Commit Graph

Select branches

Hide Pull Requests

bot/update-est-time-20260928

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• 977830e91e ci(deepseek-v4): add b300/grace-blackwell dev-branch build options (#23778) Baizhou ZhangandClaude Opus 4.7 2026-04-26 16:44:16 -07:00
• 35591c7d51 fix(lora): don't assert on non-LoRA lm_head adapter weights (#23433) Kangyan-ZhouandClaude Opus 4.7 2026-04-26 12:10:07 -07:00
• 3d95ca7546 docs(DeepSeek-V4): mark gb200|big|low-latency verified (#23737) zijiexia 2026-04-26 11:15:59 -07:00
• a392ae8879 [diffusion] feat: accelerate multiple-outputs generation (#23759) Mick 2026-04-27 01:47:33 +08:00
• 10fd0faccd [CPU] Add Qwen3.5 model optimization for CPU (#19484) 4 people 2026-04-27 01:12:36 +08:00
• 7d49564431 [NPU]Fix support_triton bug (#23604) Liwansi 2026-04-26 21:34:56 +08:00
• c7878dbb6d [MoE] Deprecate act_and_mul_triton; fold filter_expert into JIT silu/gelu_and_mul (#23707) Cheng WanandClaude Opus 4.7 2026-04-26 01:41:35 -07:00
• d49a0377de [diffusion] refactor: make timestep scheduler request-local (#23716) Mick 2026-04-26 15:59:53 +08:00
• 9003f24e2b chore: bump sglang-kernel version to 0.4.1.post1 (#23733) 4 people 2026-04-25 23:23:49 -07:00
• 8efa177f1e [CI] release-pypi-nightly: install protoc before building wheel (#23750) Kangyan-ZhouandClaude Opus 4.7 2026-04-25 22:37:29 -07:00
• 4be853e1c6 [CI] release-whl-kernel: strip +cu129 local version before PyPI upload (#23749) Kangyan-ZhouandClaude Opus 4.7 2026-04-25 22:14:45 -07:00
• 3cfd1561df docs(DeepSeek-V4): add h200|big verified recipes + tune H200 Pro parameters (#23742) Ethan (Yusheng) Su 2026-04-25 21:44:51 -07:00
• 282b47fcfd [CI] release-whl-kernel: clean root-owned build artifacts before checkout (#23747) Kangyan-ZhouandClaude Opus 4.7 2026-04-25 21:39:41 -07:00
• 049f1bf6fb docs(DeepSeek-V4): add GB200 platform to cookbook recipe (#23725) Yuhao Yang 2026-04-26 11:54:55 +08:00
• ba4e9d2ac2 Apply should_use_dp_reduce_scatterv guard to remaining MoE models (follow-up to #23731) (#23732) 4 people 2026-04-25 20:36:16 -07:00
• 71029abd64 Fix Qwen3 MoE: also guard EP all-reduce with not use_reduce_scatter (follow-up to #23731) (#23734) 3 people 2026-04-25 20:35:52 -07:00
• 714173555c chore: bump sgl-kernel version to 0.4.1.post1 (#23720) 4 people 2026-04-25 17:13:02 -07:00
• 99b59b279c Fix Qwen3 MoE double-reduce when DP attention + EP + reduce_scatterv (#23729) (#23731) Byron HsuandByron Hsu 2026-04-25 15:28:28 -07:00
• 921e14dcac [CI] release-docker-deepseek-v4: select which flavors to push (#23730) Kangyan-ZhouandClaude Opus 4.7 2026-04-25 14:09:34 -07:00
• 0c826374a8 ci: add docker release workflow for deepseek_v4 branch (#23728) Kangyan-Zhou 2026-04-25 11:45:23 -07:00
• acaa35664d [CI] sgl-kernel: prune dangling images before each wheel build (#23723) Kangyan-ZhouandClaude Opus 4.7 2026-04-25 10:44:47 -07:00
• e0a4522370 [typo] fix typo in parallel_state (#23710) AlbeeSo 2026-04-26 00:33:33 +08:00
• 03849496ad jit_kernel: tolerate FA3 kernels without out arg (#23717) Mick 2026-04-25 23:42:33 +08:00
