Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• 6cfdc18585 HiCache: Fix Flaky CI For 3FS Backend (#27358) Zhangheng 2026-06-05 22:00:19 +08:00
• c06802dc16 Fix customized_info incremental streaming (#27205) Aurick Qiao 2026-06-05 06:55:01 -07:00
• faa6286946 [BugFix]: Fix HiMamba HiCache prefetch hang after L3 sidecar transfer (#27366) Zhangheng 2026-06-05 20:37:40 +08:00
• e1955bf57a fix(pd): clear stale bootstrap_room when freeing metadata buffer slot (#27374) XiaoTianandgongxiaotian 2026-06-05 19:46:35 +08:00
• 7f919edf00 [AMD] Support alt stream for Qwen3.5 on AMD platform (#25885) Wang, FangYuan 2026-06-05 19:38:25 +08:00
• 5d691a44f4 [diffusion] fix: fix LingBot World timestep error on MUSA (#27341) R0CKSTAR 2026-06-05 19:15:08 +08:00
• 8c8281801d [AMD] update ROCm AITER commit (#27376) Bingxu Chen 2026-06-05 18:42:16 +08:00
• d8487bad06 Update best practice for qwen3-next-80b-a3b-instruct (#27353) McZyWu 2026-06-05 17:02:38 +08:00
• 4ef081b903 [diffusion] optimize: optimize LingBot realtime transport and camera conditioning (#27297) Mick 2026-06-05 16:00:48 +08:00
• 00fefef16b [PD & HiSparse] Add DeepSeek V4 support for HiSparse direct Prefill-to-Decode DRAM (#24880) huangtingwei 2026-06-05 15:39:48 +08:00
• 66b932154f [AMD] fix(ci): run partition 3 of stage-c-test-large-8-gpu-amd (#27352) 3 people 2026-06-05 15:06:36 +08:00
• 4248695b07 [NPU] add GLM model best practice docs (#27032) jianzhao-xu 2026-06-05 14:27:19 +08:00
• 4df1ccdadc [UnifiedTree]: Fix CP Reduce For L3 HiCache (#27330) Zhangheng 2026-06-05 14:03:54 +08:00
• bd47869ba4 [perf] parallelize create_flashmla_kv_indices over page-blocks (#27320) Qiaolin Yu 2026-06-04 22:11:43 -07:00
• 6cbc035dc9 FrozenKVMTPVerifyInput: add _draft_preprocess_idle call for when all requests in the verify batch finish in the same iteration (#26859) 3 people 2026-06-05 00:47:32 -04:00
• aed0808e18 6-5 nightly failed test case fix (#27335) liuxianglong17 2026-06-05 11:39:23 +08:00
• c6c1f1a29a docs: sync legacy docs/-only updates into docs_new (Mintlify) (#27308) zijiexiaandClaude Opus 4.8 2026-06-04 19:45:13 -07:00
• 46c58b5c70 bench: fix MMMU VLM eval max_tokens for CoT prompt (#27327) Yuhao Yang 2026-06-05 10:28:16 +08:00
• 2c8357f794 [XPU] Enable Gemma 4 E2B / E4B / 31B/ 26B-A4B on Intel XPU (#23280) 5 people 2026-06-04 19:05:07 -07:00
• bcf89928b4 [router] Configure experimental sgl-router via CLI flags instead of a config file (#27073) Kangyan-ZhouandClaude Opus 4.8 2026-06-05 10:02:10 +08:00
• 631db6c757 [UnifiedTree]: Sync sidecar component hits across TP ranks and make SWA prefetch all-or-nothing (#27264) Zhangheng 2026-06-05 09:23:54 +08:00
• 5af02c18ae [spec_v2] Enable trtllm_mha draft-extend CUDA graph with v2 semantics (#25002) 3 people 2026-06-04 17:50:12 -07:00
• 7dc7376697 fix(attn): delegate init_mha_chunk_metadata in HybridLinearAttnBackend (#27316) Cheng WanandClaude Opus 4.8 2026-06-04 17:44:23 -07:00
• 7425bebb6c docs(cookbook): restore Gemma 4 transformers commit pin (#27321) Xinyuan Tong 2026-06-05 01:43:36 +01:00
• bba8de8cb4 docs: sync LMSYS SGLang blog cards (#27322) sglang-botandsglang-bot 2026-06-04 17:42:09 -07:00
