ZhangandClaude Fable 5 2026-08-05 21:33:35 +08:00
• 593777c046 [FIX] [benchmark] Fix flush_cache failure after warmup by waiting for server idle (#33527) silencejade 2026-08-05 21:27:43 +08:00
• 3b4fac5b99 [XPU] DeepSeek V4: use sgl-kernel-xpu implemetation of flash_mla_sparse_fwd for prefill (#31865) Xuan LiaoandMa Mingfei 2026-08-05 21:05:53 +08:00
• 99709f734d [VLM] split multimodal scheduling from mm_utils (#32415) Mick 2026-08-05 20:24:12 +08:00
• a5888c956f [diffusion] Pack Ulysses Q/K/V input all-to-all into one collective + reusable a2a staging buffers (#33667) Xiaoyu ZhangandClaude Fable 5 2026-08-05 19:15:46 +08:00
• 2f22ed58ea [NPU] Adding a fast layernorm for diffusion models and fix BSA (#29027) 3 people 2026-08-05 04:06:00 -07:00
• 22d558b103 [Feature] Add GLM Image usage report (#33378) 3 people 2026-08-05 18:59:49 +08:00
• 8279702e0b [AMD] Stop publishing the K3 MI35X nightly image (#33689) YC Yen-Ching TsengandChen 2026-08-05 17:53:23 +08:00
• 6fa3f9df11 [Bugfix] Treat unsharded model.safetensors as HF weights in Mistral-native format detection (#33671) Alex NailsandClaude Fable 5 2026-08-05 01:54:46 -07:00
• 4a3d6ca88c [CI] Skip sglang-kernel and sgl-deep-gemm reinstall on version match (#33637) Liangsheng Yin 2026-08-05 01:46:16 -07:00
• a6e5fa7081 [Scheduler] Honor explicit min-free-slots thresholds (#33403) 3 people 2026-08-05 04:44:18 -04:00
• c0d5ebd6c4 [CI] Move CPU-only unit tests to the CPU suite and trim dead 5090 registrations (#33654) Liangsheng Yin 2026-08-05 01:43:29 -07:00
• 98ed5554bb Stop testing cu129 DeepGEMM wheels (#33675) Mohammad Miadh Angkad 2026-08-05 16:38:28 +08:00
• 81c7a54ecd [NVIDIA] Use sm_100f instead of sm_100a for sgl-kernel and FlashMLA (#33433) Trevor Morris 2026-08-05 01:36:46 -07:00
• 1478cdec9f [AMD] Fuse Kimi-K3 attn-residual aggregation (#33599) Xinyi Song 2026-08-04 23:20:04 -07:00
• d96df7bed5 [Diffusion] Batch GLM-Image AR requests (#30683) Артем СавкинandXiaoyu Zhang 2026-08-05 08:47:06 +03:00
• eac1f78568 [CI] Free hosted-runner disk space only when it is low (#33644) Liangsheng Yin 2026-08-04 21:55:06 -07:00
• 059269594c [DSV4] Add official DSV4 reasoning effort support (#33140) 3 people 2026-08-05 12:50:41 +08:00
• 198a3bc29b [Test] Route GEMM backend UTs through real layer modules and weight loaders (#33615) Liangsheng Yin 2026-08-04 20:53:26 -07:00
• 1033cae8d5 [CI] Speed up dependency install: dual-ABI Rust ext cache and prevalidation pruning (#33619) Liangsheng Yin 2026-08-04 20:33:48 -07:00
• 6c05aaae7e [trtllm_mha] perf: Stop allocating per-layer scratch inside the decode CUDA graph (#33063) Kaixi 2026-08-05 04:27:26 +02:00
• 4949b5fccf [XPU] Add qknorm_rope support for Flux (#30883) 3 people 2026-08-05 10:12:24 +08:00
• 5dc4102960 [npu] [bugfix] Fix PD‑disaggregation error (#33523) gjsheu 2026-08-05 09:54:00 +08:00