• d4c1665626 docs(DeepSeek-V4): mark h200|big|pd-disagg verified + recipe fixes (#23715) fzyzcjy 2026-04-25 22:49:12 +08:00
• 046c14a3ed [NPU] Support GGUF quantization for Ascend NPU (dense + MoE) (#17883) 1874.andronnie_zheng 2026-04-25 22:16:47 +08:00
• e708ea6d94 [diffusion] fix: restore cache-dit support for LTX2 (#23235) gjsheuandgengjinsong 2026-04-25 18:10:43 +08:00
• 50ce2708ca [diffusion] fix: Fix FLUX.1/2 graph breaks (#23648) Aleksi Vesanto 2026-04-25 12:54:52 +03:00
• 880599cd43 docs(DeepSeek-V4): bump GB300 Pro PD decode --mem-fraction-static 0.83 → 0.9 (#23698) fzyzcjy 2026-04-25 16:35:44 +08:00
• 11d77a60df [NPU] [DOC] Update supported models and features of npu (#23564) amote-i 2026-04-25 15:37:07 +08:00
• 393252f514 [AMD] fused qk gemma norm kernels to reduce four kernels (#23575) kkandroot 2026-04-25 15:30:01 +08:00
• bd523dd60d [NPU] [Bugfix] [Diffusion] Fixed gray images at the generation output (#23266) Артем Савкинandronnie_zheng 2026-04-25 10:20:38 +03:00
• 6175946db7 [Feature]Add MSProbe dump support in SGLang (#18349) Yujing 2026-04-25 15:12:50 +08:00
• 21835fb0af [HiCache] Prevent move_hybrid_indices from polluting radix-tree node host state (#23427) Yujun Dongandhzh0425 2026-04-25 14:27:42 +08:00
• 82254bd9c5 [JIT Kernel] Reland JIT activation (#22094) 4 people 2026-04-25 14:00:28 +08:00
• 0d224e5053 update: b300 container for dsv4 (#23697) ishandhananiandBaizhou Zhang 2026-04-25 00:59:20 -05:00
• adc59325bc [AMD] Optimize MiniMax-M2.5 - enable fused Triton kernel for FP8 KV cache write in aiter decode path (#23620) YC Yen-Ching Tseng 2026-04-25 13:23:49 +08:00
• fb272d27db [AMD] Optimize MiniMax-M2.5 - use aiter biased_grouped_topk for sigmoid scoring in MoE routing (#23611) YC Yen-Ching Tseng 2026-04-25 13:18:08 +08:00
• 8471c9ebe6 Skip torch.cuda.empty_cache() in weight update flush path (#22998) Shenxiu Liu 2026-04-24 21:41:39 -07:00
• 69485a176c Small udpate gb300 recipe for deepseek v4 (#23690) Baizhou Zhang 2026-04-24 21:35:16 -07:00
• 8a395994ed docs(DeepSeek-V4): mark gb300|{small,big}|{cp,pd-disagg} verified + GB300-specific fixes (#23691) fzyzcjy 2026-04-25 12:21:57 +08:00
• d2c61acf25 docs(DeepSeek-V4): mark b200|small|pd-disagg + h200|small|{cp,pd-disagg} verified (#23689) fzyzcjy 2026-04-25 11:57:25 +08:00
• fd401c2fb4 docs(DeepSeek-V4): note SGLANG_FIX_DSV4_BASE_MODEL_LOAD for base models (#23684) fzyzcjy 2026-04-25 11:27:00 +08:00
• 4a3fe2a091 model: support parakeet nemotron encoder (#23568) Yuhao Yangandtrangdough 2026-04-25 11:00:23 +08:00
• 465abadd3c Add fused moe triton config for Qwen3.5-397B-A17B-FP8 (#23682) Jackey Hua 2026-04-24 21:35:32 -04:00
• a4facdf3f6 [CI] Refactor ci_install_dependency.sh into standalone functions (#23592) Lianmin Zheng 2026-04-24 17:39:39 -07:00
• f30a6f4d7e [DOC] Add DFLASH speculative decoding documentation (#23553) shuwenn 2026-04-25 08:18:46 +08:00
• 76da28f6d6 [AMD][bugfix] add gate rocm >= 7.2 for bpreshuffle (#23671) Xinyi Song 2026-04-24 13:26:16 -07:00
• f7e840682c [AMD][MoRI] bump MoRI to v1.1.1 (#23642) jhchouuu 2026-04-25 04:12:20 +08:00