• 088f70d0c0 ci: open the LMSYS blog-sync PR with the repo sglang-bot (#27318) zijiexiaandClaude Opus 4.8 2026-06-04 17:35:33 -07:00
• 0aa72a9e76 Replace skip_attn_backend_init with a batch-carried attention plan marker (+ staleness re-plan) (#27193) Cheng WanandClaude Opus 4.8 2026-06-04 17:13:18 -07:00
• 47377525cb ci: fix LMSYS blog sync to open a PR via gh and only run on main (#27179) zijiexiaandClaude Opus 4.8 2026-06-04 16:47:46 -07:00
• 448d3afb76 fix(spec): complete CustomSpecAlgo duck-typing interface and guard against drift (#27300) Lianmin Zheng 2026-06-04 15:52:46 -07:00
• e76d36214b Changes for SM120 perf and usability for NVFP4 (#26496) 4 people 2026-06-04 15:29:25 -07:00
• 07f326c184 Fix multimodal synthetic benchmark prompt generation to exclude special tokens (#26864) Bowen WangandXinyuan Tong 2026-06-04 23:27:43 +01:00
• 0e4aa081ba Add --enable-symm-mem for Qwen3.5 (#27296) Faradawn Yangandzijiexia 2026-06-04 17:23:37 -05:00
• 4cfebbb95f [FlashInfer v0.6.12] Support FlashInfer 4over6 NVFP4 (#25239) Ziang Li 2026-06-04 14:35:07 -07:00
• 5bf90ad988 Enable DeepGEMM PDL on by default (#23979) 3 people 2026-06-04 14:13:45 -07:00
• cd98d97037 Use level-1 (quiet) busy memory check in chunked-prefill and streaming tests (#27303) Liangsheng Yin 2026-06-04 17:04:27 -04:00
• 687cfe9198 Enable runtime busy memory check for speculation topk>1 (#27228) Liangsheng Yin 2026-06-04 16:46:20 -04:00
• 88a9d513e0 [Quant] Support asymmetric weight quant in compressed-tensors WNA16 (#25292) Ilia Iliev 2026-06-04 23:15:47 +03:00
• 69623f4b11 [AMD] Guard aiter greedy_sample OOB token id (fixes VLM MMMU CI) (#27247) YC Yen-Ching Tseng 2026-06-05 03:53:58 +08:00
• 75be922451 docs(cookbook): add Docker install option for Gemma 4 (#27287) Douglas YangandClaude Opus 4.8 2026-06-04 12:45:32 -07:00
• efe24704b7 [HiCache] feat: truncate mamba prefetch length to available host KV size (#26945) shuwenn 2026-06-05 03:01:37 +08:00
• 8e836e7dc9 Support optional kwargs in AITER fused_moe runner (#26746) Chengze FanandHaiShaw 2026-06-04 08:25:03 -07:00
• 133254086b Plug mamba_extra_buffer ping-pong slot leaks (#26941) Sam Shleifer 2026-06-04 09:46:36 -04:00
• b97a3dbb46 [HiSparse PD & PP]Fix HiSparse compatibility with PP decode (#27258) huangtingwei 2026-06-04 21:37:42 +08:00
• 8933ec8772 fix test cases failed on 5/30 in nightly pipeline (#26775) liuxianglong17 2026-06-04 20:07:20 +08:00
• 4167f211b1 Solving the problem of test case failures caused by timeouts (#27027) liuxianglong17 2026-06-04 20:07:04 +08:00
• 364c2bfea8 Test case restoration in the full test. (#26933) liuxianglong17 2026-06-04 20:06:44 +08:00
• e4a01d5ee2 [diffusion] fix: fix realtime webui recording timeline (#27249) Mick 2026-06-04 19:58:56 +08:00
• a5c7e9d236 [HiCache] fix PD L3 cache hit details from decode responses (#27046) huangtingwei 2026-06-04 18:01:21 +08:00
• 9e2ad6054c [diffusion] feat: speed up lossless realtime rgb transport (#27236) Mick 2026-06-04 16:59:23 +08:00
• ef170c27b6 Add quiet mode for busy mem check (level 1: buffer + dump on leak) (#27238) Liangsheng Yin 2026-06-04 04:31:57 -04:00