• a0b04dbe4c feat(grpc): add generation request semantics (#32588) 3 people 2026-08-04 18:53:45 -07:00
• 29831d58ef fix mm-chunk-embedding test suite (#32895) Zaili WangandMa Mingfei 2026-08-05 09:49:39 +08:00
• d2c405f19d [Intel GPU] DeepSeek V4 8/N: use sgl-kernel implementation of fused_k_norm_rope_flashmla on XPU (#28040) Polisetty V R K Jyothendra VarmaandMa Mingfei 2026-08-05 06:58:43 +05:30
• 9303e26f03 [ci] add qwen 3.5 mtp + replayssm + flashinfer gdn test (#33607) Qiaolin Yu 2026-08-04 18:08:42 -07:00
• 211ee64249 [rotary] Rebuild the shared RoPE cache entry when its buffers are dead (#33575) 3 people 2026-08-04 17:24:42 -07:00
• 87ed82ff7e Remove custom all-reduce disable from Kimi-K3 B300 recipe (#33612) Baizhou Zhang 2026-08-04 16:07:31 -07:00
• 76dc89f5aa [Test] Replace NVFP4 MoE runner backend e2e matrix with a layer-level unit test (#33611) Liangsheng Yin 2026-08-04 16:03:29 -07:00
• 0d99d91e49 [CI] Make B200 base-b suites single-GPU as prep for 1-gpu B200 runners (#33605) Liangsheng Yin 2026-08-04 15:53:01 -07:00
• a0b3f1dde6 [Test] Replace GEMM backend e2e matrices with layer-level unit tests (#33596) Liangsheng Yin 2026-08-04 15:50:41 -07:00
• b0fd31ba07 Multiple flexibility fixes for DP attention (#33537) Lianmin Zheng 2026-08-04 15:40:40 -07:00
• 6808c6d571 [Tiny] Little enhancement of Kimi-K3 test (#33609) Baizhou Zhang 2026-08-04 15:36:41 -07:00
• 58da9859c4 [CI] Extract download-rust-ext and give every install step a cache fallback (#33597) Liangsheng Yin 2026-08-04 15:33:34 -07:00
• b327d76682 [AMD] Bump mori to latest in sglang (#33462) Zhaoyi Li 2026-08-04 17:12:40 -05:00
• c8822fd990 Clarify post-capture KV reservation logs (#33598) cctry 2026-08-04 15:06:11 -07:00
• 19d3f86895 [LoRA] Laguna: per-layer LoRA hidden-dim resolution for packed attention (#30298) FilipandClaude Opus 4.8 2026-08-04 22:48:05 +01:00
• a9c3b55435 [Refactor] Keep chat template validation out of ServerArgs dispatcher (#33392) Xinyuan TongandAlex Nails 2026-08-05 05:45:47 +08:00
• e510dc58ba Add @Jiminator as codeowner for Laguna model and config (#33472) Jimmy ShongandClaude Opus 5 2026-08-04 14:10:19 -07:00
• 34af3ff386 Allow optimistic prefill with L2 hierarchical cache and write-back policy (#33545) Lianmin Zhengandcctry 2026-08-04 13:23:08 -07:00
• abddb1c7e9 [Kimi] Support kimi-k3 (#32541) +2636 people 2026-08-04 13:22:49 -07:00
• 0753663b8e [CI] Trim redundant B200 test registrations (#33586) Liangsheng Yin 2026-08-04 13:22:00 -07:00
• aa06433709 Avoid TRTLLM prefill output copy (#33306) Xingyu Liu 2026-08-04 12:54:04 -07:00
• 38dc2d6cf8 [metrics] Split tokenizer request metrics by stream (#32734) 3 people 2026-08-04 12:53:52 -07:00
