Commit Graph

Select branches

Hide Pull Requests

bot/update-est-time-20260928

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• ebd64f5d40 Support user-supplied recv-side transform in dumper grafter (#24509) fzyzcjy 2026-05-06 16:56:52 +08:00
• 9a65f0ac26 Support t2b direction and overlap protection in dumper grafter (#24508) fzyzcjy 2026-05-06 16:56:24 +08:00
• 58487e68e5 Support cross-system tensor grafting in dumper (#24507) fzyzcjy 2026-05-06 16:55:40 +08:00
• 61104d7d0a Add prefixed _log helper in dumper (#24506) fzyzcjy 2026-05-06 16:54:20 +08:00
• 23172224b9 chore: Gitignore Claude scheduled_tasks.lock (#24505) fzyzcjy 2026-05-06 16:53:56 +08:00
• 094b90b1ec ci: drop 1-gpu-h100-h200 shared label (#24495) Alison Shao 2026-05-06 01:02:31 -07:00
• fbebfdec9a [diffusion] fix: fix diffusion FSDP sharding (#24431) Mick 2026-05-06 14:55:51 +08:00
• 660a77f221 Silence noisy health-check race log in TokenizerManager (#24466) cctry 2026-05-05 21:06:43 -07:00
• 3da87902d7 [HiSparse] Support FP8 KV cache by routing to flashmla_kv backend (#23013) 4 people 2026-05-06 11:18:30 +08:00
• b2420d72ff [RL] DeepEP support for --enable-return-routed-experts (#16859) 3 people 2026-05-05 20:01:07 -07:00
• d7385b575f [Diffusion] Optimize Hunyuan3D shape denoising (#24287) Xiaoyu Zhang 2026-05-06 10:10:09 +08:00
• c7019ff33d [NIXL][XPU] Use np.uint64 for pointer/length arrays in disaggregation KV transfer (#24188) Jianhong Zhang 2026-05-05 19:09:03 -07:00
• a965f886bf [misc] update CI_PERMISSIONS.json (#24468) Qiaolin Yu 2026-05-06 09:37:18 +08:00
• b91b05ae27 Add --random-input-len to send_one.py (#24464) Lianmin Zheng 2026-05-05 17:49:33 -07:00
• 22cf7d2b42 [Fix] Handle nixlRemoteDisconnectError in NixlKVSender (#24296) cctry 2026-05-05 17:23:42 -07:00
• 64f80eabbe Register aten::rms_norm and aten::mm.dtype in batch invariant mode (#24459) Lianmin Zheng 2026-05-05 17:21:34 -07:00
• 46bde1f426 Add fwd_occupancy metric to SchedulerStats and Prometheus collector (#24458) Lianmin Zheng 2026-05-05 17:04:34 -07:00
• 1e404afec2 fix(req_pool): bump pool.size to match actual tensor row count after #24243 (#24439) Xinyuan Tong 2026-05-06 00:58:26 +01:00
• 710fed10fb Revert "[fix] /pause_generation and /continue_generation wrong for --tokenizer-worker-num > 1" (#24461) Lianmin Zheng 2026-05-05 16:44:34 -07:00
• 431ca54334 [fix] /pause_generation and /continue_generation wrong for --tokenizer-worker-num > 1 (#24445) maocheng23andCursor 2026-05-05 16:27:58 -07:00
• 6764155914 chore: bump sgl-kernel version to 0.4.2.post1 (#24457) 3 people 2026-05-05 16:19:58 -07:00
• 08d4c2072b move topk capturers to srt/state_capturer/ (#24450) 4 people 2026-05-05 15:54:01 -07:00
• 47a416fc62 add indexer-topk capture (V3.2 NSA + infra) (#24392) Liangsheng Yin 2026-05-05 15:05:15 -07:00
• 83b48fd523 [codex] update Nemotron3 Nano Omni cookbook benchmarks (#23998) zijiexia 2026-05-05 14:53:15 -07:00
• e0faed8ebc [Docs] Add B200, GB200, GB300 NVIDIA hardware platform support for Kimi-K2.6 (#24441) zijiexia 2026-05-05 14:53:05 -07:00