• 0783813fdf ci: cache HF hub for base-a-test-cpu to avoid Hub 429 flakes (#27092) Alison Shao 2026-06-04 01:25:00 -07:00
• 7aee2ff31b [AMD] Remove BF16-to-FP32 elementwise cast from compressor GEMM on HIP (#26914) jacky.cheng 2026-06-04 15:58:02 +08:00
• 10b6b45cad docs: add DeepSeek V4 FP4 indexer usage (#27035) Jinyan ChenandJinyan Chen 2026-06-04 15:44:21 +08:00
• ff93a576e5 [AMD] Minimax M25 : FP8 block-scale GEMM dispatch for ROCm 7.0 on gfx950 (#27111) YC Yen-Ching Tseng 2026-06-04 15:41:42 +08:00
• b89686710d [Docs] re-organize nemotron cookbook (#27240) zijiexia 2026-06-04 00:40:13 -07:00
• 1463e5fbdd docs: add Nemotron 3 Ultra cookbook entry (#26969) 4 people 2026-06-04 00:14:19 -07:00
• 1af53f67b2 [mem_cache][2/N] refactor: move SWATokenToKVPoolAllocator to allocator/swa.py (#26676) shuwenn 2026-06-04 15:10:37 +08:00
• a10bd785be Reduce mamba prefill allocation overhead (#25000) Leon Gao 2026-06-04 00:10:16 -07:00
• 04c16fc1e5 [AMD][CI] Remove transformers pin from GLM-5.x nightly jobs (#27232) 3 people 2026-06-04 14:34:07 +08:00
• f3acb6d4de [CI] Fix multimodal-gen path filter for shared trace code (#27222) Shangming Cai 2026-06-04 14:20:27 +08:00
• e4191708c9 [Qwen3.5][AMD] Fix shared-expert ×ep_size over-count under allreduce-EP (#26845) Alex SunandHaiShaw 2026-06-04 14:04:39 +08:00
• cc67f922cd [misc] Update Codeowner for Lora (#27224) Baizhou Zhang 2026-06-03 22:50:15 -07:00
• 858e5a5109 [diffusion] chore: disagg server args, launch helpers, and warmup utils (#26119) Heyang Huang 2026-06-04 13:40:39 +08:00
• 0e0ecc11ff Add nightly Intel XPU Docker release workflow (#27182) Jitendra PatilandMa Mingfei 2026-06-04 00:39:57 -05:00
• 6dcd78a37f [AMD] Add MiniMax-M2.5 TP=4 nightly accuracy test for MI355X (#27126) YC Yen-Ching Tseng 2026-06-04 13:17:40 +08:00
• ff5c4d7b57 Add zx3xyy to CI_PERMISSIONS.json (#27213) Lianmin Zheng 2026-06-03 22:12:00 -07:00
• e541bc3881 [EPD] feat: encoder DP mode with per-rank subprocess workers (#26576) Zhonghua Deng 2026-06-04 12:37:41 +08:00
• 71c759ebb7 [loader] Reduce transient allocations in NVFP4 MoE setup (#26861) Yinghai Lu 2026-06-03 21:13:25 -07:00
• 5c8a04ac4e [XPU CI] Expand stage-a and consolidate stage-b tests into stage-a (#27156) 3 people 2026-06-04 09:30:25 +05:30
• 10ab7c919f [refactor] Retire DecodeInputBuffers / PrefillInputBuffers in favor of CudaGraphBufferRegistry (#27192) Cheng WanandClaude Opus 4.8 2026-06-03 20:52:56 -07:00
• 3b7a258f63 [CPU] upgrade dependent torch ver to PT2.12 (#21456) Zaili WangandMa Mingfei 2026-06-04 11:04:11 +08:00
• 29d23e198f [diffusion] fix: preserve _explicit_fields across dataclasses.replace in DiffGenerator (#25308) 王鹤男 2026-06-04 11:02:23 +08:00
• 11605767e0 [diffusion] optimize: skip unused wanvae halo send copies (#27151) Mick 2026-06-04 10:23:01 +08:00
• 736263f3dc [UnifiedTree]: Support l3 storage for swa and deepseek v4 (#26881) Zhanghengand晟海 2026-06-04 10:17:34 +08:00
• 1a57145975 [codex] Fix adaptive metrics test flake (#27135) shuwenn 2026-06-04 10:07:17 +08:00
