Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• 9ef1830701 [Scheduler] Extract DFlash prefill refill into a standalone MinFreeSlotsDelayer (#29089) Liangsheng Yin 2026-06-23 17:04:55 -07:00
• f444b5897b [Spec][1/N] Decoupled speculative decoding: IPC protocol + cross-process request id + server flags (#27634) Jackey Hua 2026-06-23 17:04:45 -07:00
• 34dd9c28ca [Refactor] Introduce sock_send/sock_recv wrappers for zmq IPC (#29012) Lianmin Zheng 2026-06-23 15:54:36 -07:00
• ecab3f322e Revert "Improve MFU metrics for prefill and verify timing" (#29079) Lianmin Zheng 2026-06-23 15:46:22 -07:00
• 11e7c9e0e6 [misc] Add sglang.bench_one_batch deprecation shim (#29082) Liangsheng Yin 2026-06-23 14:57:35 -07:00
• f74a1722e6 [NVIDIA] Support TF32 matmul to improve MiniMax gate gemm performance (#22744) Trevor Morris 2026-06-23 14:54:54 -07:00
• c864c8d9c2 [misc] Move bench_one_batch into sglang/benchmark/ with a back-compat shim (#28687) Liangsheng Yin 2026-06-23 14:48:35 -07:00
• 6c5f466023 [server_args] Reland FA4 page_size auto-force for combined --attention-backend fa4 (#28976) Liangsheng Yin 2026-06-23 14:21:59 -07:00
• 93015a9e6b fix(runner): autotune flashinfer MoE on a decode-shaped buffer (#29069) YAMY 2026-06-23 13:31:47 -07:00
• fc27ce0666 Add GB10 FP8 fused MoE Triton config (#25665) Jan Bernlöhr 2026-06-23 22:30:51 +02:00
• cedb43d522 [DeepEP] Gate DeepEP MNNVL on fabric support (#28942) Mohammad Miadh Angkad 2026-06-24 04:19:47 +08:00
• b60185c41c Improve MFU metrics for prefill and verify timing (#29000) Lianmin ZhengandPranjal Shankhdhar 2026-06-23 12:26:56 -07:00
• 0c6e8e9477 Expand parser auto detection coverage (#28449) Xinyuan Tong 2026-06-23 20:26:37 +01:00
• e0dc8b7137 [AMD] Fuse topk padded-token masking into a single Triton kernel (#28084) karverma-amdandCursor 2026-06-23 13:15:12 -05:00
• ed26a109ee [Fix] Return streaming logprobs when reasoning/tool parser is active (#28601) 3 people 2026-06-23 09:56:09 -07:00
• 83d32fbc2f remove lora section (#29056) Yihao Wang 2026-06-23 09:26:51 -07:00
• b5e0965b07 [Diffusion][Cookbook] Add Krea-2 cookbook (#29051) Yihao Wang 2026-06-23 08:02:08 -07:00
• 31d71c47af [Diffusion] Fix SANA VAE dtype and TurboWan backend selection (#28769) Xiaoyu Zhang 2026-06-23 22:57:08 +08:00
• 12b08e620b [Diffusion] [NPU] enable Helios on npu (#29011) Thomasandronnie_zheng 2026-06-23 22:25:10 +08:00
• c67d338637 [Diffusion] enable cache-dit for ERNIE-Image model (#28266) Thomas 2026-06-23 21:08:03 +08:00
• 0460f277b7 Fix manual chunked-prefill test to use req.fill_len after fill_ids refactor (#29032) Shangming Cai 2026-06-23 17:52:26 +08:00
• ed0a62e4dd [Mem] Add KV-page double-free checks to the invariant checker (#27731) Liangsheng Yin 2026-06-23 02:29:40 -07:00
• af9027f6c9 [AMD] Improve performance of dsv4 in high concurrency (#28938) kkandwunhuang 2026-06-23 17:20:48 +08:00
• a65c68c1fb Revert "[CI] Fix flaky optimistic test by adding contention handling" (#29023) Shangming Cai 2026-06-23 16:53:32 +08:00
• e67b228d4c feat: session radix cache (#27058) ishandhanani 2026-06-23 10:51:41 +02:00