• 4794b401d5 [Observability] Add startup, memory, and hybrid SWA diagnostics (#33375) Lianmin Zheng 2026-08-04 12:50:09 -07:00
• 5081c063c0 fix(metrics): clear forward occupancy on idle (#33562) Lianmin ZhengandJialin Ouyang 2026-08-04 12:49:17 -07:00
• dea2be5ae3 [CUDA Graph] Allow custom decode graph runners (#33553) Lianmin ZhengandItai Gat 2026-08-04 12:48:56 -07:00
• e76d0acdc9 migrate NPU PR/nightly test cases to a3-560T (#33346) 3 people 2026-08-05 01:17:58 +08:00
• 95d0e57e83 [diffusion] Fuse DiT FFN tanh-GELU into up-proj GEMM (cublasLt epilogue) behind quality=high (Qwen-Image 1024^2 denoise 12.36 -> 12.05 s on H200) (#33536) Xiaoyu ZhangandClaude Fable 5 2026-08-04 23:49:43 +08:00
• 0d0c7d853f [diffusion] FLUX.2 VAE decoder fast path behind quality=high (H200: 1024^2 97.6->29.2 ms, 2048^2 437.2->168.5 ms) (#33451) Xiaoyu Zhang 2026-08-04 23:48:19 +08:00
• 7adf2f4a9a Inline _set_gc into _set_envs_and_config (#33538) Lianmin Zheng 2026-08-04 07:31:56 -07:00
• d257b58e67 [Router] Report accelerator count in /v1/loads (#33548) Lianmin ZhengandYinghai Lu 2026-08-04 05:55:35 -07:00
• 8f2a3ad6d7 [mem_cache] Label HiCache host pools and clarify post-capture KV sizing logs (#33445) Lianmin Zheng 2026-08-04 04:21:36 -07:00
• 723c277640 [AMD] [Fix] Enable aiter hd256 FP8 prefill FMHA on gfx950 (#33399) jacky.cheng 2026-08-04 17:40:33 +08:00
• 5e6c37f2b4 [cuda_graph] Gate breakable-CG capture_inputs retention to DP-gather paths (#32678) Xingyu Liu 2026-08-04 02:23:34 -07:00
• b57721ccf7 Enable post-capture KV sizing with DP attention (#33427) 3 people 2026-08-04 02:20:24 -07:00
• 16d3b118a2 Reduce startup log noise and fix Dynamo / CUDA-graph edge cases (#33428) Lianmin Zheng 2026-08-04 02:19:56 -07:00
• b6d548afd7 [Fix] Resolve VLM test image placeholders from the model's own chat template (#33509) Liangsheng Yin 2026-08-04 02:01:39 -07:00
• 26a542722f [CI] Replace the rust-ext-build venv instead of failing when it exists (#33512) 3 people 2026-08-04 01:56:16 -07:00
• eaf5c29cc5 [AMD] Enable block-fp8 + quick INT4 all-reduce in MiniMax-M3 MI35x nightly Test (#33402) YC Yen-Ching Tseng 2026-08-04 16:55:06 +08:00
• 53804d609c [CI][XPU] Stabilize XPU CI: pin UMD/IGC, retry infra flakes, right-size EAGLE3 (#32438) ashwini rathiandMa Mingfei 2026-08-04 13:58:52 +05:30
• 17d19081d9 [mm] sglang-mm: server vision pipeline core (fetch/driver/pipeline) + Qwen VL (#32364) Kan WuandClaude Fable 5 2026-08-04 00:47:00 -07:00
• 154f0ac662 Fix DSpark and DP/EP (#33098) Vladislav NosivskoyandXinyuan Tong 2026-08-04 10:35:57 +03:00
• bfa4e4a57b [Nemotron] Hoist mamba track-mask host syncs out of the per-layer prefill path (#32589) Alex NailsandClaude Opus 5 2026-08-03 23:58:28 -07:00
• 23ea7b6481 Prewarm DSV4 MHC post kernel at model load (#30741) weireweireandweireweire 2026-08-04 14:41:56 +08:00
