Commit Graph

Select branches

Hide Pull Requests

bot/update-est-time-20260928

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• b7f856df70 DeepSeek V4 w4a4 MegaMoE (#25052) Baizhou Zhangandpranjalssh 2026-05-13 18:35:32 -07:00
• 34c0029f0a [diffusion] [AMD] feat: support online MXFP4 and fp8 quantization (#21431) 3 people 2026-05-13 17:52:01 -07:00
• af7511e0e8 [sgl-model-gateway] Close PyO3 binding gaps and add regression tests (#24719) Kangyan-ZhouandClaude Opus 4.7 2026-05-13 17:43:01 -07:00
• 7618ad7075 [attn backend] Integrate tokenspeed_mla prefill/decode kernels (fp8 kv cache, blackwell) (#24925) Qiaolin Yu 2026-05-13 17:36:17 -07:00
• 22d3f3996c ci: decouple stage and runner for cuda registry (#25197) Liangsheng Yin 2026-05-13 17:28:21 -07:00
• dd98153d17 chore(ci_monitor): drop post_bisect_to_slack (#25192) Alison Shao 2026-05-13 17:05:38 -07:00
• 6c0633b0b1 fix(nvfp4): make process_weights_after_loading hot-reload-safe via alias-when-same-shape (#25190) Cheng WanandClaude Opus 4.7 2026-05-13 16:57:39 -07:00
• 37f18438c5 [rebase]Deepseek_v4 support w4(mxfp4)a16 on hopper (#24986) shiyu7 2026-05-14 07:33:46 +08:00
• 371cb2ade2 [Bench] Add MEM profile activity to bench_serving (#25026) Khoa PhamandClaude Opus 4.7 2026-05-13 16:22:11 -07:00
• 9a32a0272f ci: compute matrix partition counts from est_time (#25193) Liangsheng Yin 2026-05-13 16:18:54 -07:00
• 8438709e9c Fix scheduler admission for near-full KV requests (#25126) ziang663 2026-05-14 06:30:49 +08:00
• f9ff5fc154 multi_layer_eagle: add tracing hooks (#24858) Liangsheng Yin 2026-05-13 15:29:38 -07:00
• 0a2615df24 chore: add vLLM SPDX copyright headers to ported files (#25182) sglang-botandClaude Opus 4.7 2026-05-13 15:17:30 -07:00
• 01a225ac6f [LoRA] MLA attention LoRA: q_b_proj / kv_b_proj support (#25001) Yanbin Jiangandgh1595 2026-05-13 15:15:30 -07:00
• 28758d37dd Add FlashInfer SM90 cutlass MXFP4 MoE backend (W4A16) for GPT-OSS + DeepSeek-V4 (#24816) Yuan Luoandluoyuan.luo 2026-05-14 05:53:18 +08:00
• d6b28b4a69 [Refactor] Remove dead key_convert_fn / convert_to_bigram_key (#25161) shuwennandClaude Opus 4.7 2026-05-14 04:54:36 +08:00
• 4d91a4f3a1 ci: merge sgl-kernel-build-wheels x86+arm into reusable workflow (#25135) Liangsheng Yin 2026-05-13 13:51:59 -07:00
• b1db9f71ee [SMG] Fix matrix-sibling concurrency collision in PR Test (SMG) (#25188) Kangyan-ZhouandClaude Opus 4.7 2026-05-13 13:25:40 -07:00
• 72b49bfac6 docker, ci: swap GB DeepEP source from fzyzcjy fork to deepseek-ai/DeepEP@hybrid-ep (#25113) Cheng WanandClaude Opus 4.7 2026-05-13 13:25:15 -07:00
• 3178a70577 [PD Disaggregation] Fix priority scheduling in PD disaggregation mode (#25062) Ata Fatahi 2026-05-13 11:43:21 -07:00
• 22012ba1bc Add BF16 support to EP-MoE for DeepGEMM (#17392) oriandzhiguo.qin 2026-05-14 02:06:37 +08:00
• 6ac30192fa [MLX] Add on-the-fly --quantization mlx_q4 / mlx_q8 for Apple Silicon (#24907) Le Zhangandlezhang 2026-05-13 11:06:13 -07:00
• ff70aeac30 [diffusion] feat: add performance mode server args (#24491) Mick 2026-05-14 00:57:46 +08:00
• e2290b155a Port KV Compression V2 from deepseek_v4_dev (#24890) 3 people 2026-05-13 22:40:38 +08:00