• e03dfa8182 [3/N][Sync sglang-miles] TITO Support (#23751) Yuzhen ZhouandJiajun Li 2026-06-03 18:45:33 -07:00
• 084c6a7e2a Refactor simulated acceptance length generation (#26768) Jonny Kong 2026-06-03 18:31:32 -07:00
• f6cd1a9822 Add num_waiting_uncached_tokens load metric (#27174) 3 people 2026-06-03 18:29:49 -07:00
• 687baf9471 fix(load-snapshot): avoid duplicate zmq bind in multi-tokenizer mode (#27145) ybyangandLianmin Zheng 2026-06-04 09:24:56 +08:00
• 14ed9b448e Add ZMQ IPv6 support, bench_serving sampling params, and reduce routed_dp_rank log noise (#27180) 3 people 2026-06-03 17:49:34 -07:00
• 5dbc52c2b7 [diffusion] doc: add ernie Image diffusion (#27195) Mick 2026-06-04 08:45:56 +08:00
• 7bb5c96685 Trigger scheduler diagnostics on health failure (#26757) Yinghai Lu 2026-06-03 17:19:55 -07:00
• d097cd2212 [GLM-5] Apply trtllm MHA kernel for GLM-5 on Blackwell (#21332) Baizhou Zhang 2026-06-03 17:16:54 -07:00
• 3790173b3b [diffusion] fix: avoid flashattention forward context lookup (#27153) Mick 2026-06-04 08:11:50 +08:00
• cfb7fb4fad [AMD] Fix TP2 DeepSeek-R1 nhead=64 MLA decode crash and add nightly coverage (#27188) Clintandclintg6 2026-06-03 18:56:05 -05:00
• c9ca56da8c Unify full→SWA index translation in init_forward_metadata; drop pool caches (#27091) Cheng WanandClaude Opus 4.8 2026-06-03 16:12:27 -07:00
• 8980eb82de [Docs] Update Nemotron3-Nano-Omni cookbook to reflect new model paths (#25198) Yihao Wang 2026-06-03 14:55:30 -07:00
• 61aa3293d3 Revert "Fix TokenizerManager crash on top_logprobs with tensor values" (#27187) Cheng Wan 2026-06-03 14:53:28 -07:00
• 90985117a5 docs: fix Nemotron Super MTP deployment command (spec-v2 + B200) (#27184) Jiajun Liandzijiexia 2026-06-03 14:52:47 -07:00
• 978fb6ed1a hicache kv events: publish split write-through fragments (#27072) ishandhanani 2026-06-03 23:52:25 +02:00
• 1dd9432889 [AMD] Fuse compress norm+rope+hadamard into single Triton kernel (#26894) jacky.cheng 2026-06-04 05:20:40 +08:00
• c670609ac5 [NPU] Diffusion CI Ground Truth Generation (NPU) (#24630) 5 people 2026-06-04 00:14:37 +03:00
• e485ad6ac1 Fix hybrid linear attention dispatch by layer id with draft-worker awareness (#27120) Cheng WanandClaude Opus 4.8 2026-06-03 14:11:14 -07:00
• 7716fa00e0 Fix TokenizerManager crash on top_logprobs with tensor values (#26825) Kevin Flansburg 2026-06-03 14:55:02 -06:00
• d1bc06b63b [AMD] Disable AITER custom all-gather in DeepSeek-R1-MXFP4 8-GPU test (#27163) YC Yen-Ching Tseng 2026-06-04 04:38:34 +08:00
• 293816ab14 [AMD][MXFP4] Online MXFP4 quantization 1/N - dense and MOE models w. original BF16 weight (#18005) 3 people 2026-06-03 21:55:24 +02:00
• e0b692600f [Mamba] extra buffer lazy support (#27118) Hanming LuandYAMY 2026-06-03 12:42:11 -07:00
• ac99794e64 Reland spec v2 tree drafting (eagle topk>1) with page_size==1 (#26866) (#26997) Liangsheng YinandAlison Shao 2026-06-03 15:40:05 -04:00
• 7f706f4cfb [Deps] Bump FI to 0.6.12 and cutedsl to 4.5.2 (#26854) Mohammad Miadh Angkad 2026-06-04 03:09:18 +08:00
• 578f232e5e Fix trace_modules gate disabling default trace contexts (#27173) Liangsheng Yin 2026-06-03 14:03:42 -04:00