• 157401f050 [CI] Build the Rust extensions on the 5090 pool and seed the cache from main (#33460) Liangsheng Yin 2026-08-03 23:33:20 -07:00
• 101bb2327c [diffusion] fix: fix local-path detection for MiniMax-H3 and other non-diffusers models (#33365) TobyMintandMick 2026-08-04 14:03:35 +08:00
• 4494fb96b2 refactor the tcp listener binding logic (#33420) Rain Jiang 2026-08-03 22:10:52 -07:00
• 48dcadc770 [AMD][DI][CI] 5/N Add DSV4 wide-EP16 4-node 2P1D nightly recipes (#31500) Zhaoyi LiandChen 2026-08-04 00:08:00 -05:00
• afc868517b [Perf] Speed up the Kimi-K2.5 vision path and match PIL bicubic in the GPU resize (#33349) Liangsheng YinandMick 2026-08-03 21:57:49 -07:00
• 7ba393dd15 [CI] Dispatch base-a-test-cpu through its own reusable stage workflow (#33461) Liangsheng Yin 2026-08-03 20:49:23 -07:00
• c6f2a9c1d4 [diffusion] Restrict request-level quality to two validated tiers: lossless (default) and high (#33453) Xiaoyu ZhangandClaude Fable 5 2026-08-04 11:43:16 +08:00
• 614825fd38 [vla] fix: pi05 models does not apply scale factor for language embeddings (#33367) Jinchen HanandMick 2026-08-04 11:25:54 +08:00
• b058dc9106 [diffusion] fix: reject ring parallelism where it would silently miscompute (#33353) Mick 2026-08-04 11:15:23 +08:00
• 7dd8a3d5ea [CI] Fix scheduler max new tokens test fixture (#33467) Mohammad Miadh Angkad 2026-08-04 11:12:21 +08:00
• 1e64fc1563 [Fix] Honor FlashMLA natural-log LSE in DCP reduction (#33065) EchO 2026-08-04 10:55:52 +08:00
• 91fae8a72c [DCP] Bound a request by the aggregate KV pool, not one rank's share (#33448) Khoa Pham 2026-08-03 19:36:25 -07:00
• c113ead98a Bump helion version to 1.4 (#32562) Oguz UlgenandCheng Wan 2026-08-03 19:10:31 -07:00
• 92087ef4d2 fix(mem_cache): state the MLA KV bound in the DCP index space (#33432) Yuwei AnandClaude Opus 5 2026-08-03 19:07:49 -07:00
• eb31a53338 Revert "Add flashinfer rmsnorm + quant fusion support SM90, SM100, SM120" (#33455) Baizhou Zhang 2026-08-03 19:03:56 -07:00
• cdff33d738 [CI] Build the Rust extension modules once per run instead of in every CUDA job (#33384) Liangsheng Yin 2026-08-03 18:56:48 -07:00
• 572924634b [mem_cache] Build empty-prefix last_loc sentinel on-device to avoid per-call H2D sync (#32575) Xingyu Liu 2026-08-03 18:55:03 -07:00
• a84e70eb1e [AMD][DI][CI] 6/N Add Kimi-K2.6 MXFP4 wide-EP16 2P1D nightly recipes (#33333) Zhaoyi Li 2026-08-03 20:51:55 -05:00
• 03f44c978a [AMD][DI][CI] Auto-mount latest host ionic/ibverbs userspace in the MI355X nightly launcher (fix ABI mismatch) (#33374) Zhaoyi LiandChen 2026-08-03 20:14:14 -05:00
• 7f6a2e2b50 [Refactor] Clean up and split DSA indexer (#33443) Baizhou Zhang 2026-08-03 18:05:01 -07:00
• 3960983753 Add flashinfer rmsnorm + quant fusion support SM90, SM100, SM120 (#32994) 3 people 2026-08-03 17:37:42 -07:00