• c4c0376fcb consolidate routed-experts capturer onto reusable base (#24403) Liangsheng Yin 2026-05-05 12:41:49 -07:00
• 8d1b6f0c00 Add zRzRzRzRzRzRzR to CI permissions (#24432) Xinyuan Tong 2026-05-05 20:37:50 +01:00
• d23ef408f7 [diffusion] fix: fix RowParallel LoRA merged forwarding (#24410) Mick 2026-05-06 00:30:16 +08:00
• 932d89690a Gemma4-mtp cookbook (#24433) Khoa Pham 2026-05-05 08:04:41 -07:00
• cc54d8e8d0 [diffusion] chore: clean CUDA cache only at explicit release points (#24397) Mick 2026-05-05 22:30:43 +08:00
• fdfc46f3a5 [Intel GPU] Enable DeepSeek V3.2 inference on XPU (#24356) Polisetty V R K Jyothendra Varma 2026-05-05 18:17:40 +05:30
• e299ec1bff [UnifiedRadixTree]: Fix flaky ci (#24421) Zhangheng 2026-05-05 20:22:19 +08:00
• d22853480d Fix deterministic inference on models with SWAKVPool (#24395) Khoa Pham 2026-05-05 05:20:46 -07:00
• 9fb9a1cca6 [sgl] expose swa and mamba cache metrics (#24396) Bi Xue 2026-05-05 05:19:50 -07:00
• 67e8bd7a80 [codex] Optimize Helios fused norm modulation (#24059) Xiaoyu Zhang 2026-05-05 19:28:37 +08:00
• 8c703f215e Add HunyuanVideo ModelOpt FP8 diffusion support (#23199) Xiaoyu Zhang 2026-05-05 19:27:28 +08:00
• 80ccb6b93c [AMD] fix tbo specv2 seq_lens_cpu NoneType error (#24319) billishyahao 2026-05-05 16:54:43 +08:00
• e0af36a907 [Fix] Fix pypi release workflow (#24417) Baizhou ZhangandClaude Opus 4.7 2026-05-05 01:21:06 -07:00
• 177babcc38 [diffusion] optimize: fuse LTX2 split rotary embedding (#24411) Mick 2026-05-05 16:07:40 +08:00
• 64613543ea [AMD] fix(docker): unbreak nightly when archive.ubuntu.com:80 is unreachable (#24407) YC Yen-Ching Tseng 2026-05-05 15:39:14 +08:00
• c2db19ffa4 [AMD] Enable EAGLE speculative decoding for Qwen3.5 FP8 and MXFP4 models with aiter's unified attention (#23146) 3 people 2026-05-05 00:09:40 -07:00
• 244531bc4f [AMD] Add Kimi-K2.6 in nightly tests for MI30x and MI35x (#23848) Michael 2026-05-05 14:37:14 +08:00
• 6279aee716 [docs] Update B300 Pro cookbook with accuracy-verified serving configs (#24367) Yuhao YangandClaude Opus 4.6 2026-05-05 14:26:03 +08:00
• 04926e1d9f [diffusion] feat: cache encoder results for default negative prompt (#24304) Mick 2026-05-05 11:56:01 +08:00
• e483e60b72 [diffusion] CI: pin diffusion consistency GT revision (#24400) Mick 2026-05-05 11:53:22 +08:00
• 2b769d37a4 (2/n - prefill optimize)perf(lora): remove GPU-CPU sync barrier (.item()) in MoE LoRA path and remove duplicate code (#24246) Ethan (Yusheng) SuandCursor 2026-05-05 03:11:28 +02:00
• 2f7d99b7f7 [diffusion] cli: support component attention backend overrides (#24320) Mick 2026-05-05 08:39:27 +08:00
• 078f84d80d [SKILL] Add diffusion benchmark presets for edit and Hunyuan3D models (#24288) Xiaoyu ZhangandBBuf Codex 2026-05-05 08:18:12 +08:00
• 4b487ca98b [Fix] NGRAMWorker.update_weights_from_tensor — delegate to target worker (#24344) Ji Zeng 2026-05-05 07:23:17 +08:00
• 6a62eabed6 consolidate NSA pool construction (#24389) Liangsheng YinandZhiqiang Xie 2026-05-04 16:04:31 -07:00
• d7c93e183b [sgl] reduce specdec cpu overhead (#23321) Tarushii Goel 2026-05-04 15:02:03 -07:00