• d0913fca8d Port fused SiLU+clamp+FP8 quant from DSV4 dev branch (#24897) 4 people 2026-05-13 22:36:44 +08:00
• 9e00b7ca95 [NPU] add zbal support for npu (#24575) litangss 2026-05-13 20:52:42 +08:00
• 4984552cc9 Fix tests for decode radix cache (#25145) Shangming CaiandHanHan009527 2026-05-13 20:31:44 +08:00
• 2a4d382b07 [Disagg][NIXL] Add staging buffer support for heterogeneous TP KV transfer (#22536) 4 people 2026-05-13 04:54:02 -07:00
• 5227b07669 [CI] Add DSV4 Flash disaggregation test (#24973) Shangming Cai 2026-05-13 18:42:42 +08:00
• cf92ccbf18 [AMD] Run jit kernel PR test through run_suite.py register mechanism (#24987) YC Yen-Ching Tseng 2026-05-13 17:57:08 +08:00
• fc20f5b114 [AMD] Skip redundant CatArrayBatchedCopy in GLM-5 NSA TileLang decode (#24125) Jacob0226 2026-05-13 17:55:28 +08:00
• a9359707c1 [AMD] Enable preshuffle paged MQA and page_size=64 for NSA indexer (#23562) Thomas Wang 2026-05-13 17:33:57 +08:00
• 1ae3218d03 Add jasonjk-park and charlotte12l to CI_PERMISSIONS.json (#25104) Lianmin Zheng 2026-05-13 01:51:05 -07:00
• 9d0be860a4 [NPU] recover accuracy for gemma3-4b-it from 54% to 72% (reduced by transformer5.3) (#21537) McZyWu 2026-05-13 16:46:04 +08:00
• d8f7b78a29 [diffusion] fix: plumb max_sequence_length via diffusers_kwargs (#20930) 3 people 2026-05-13 16:32:47 +08:00
• 7d515c6d1f docs: prepend SGLANG_JIT_DEEPGEMM_PRECOMPILE=0 for H200 FP8 Flash max-throughput (#25152) zijiexiaandClaude Opus 4.7 2026-05-13 00:50:15 -07:00
• c32f2dc1ac fix(nixl): close file descriptors after each FILE transfer (#24671) Lukas HumbelandClaude Sonnet 4.6 2026-05-13 09:34:52 +02:00
• 839f7f2696 [AMD] Add _skip_rope_for_aiter_fused_mla method and check to avoid double rotating with gfx950 and Aiter backend (#24148) Matti Varjokallio 2026-05-13 10:27:09 +03:00
• 51a9403104 Update flashinfer to 0.6.11.post1 (#25129) Baizhou Zhang 2026-05-13 00:12:19 -07:00
• 10a005e4ea [CI][NPU] use internal HTTP cache for Rust toolchain via RUSTUP_CACHE (#24801) Goalina 2026-05-13 14:58:07 +08:00
• f2a90094c9 bench: fix wrong flag names in bench_one_batch{,_server} docstrings (#25143) Cheng WanandClaude Opus 4.7 2026-05-12 23:46:47 -07:00
• 66a9234246 [AMD] support fp8 blockwise quantization combine for mori ep (#24879) billishyahaoandDuyi-Wang 2026-05-13 14:24:40 +08:00
• 72b266d59b [AMD] Disable unittest fail-fast for deepseekv4 perf test (#25039) YC Yen-Ching Tseng 2026-05-13 13:47:31 +08:00
• 245f7d8026 [AMD] Register 5 server-style 1-GPU tests for AMD PR CI (#24572) Michael 2026-05-13 13:45:52 +08:00
• c665edec6e [env] Make max KV chunk capacity configurable via SGLANG_MAX_KV_CHUNK_CAPACITY (#25120) Khoa Pham 2026-05-12 22:37:45 -07:00
• b0018ad015 [Doc]: refactor Intern-S2-Preview cookbook with interactive command generator (#25134) zijiexiaandClaude Opus 4.7 2026-05-12 22:26:18 -07:00
• adae4042a7 spec: defer verify() idle hidden_size to worker fixup (#25109) Liangsheng Yin 2026-05-12 22:13:52 -07:00
• 938198e91c ci: extract check-changes into reusable workflow (#25132) Liangsheng Yin 2026-05-12 22:05:11 -07:00
• 3f048c80b8 Reject repetition_penalty=0 in SamplingParams.verify() (#24874) RulinJuiceandRulinJuice 2026-05-13 12:25:23 +08:00