• 587fd15bd2 perf: eliminate attention DtoD copy by passing pre-allocated output to FA (#21985) Jia GuoandClaude Opus 4.6 2026-04-24 12:05:16 -07:00
• 6d03861476 support Hy3 preview (#23533) 5 people 2026-04-24 20:03:24 +01:00
• 6344b546c8 Deprecate --collect-tokens-histogram, auto-collect with --enable-metrics (#23595) Lianmin Zheng 2026-04-24 12:00:16 -07:00
• 05696527ea [diffusion] feat: support LoRA for LTX2.3 (#23649) Mick 2026-04-25 01:52:41 +08:00
• baa0aa670f [HiCache & HybridModel] 3FS backend support DSA & mamba model (#23241) 3 people 2026-04-25 00:48:01 +08:00
• 92d262f710 [diffusion] RL: add per-step rollout options for SDE and trajectory capture (#23151) Kangrui Du 2026-04-24 08:26:16 -07:00
• bca3dd958a [Intel GPU] Enable pipeline parallelism on XPU (#23645) Siju Samuel 2026-04-24 17:22:44 +05:30
• 60bbb800db [Experimental] Breakable Piecewise Cuda Graph (#22218) Yuwei AnandClaude Opus 4.6 2026-04-24 04:33:05 -07:00
• b3b03369a5 [diffusion] fix: unify LTX-2.3 HQ codepath gates for all LTX-2.3 variants (#23624) Mick 2026-04-24 17:44:08 +08:00
• b060a5ccfd [AMD] Fix nightly version tag selection (#23644) YC Yen-Ching Tsengandbingxche 2026-04-24 17:39:47 +08:00
• b8d883398d Revert "[Intel GPU] Enable pipeline parallelism on XPU" (#23641) Shangming Cai 2026-04-24 17:36:35 +08:00
• 1758856762 [CI] Fix mxfp8 TrtllmGenMoe test (#23125) Ziang LiandBrayden Zhong 2026-04-24 02:02:11 -07:00
• 92bb5c6bbe Update pro fp8 checkpoint in DeepSeek V4 cookbook (#23634) fzyzcjy 2026-04-24 15:58:04 +08:00
• 3a620cb761 Again update DeepSeek V4 cookbook (#23622) fzyzcjy 2026-04-24 15:12:35 +08:00
• 1a37e57fb1 [codex] docs: note H200 DeepSeek-V4 checkpoint (#23628) zijiexia 2026-04-24 00:06:30 -07:00
• 4cb0c4e1f3 [AMD] Fix memory access fault when --page-size > 1 with speculative decoding on AMD GPUs (#23596) Hubert Lu 2026-04-23 23:56:36 -07:00
• cd1fa7506a [diffusion] model: support LTX2.3 high quality pipeline (#23366) Mick 2026-04-24 14:18:20 +08:00
• 734e1e2965 Further update Deepseek V4 docs (#23617) fzyzcjy 2026-04-24 13:23:50 +08:00
• 492883c8ca Add DeepSeek V4 cookbook (#23605) 3 people 2026-04-23 22:10:28 -07:00
• 30909cbeeb [AMD] upd local registry address (#23607) YC Yen-Ching Tsengandbingxche 2026-04-24 12:07:09 +08:00
• 59724e90a9 model: support Moss-VL (#23454) Shaojun Zhou 2026-04-24 11:14:29 +08:00
• bf98eb3ab7 [Intel GPU] Enable pipeline parallelism on XPU (#23472) Siju SamuelandShangming Cai 2026-04-24 08:11:51 +05:30
• b35213be11 [MUSA][16/N] Add MUSA backend support for layers and DeepSeek models (V2/V3/R1) (#22774) 3 people 2026-04-24 09:59:51 +08:00
• cbc2bee547 [Intel CPU/XPU] SGL doc updates (#23547) Zaili Wang 2026-04-24 09:23:27 +08:00
• 23e4d381f0 [CPU] remove RECORD_FUNCTION (#23528) Ma Mingfei 2026-04-24 09:18:29 +08:00
• 87e50f20f6 [Apple Silicon][MLX] Cache seq_lens-derived tensors in BatchedDecodeContext (#23470) R0CKSTAR 2026-04-24 09:12:26 +08:00
• 74c2e5bacd [MUSA][8/N] Port CUDA kernels that are compatible with MUSA (#17946) MARATRIXandAlex Nails 2026-04-24 09:04:58 +08:00
• c0166355ae [diffusion] CI: minor refactor CI (#23576) Mick 2026-04-24 08:48:31 +08:00