• 4743cf6051 misc: add marlin to moe runner choices; drop dead env var doc (#24384) Liangsheng Yinandfzyzcjy 2026-05-04 15:01:47 -07:00
• f059e0264a Fix sgl-deep-gemm release workflow (#24385) Baizhou ZhangandClaude Opus 4.7 2026-05-04 14:36:59 -07:00
• 29dd3a36c0 Refactor device timer installation and rename prefill prealloc to bootstrap (#24341) Lianmin Zheng 2026-05-04 13:57:13 -07:00
• e93bb638bf [docs]update sm75 to sm80 min gpu support (#24336) Praneth Paruchuri 2026-05-05 02:20:20 +05:30
• 0f283a515a [Docker] fix: install nixl stub alongside nixl-cuXX binary (#24369) Kangyan-ZhouandClaude Opus 4.7 2026-05-04 12:46:38 -07:00
• 60a1dacd89 [HiCache] return cached_tokens_details in sglext for streaming responses (#22055) Vladislav Nosivskoy 2026-05-04 22:30:17 +03:00
• 8ffd39e140 [CI] Exclude flaky h20 stage from check-stage-health root cause set (#24374) Kangyan-ZhouandClaude Opus 4.7 2026-05-04 11:41:25 -07:00
• 6dd7aebb36 Minor scheduler fixes (#24359) Ke Bao 2026-05-05 02:01:23 +08:00
• e6f252e9b8 Cache FlashInfer autotune configs (#24156) Sam Shleifer 2026-05-04 11:00:40 -07:00
• c545a5b1c0 [SMG][CI] Add K8s integration tests + wire into pr-test-rust (#24278) Kangyan-ZhouandClaude Opus 4.7 2026-05-04 10:27:37 -07:00
• e5c58eb9d6 [VLM] Optimize Gemma4 VLM with PCG and fuse RMSNorm + residual add + scalar (#24048) Yuan Luoandluoyuan.luo 2026-05-05 00:36:26 +08:00
• 1be3163011 [diffusion] fix: use direct all-to-all for USP collectives (#24366) Mick 2026-05-05 00:08:48 +08:00
• 62a4df0067 [docker] Fix silently-masked cubin download failure; skip prebuilt cubins on aarch64 (#24234) mispa-msandClaude Opus 4.7 2026-05-04 09:08:01 -07:00
• 4b6d44641b [diffusion] chore: enable channels-last 3D VAE convs by default (#23200) Xiaoyu Zhang 2026-05-04 22:59:31 +08:00
• 05aed5e1d5 [UnifiedRadixTree]: Add KL accuracy CI for UnifiedTree with HiCache (#24346) Zhangheng 2026-05-04 20:18:10 +08:00
• 84f3b44916 [tiny] misc cleanups across configs, attention, jit_kernel (#24350) Liangsheng Yin 2026-05-04 03:17:14 -07:00
• 952b3caf18 feat: use structural tags to enable strict tool calling and reasoning for more models (#21722) 4 people 2026-05-04 17:30:28 +08:00
• ef2b1b6d89 Fix flashinfer workspace OOM (#24172) Khoa Pham 2026-05-04 01:26:35 -07:00
• de08c804c5 Add release workflow for sgl-deep-gemm wheels (#24348) Baizhou ZhangandClaude Opus 4.7 2026-05-04 00:58:44 -07:00
• 52b4609789 [Docker] Prep for torch 2.11: cu129 fix, image validator, dep cleanup (#23593) 3 people 2026-05-04 00:37:55 -07:00
• 5eff3c489a [AMD] Deepseek v4 Flash / Pro nightly tests for MI35x ROCm 7.2 (#24203) Bingxu ChenandYC Yen-Ching Tseng 2026-05-04 15:02:33 +08:00
• aea527afdc Fix swa chunk req deferred (#24318) Ke Bao 2026-05-04 14:52:15 +08:00
• c2d60c3667 chore(codeowners): add @kpham-sgl as codeowner for ngram files (#24349) Khoa PhamandClaude Opus 4.7 2026-05-03 23:27:33 -07:00
• a91ae6af9e nextn subclass owns post_load_weights is_nextn (#24333) Liangsheng Yin 2026-05-03 22:04:44 -07:00