• 349a6af6b8 [HiCache] Fix hicache host memory leak by bounding PP-sync work_list (#28916) ybyang 2026-06-23 16:39:03 +08:00
• 7b1a20344c Re-enable SM90 FlashInfer allreduce fusion with safe backend defaults (#28789) Mohammad Miadh Angkad 2026-06-23 16:29:19 +08:00
• 854c688121 [Spec] Unify decode KV-commit bookkeeping across spec-v2 workers (#28754) Liangsheng Yin 2026-06-23 00:53:05 -07:00
• 743ce88bc5 Fix flaky optimistic prefill retry test (#28995) cctryandcctry 2026-06-23 00:18:01 -07:00
• 219742c394 [diffusion] optimize: optimize realtime causal attention fastpath (#28760) Mick 2026-06-23 15:15:30 +08:00
• c4376aaa88 [Refactor] Remove dead out_cache_loc_swa buffers (#28968) Cheng Wan 2026-06-23 00:03:30 -07:00
• 52a90c9a36 docs(minimax-m3): use published AMD ROCm images (#28777) zijiexiaandClaude Opus 4.8 2026-06-22 23:48:16 -07:00
• 7e6587c94a [AMD] Update v4 cookbook to clean env vars (#28981) Thomas Wang 2026-06-23 13:16:29 +08:00
• e63b57da0b [Fix] model init / XPU / transformers-v5 / bench-image fixes (#28292) vikram singh shekhawatandClaude Sonnet 4.6 2026-06-23 10:28:13 +05:30
• b42c79c4eb [AMD] Fix AMD extra scheduled runs cancelling each other (#28989) 3 people 2026-06-23 11:48:32 +08:00
• 4740f23e1f Revert "[server_args] compute mem_fraction_static after dp chunked-prefill division" (#28991) Liangsheng Yin 2026-06-22 20:48:16 -07:00
• fc0d8fa262 [Lint] fix main's lint (#28990) Liangsheng Yin 2026-06-22 20:40:46 -07:00
• abb0717174 [CI] Fix lint brought by #27527 (#28988) Yuan Luoandluoyuan.luo 2026-06-23 11:40:06 +08:00
• 6cd8d2869b Vectorize _create_custom_4d_mask in CustomQwen2Decoder (#27527) Chetan Kumar Verma 2026-06-23 08:26:35 +05:30
• 62f7ffc492 feat: add Mooncake group semantics (#26574) 3 people 2026-06-23 10:38:58 +08:00
• 28d5627fd8 [AMD] register kv_canary + mock_model e2e tests to extra-a (1-gpu-small + 2-gpu-large) (#28850) Michael 2026-06-22 19:06:31 -07:00
• 84338df6f0 [NPU] [DOC] Update contribution guide of Ascend NPU (#28909) amote-i 2026-06-23 10:00:12 +08:00
• a17753e449 Fix EAGLE draft graph seq_lens_sum padding (#26880) Terry-UVandhnyls2002 2026-06-23 09:00:25 +08:00
• de3ec2c437 [server_args] compute mem_fraction_static after dp chunked-prefill division (#28884) Xinyuan Tong 2026-06-23 01:59:05 +01:00
• ba9d5aed98 Fix nightly CI test for Kimi K2.5 INT4 + H200 (#28746) Brayden Zhong 2026-06-22 17:27:33 -07:00
• d70726a31a Revert "[server_args] fix FA4 page_size auto-force for combined --attention-backend fa4" (#28972) Liangsheng Yin 2026-06-22 17:01:37 -07:00
• e00703bb2a fix(frontend): return 400 for missing completions json_schema (#28090) Ting SUN 2026-06-23 05:47:05 +07:00
• 6c212a5d6b [server_args] fix FA4 page_size auto-force for combined --attention-backend fa4 (#28825) 3 people 2026-06-22 23:35:10 +01:00
• a32f7111e4 [CI] Disable async-assert for base-a tests in rerun-test (#28965) Liangsheng Yin 2026-06-22 15:29:45 -07:00
• 770d6b2825 [Spec] Add sync-free fast_prefill_plan for EAGLE draft-extend CUDA graph (#28854) Liangsheng Yin 2026-06-22 15:15:15 -07:00