• d9c72bdd2b Skip unselected experts in flashinfer_trtllm (#23493) Cheng Wan 2026-04-23 17:30:19 -07:00
• 000a2525e1 Move expert_mask_gpu from FusedMoE layer to StandardDispatcher (#23585) Cheng WanandClaude Opus 4.7 2026-04-23 17:17:27 -07:00
• 95d021b523 Pre-set SWA cache location in CudaGraphRunner (#23552) Lianmin Zheng 2026-04-23 16:51:29 -07:00
• bb962b0046 Fix MoE no_combine: skip router weight in down projection (#23545) Lianmin Zheng 2026-04-23 16:47:58 -07:00
• cf88fdcc9c Expose child process PIDs from Engine for health check support (#23320) Sundara Raman Ramachandran 2026-04-23 16:44:49 -07:00
• 9891572c4a [CI] Export GB200 nightly logs to S3 (#23502) Sahithi Chigurupati 2026-04-23 15:35:10 -07:00
• f3b88e080a chore: bump flashinfer version to 0.6.8.post1 (#23281) sglang-botandsglang-bot 2026-04-23 15:23:03 -07:00
• 6428392b6f ci: fix cu129 wheel tagging + pipefail-abort in install script (follow-up to #23497) (#23587) Jia Guo 2026-04-23 14:52:58 -07:00
• 17210350fd [PD+DP] Allow PrefillDelayer in disaggregated-prefill mode (#23588) Byron Hsu 2026-04-23 14:51:16 -07:00
• 2882a136bf [CI] Consolidate Docker release workflows into reusable workflow (#22541) Kangyan-ZhouandClaude Opus 4.6 2026-04-23 14:06:01 -07:00
• 579bd0b152 [bug fix] has_fp8_weights_in_checkpoint: handle HF repo IDs, not just local paths (#23542) Alex NailsandClaude Opus 4.7 2026-04-23 12:56:49 -07:00
• 9b2f7f8a91 docs: split MI300X and MI325X options in GLM-5.1 generator (#23540) zijiexiaandgemini-code-assist[bot] 2026-04-23 12:01:58 -07:00
• 80125febb1 [BUGFIX]Fix Ascend backend pre-allocated range in NPU Graph Mode. (#22778) WangHao-hw 2026-04-24 01:23:35 +08:00
• c6872fc8fb Fix: fallback to torch API when NVML memory query is not supported (#23426) Jinghong Liandronnie_zheng 2026-04-24 00:26:04 +08:00
• 86ed0680d7 feat: add OpenTelemetry tracing to DiffGenerator (#21254) Jie Hao 2026-04-23 12:25:23 -04:00
• 76e4c5a1f8 [Diffusion][NPU][Bugfix] Ascend_fa crashes when sequence parallelism is used. (#23572) Arseniy MironovandNapkin-AI 2026-04-23 19:21:30 +03:00
• 54e21bb3a5 [fix] Fix dynamic chunking profiling crash on GLM-5 models (#23060) Baichuanandliubaichuan 2026-04-23 19:30:57 +08:00
• 4868e367f8 docs: add Hunyuan 3 Preview cookbook (#23532) Xinyuan Tong 2026-04-23 10:44:47 +01:00
• cd459af4e2 [AMD] Use bpreshuffle FP8 blockscale GEMM to replace ABScale GEMM (#23319) Xinyi SongandHaiShaw 2026-04-23 01:51:30 -07:00
• fd88a1c562 [AMD] skip deterministic inference for MLA FP8 test (#23382) Bingxu ChenandYC Yen-Ching Tseng 2026-04-23 15:43:23 +08:00
• 1340d8b3bd [CI][NPU]use rsproxy.cn mirror to speed up Rust toolchain installation on NPU runners (#23514) Goalina 2026-04-23 14:52:14 +08:00
• 2ef1a21d5e [bug fix] fix: detect FP8 weights from safetensors header instead of ass… (#23414) Ethan (Yusheng) Su 2026-04-22 23:49:57 -07:00
• f1a70b4666 [Observability] Add HTTP sidecar endpoints and FlushCache gRPC RPC for gRPC mode (#22500) Kangyan-ZhouandClaude Opus 4.6 2026-04-22 23:06:10 -07:00
• 3c5b1f0810 [diffusion] fix: fix --warmup-resolutions hang with --enable-cfg-parallel (#23198) mispa-ms 2026-04-22 22:39:20 -07:00