• 1dd8f6d5ae dedup state_kv_args setup into helper (#24340) Liangsheng Yin 2026-05-03 20:45:26 -07:00
• 91fa2340ed extract adjust_hybrid_swa_layers_for_pp (#24334) Liangsheng Yin 2026-05-03 18:52:54 -07:00
• b7fefc0e85 feat(lora): enable csgmv backend with virtual experts for MoE LoRA (#24007) Ethan (Yusheng) Su 2026-05-04 03:44:17 +02:00
• c611a3fb78 [diffusion] chore: disable VAE cpu offload by default (#24315) Mick 2026-05-04 08:24:51 +08:00
• 00d620b77d introduce arg_groups/ with nemotron_h hook (#24328) Liangsheng Yin 2026-05-03 16:28:11 -07:00
• c3b6d20a80 Register deepseek_v32 alias instead of rewriting config.json (#24295) Liangsheng Yin 2026-05-03 16:02:17 -07:00
• 53df43d0a3 rerun-test: route deepep h200 suite to deepep runner (#24325) Liangsheng Yin 2026-05-03 15:57:53 -07:00
• 9a5450ad73 [PD]: Support incremental transfer for mooncake transfer engine (#24257) ZhanghengandShangming Cai 2026-05-04 00:57:59 +08:00
• 62265ca7fc [diffusion] feat: initial support for dynamic batching (#18764) Chi McIsaacandJunhao Liu 2026-05-03 12:44:42 -04:00
• f2d1390909 [Diffusion] Add Qwen Image ModelOpt FP8 support (#23155) Xiaoyu ZhangandMick 2026-05-04 00:24:22 +08:00
• 5925572c95 [diffusion] CI: switch CI data references to sgl-project/ci-data (#24299) Mick 2026-05-03 23:05:12 +08:00
• c0f5950636 [UnifiedRadixTree]: Support HiCache Framework for UnifiedRadixTree (#23316) 3 people 2026-05-03 22:13:22 +08:00
• e37f46fcf7 [NPU] Fix Z-Image negative-branch rotary embeddings for CFG (#23538) GXINandronnie_zheng 2026-05-03 21:18:26 +08:00
• 44ca2d01fc [pd]: (Bug Fix) Incorrect out_cache_loc slicing in prepare_for_prebuilt (#24230) ZhanghengandShangming Cai 2026-05-03 18:35:16 +08:00
• 2bfc5d3bb1 [diffusion] optimize LTX2.3 HQ denoising split passes (#24298) Mick 2026-05-03 16:37:46 +08:00
• 8b12367b92 [diffusion] CI: add diffusion GT generation (official implementation) to the workflow (#24270) Mick 2026-05-03 15:10:35 +08:00
• fcc8b7b126 Rename SGLANG_USE_JIT_ALL_REDUCE to SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2 (#24297) Liangsheng YinandDarkSharpness 2026-05-02 23:59:46 -07:00
• b7d8ceb444 [CI] Keep custom sgl-kernel wheel in CUDA CI (#24291) Mohammad Miadh Angkad 2026-05-03 12:53:21 +08:00
• 266d994368 [CI] Fix nightly NV jobs cancelling each other via shared concurrency groups (#24282) Kangyan-ZhouandClaude Opus 4.7 2026-05-02 21:53:03 -07:00
• 76b9c8de6f [Feature] add LoRADrainer to address high P99 TTFT (#17913) Glen Liu 2026-05-02 19:13:43 -04:00
• 200944b415 Update kernel installation instructions after shifting default cuda to 13 (#24181) Baizhou Zhang 2026-05-02 16:07:20 -07:00
• 88bb5dffe4 [Dependency] Upgrade to Torch 2.11.0 (#21247) 6 people 2026-05-02 15:25:36 -04:00
• 24a6b3084d [CI] drop --prerelease allow from uv pip install suffix (#24265) 3 people 2026-05-02 12:25:01 -07:00
• d74d9bda49 [gateway] Align /v1/loads and /model_info with sglang server; drop dead /rerank (#24167) Kangyan-ZhouandClaude Opus 4.7 2026-05-02 11:43:30 -07:00
• e0474fdd9b throw ValueError for DoRA adapters (#22125) Glen LiuandEthan Su 2026-05-02 10:54:19 -04:00