• c0198fc277 [CI] Refactor int checkpoint tests style (#28813) Yuan Luoandluoyuan.luo 2026-06-23 05:59:06 +08:00
• 4f60378ff5 Fix Kimi-VL GPU image preprocessing crash on non-RGB images (#28647) Kevin FlansburgandYuhao Yang 2026-06-22 22:23:33 +01:00
• bbc853df46 fix(schedule_batch): trim stop string when EOS matches in the same step (#28802) Maxwill LinandClaude Opus 4.8 2026-06-22 13:14:00 -07:00
• 7c23d2255a [minimax-m3] Split 1/4: sparse attention ops + JIT kernels + config foundation (#28712) Xinyuan Tong 2026-06-22 21:10:43 +01:00
• b5e4e289b1 [gRPC] Native server: Python bridge entrypoint (2/4) (#23507) Alex NailsandClaude Opus 4.7 2026-06-23 04:08:33 +08:00
• 8dc27f6326 [MoE] dedup triton_kernels backend quant-arg asserts and fill weight dtype guard (#28689) Wenyao Gao 2026-06-22 13:01:21 -07:00
• 669be5448b [cuda graph] Enable prefill piecewise CUDA graph for Cohere2Vision (text path) (#28686) zijiexiaandClaude Opus 4.8 2026-06-22 12:20:14 -07:00
• adc203dfee fix(router): use full conversation for PD chat cache-aware routing (#26263) (#27430) Prajjwal ChittoriandClaude Opus 4.8 2026-06-22 13:55:48 -05:00
• 43b1fb95f6 [CI] Optimize nixl dependency installation script (#28950) Shangming Cai 2026-06-23 01:58:22 +08:00
• 99c18cceec Sync server arguments and environment variables + update various documentation (#28674) 3 people 2026-06-22 10:23:46 -07:00
• 34e5e38604 [Cookbook] Nemotron3-Ultra: Add mamba-backend and SSM dtype flags (#28675) Brayden ZhongandBrayden Zhong 2026-06-22 09:50:22 -07:00
• ab21dc984a [CI] Fix flaky optimistic test by adding contention handling (#28947) Shangming Cai 2026-06-23 00:35:13 +08:00
• 70883cb1b0 [UnifiedTree]: Rollback mamba hicache test to direct io backend (#28904) Zhangheng 2026-06-22 23:48:48 +08:00
• bbe8b7dd8a [diffusion] CI: restore Hunyuan3D-2 image-to-3D (#28781) Mick 2026-06-22 23:36:18 +08:00
• b28e990161 Migrate all ServerArgs fields to Annotated style, reduce add_cli_args by ~2400 lines (#28919) Lianmin Zheng 2026-06-22 08:34:37 -07:00
• 6b2c730bf7 [codex] Fix DSA indexer in prefill piecewise CUDA graph (#28644) Xiaoyu Zhang 2026-06-22 22:39:21 +08:00
• b43bd6824f [B300] Enable FlashInfer allreduce for Qwen3-VL MoE (#28786) Xiaoyu Zhang 2026-06-22 22:38:45 +08:00
• cee1caaf47 [AMD] Fix nightly-8-gpu-mi35x-deepseek-v4-flash-rocm720 OOM issue (#28941) Thomas Wang 2026-06-22 22:35:45 +08:00
• 04d952ea10 [AMD] deepseek-v4 clean env vars (#28920) Thomas Wang 2026-06-22 22:32:43 +08:00
• dd39ef6cec [AMD] Pin httpx>=0.25.0 to fix anthropic SDK socket_options error (#28869) YC Yen-Ching Tseng 2026-06-22 22:25:22 +08:00
• 4923bb93ae [diffusion] CI: fix turbo_wan/flux invisible CI cases (#28913) Mick 2026-06-22 21:21:16 +08:00
• ad9723af03 Clean up CUDA graph capture logs (#28937) Lianmin Zheng 2026-06-22 06:15:26 -07:00
• ead39d38fc [diffusion] refactor: refactor causal KV local head cache updates (#28888) Mick 2026-06-22 21:00:38 +08:00
• 1adb53f147 Fix CP page filtering by request-local position (#28718) shihaozhou 2026-06-22 20:47:29 +08:00
• 06e001347a [CI] Update nixl installation to include nixl-cu13 for h20 runner (#28930) Shangming Cai 2026-06-22 20:35:14 +08:00
• b8e64fb56d [CI] Modify nixl installation to force reinstall (#28927) Shangming Cai 2026-06-22 19:26:26 +08:00
• 0c9e775f2c [Diffusion] Fix FastWan2.1 default 480p resolution (#28733) Xiaoyu Zhang 2026-06-22 18:43:24 +08:00
• 4e1d25117b [NPU] update best practice docs from testcase (#28621) jianzhao-xu 2026-06-22 16:58:48 +08:00
• 2ce32366a0 [Fix][BCG][Spec] Restore EAGLE prefill plumbing dropped by #23906 (#28870) Yuwei An 2026-06-22 01:54:54 -07:00
• 93553a67a3 [NPU] [DOC] Create deployment tutorials for mainstream models on Ascend NPU (#27893) amote-i 2026-06-22 16:21:13 +08:00
• db12bfcdc8 [JIT] Add kpool_topk_transform JIT kernel (#28670) Xinyuan Tong 2026-06-22 09:04:21 +01:00
• 106d2930a6 [core] Gate the overlap WAR barrier on forward reads to recover decode throughput (#28363) Liangsheng Yinandthanhhao98 2026-06-22 00:40:56 -07:00
• 441ae9a5ae [Lint] Fix black formatting of DeepSeek-R1-MXFP4 MI35x tests (#28885) Xinyuan TongandXinyuan Tong 2026-06-22 07:20:34 +01:00
• 62b3c8e177 [Intel GPU] Guard tvm_ffi import in dsv4 online mtp module under TYPE_CHECKING to fix import error on XPU (#28531) Polisetty V R K Jyothendra VarmaandMa Mingfei 2026-06-22 11:26:21 +05:30
• 64e455d4bf Fix lint break on main (#28886) Liangsheng Yin 2026-06-21 22:46:08 -07:00
• e2540188ce [AMD] Clean up DeepSeek-R1-MXFP4 TP2/TP4 MLA GSM8K tests (#27243) Bingxu Chen 2026-06-22 12:41:19 +08:00
• 0642cd5020 (chore): bump tokenspeed_mla to 0.1.7 (#28759) Khoa PhamandCursor 2026-06-21 21:25:37 -07:00
• 018d0c21dc [Docs] Add Anthropic-compatible API documentation (#28522) 4 people 2026-06-22 05:01:09 +01:00
• be774d0acd [docs][cookbook] Laguna-M.1 playground: add HiCache; refresh EP / DP-Attention notes (#28774) 3 people 2026-06-21 20:58:42 -07:00
• 886b96621d Migrate more server args to annotated style (#28830) Lianmin Zheng 2026-06-21 20:50:17 -07:00
• 0c065671c9 [Spec] Redo: split init_backends; account draft weights in --mem-fraction-static (#28855) cctryandcctry 2026-06-21 20:45:16 -07:00
• fd7874d11b [AMD] Register DP attention test (#28495) Bingxu Chen 2026-06-22 11:21:41 +08:00
• 73448b0d70 [AMD] Temporarily disable deepseek V4 in AMD PR test (#28871) YC Yen-Ching Tseng 2026-06-22 10:48:38 +08:00
• 2b2cd21783 [diffusion] fix: reject cache-dit with fsdp (#28834) Mick 2026-06-22 10:47:40 +08:00
• c0bb04b67f [NVIDIA] Support NVFP4 MoE for DeepSeek-V4 (#25820) Trevor Morris 2026-06-21 19:35:14 -07:00
• 5deca2d39f [DOC] [NPU] Update features on Ascend NPU (#28643) amote-i 2026-06-22 09:50:51 +08:00
• 6779ca8d7f Fix Qwen MoE precision issue with PP and all-reduce fusion (#28619) 3 people 2026-06-22 08:20:16 +08:00
• 4f5ff39bc9 [Spec] Enable FR-Spec in EAGLE draft-extend CUDA graph by sizing logits buffer from the draft head (#28856) Liangsheng Yin 2026-06-21 15:35:57 -07:00
• e6722c751b [Feature] Add graceful scheduler shutdown; free hisparse host buffer on exit (#28779) Liangsheng Yin 2026-06-21 15:08:10 -07:00
• 8e890391f5 [Spec] Support FlashInfer CUDA graph for EAGLE draft-extend (#28782) Liangsheng Yin 2026-06-21 14:46:25 -07:00