• 622baa17bd [Doc]: add interns2preview in cookbook (#25115) RunningLeon 2026-05-13 12:05:59 +08:00
• 4e35c30cbe [Bugfix, NSA HiCache] Fix missing override_kv_cache_dim in attach_hybrid_nsa_pool_to_hiradix_cache (#25022) 3 people 2026-05-13 12:45:08 +09:00
• 5ed9a494d0 [UnifiedTree] fix: allow partial match on evicted+backuped nodes (#24943) shuwenn 2026-05-13 11:43:12 +08:00
• 6140e45ef3 [UnifiedTree]: Fix the leaf determination logic in _cascade_evict. (#25068) ZhanghengandJINZ 2026-05-13 10:57:32 +08:00
• 409d350fb6 Bugfix: fix symm not enabled due to incorrect registration of comm (#19329) sky 2026-05-13 10:01:04 +08:00
• 642ac9c916 [NPU]pp support mla kv transfer (#23893) chenxu214 2026-05-13 09:10:02 +08:00
• d6d3d0f599 Optimize SWA memory preallocation for disaggregated decode (#24857) 3 people 2026-05-13 09:09:34 +08:00
• 4fb40bffac perf(nvfp4): free unused source scales after weight processing (#25107) Cheng WanandClaude Opus 4.7 2026-05-12 16:53:50 -07:00
• b49d05fd0e feat: add Crusoe managed inference backend (#20475) Emmanuel AcheampongandClaude Sonnet 4.6 2026-05-12 16:23:23 -07:00
• 49ac447c94 [bench] Agentic support for bench_serving.py (#25016) Khoa Pham 2026-05-12 16:00:45 -07:00
• a4109e87ac Fix TRTLLM MHA routing for draft extend (#24856) YAMY 2026-05-12 15:48:32 -07:00
• d5f3254ed1 [Dependency] Flashinfer 0.6.8post1 -> 0.6.11 (#24452) Brayden Zhongandb8zhong 2026-05-12 17:38:32 -04:00
• 48fc26a814 Fix Eagle draft decode positions (#25015) Lianmin Zheng 2026-05-12 14:04:58 -07:00
• 486b54714b [NVIDIA] Deterministic inference backend order on Blackwell (#21450) Kaixi Hou 2026-05-12 13:35:50 -07:00
• 52d4c697bb Fix fused_moe import for non-NPU devices (#25076) Polisetty V R K Jyothendra VarmaandBrayden Zhong 2026-05-13 01:35:51 +05:30
• 00d8730d05 [Bug Fix] Fix broken sgemm_lora_a_graph_fwd due to invalid torch.mm() call (#24760) flutist-alibaba 2026-05-13 03:29:01 +08:00
• 1419bcc39e [CI] Temporarily disable multimodal-gen test_update_weights_from_disk (flaky) (#25032) Alison Shao 2026-05-12 12:28:33 -07:00
• 538832c8b7 spec: STANDALONE skips hidden_states end-to-end (Optional schema + None-safe consumers) (#25037) Liangsheng YinandQiaolin Yu 2026-05-12 12:27:21 -07:00
• e86fb42736 feat: emit per-iteration forward pass metrics via ZMQ PUB (#22789) 4 people 2026-05-12 10:28:17 -07:00
• fd3eb77d45 [Cookbook]: add Laguna-XS.2 (Poolside) (#24730) Jimmy Shongandgithub-actions[bot] 2026-05-12 10:06:26 -05:00
• 6be1a45d99 Fix swa component host hit (#25085) Ke Bao 2026-05-12 21:03:39 +08:00
• 693f49793a [NPU] use causal_conv1d_update_v2 for performance (#24595) iridiumineandiridiumine 2026-05-12 17:04:02 +08:00
• 1efe9e2164 [Bug Fix] Reject incompatible combination of --disable-cuda-graph-padding and --enable-torch-compile (#23903) Praneth Paruchuri 2026-05-12 13:48:36 +05:30
• 8d27ce7371 Optimize uvicorn startup command (#25041) lw9527 2026-05-12 15:28:13 +08:00
• b35fd5f104 [fix] skip legacy minicpmv conv template for MiniCPM-V 4.6 (#24998) Yihao Wang 2026-05-12 00:27:28 -07:00
• 7582237f8e [Tiny Fix] Disable BCG when inner layer_model unresolved (#25021) Yuwei AnandClaude Opus 4.7 2026-05-11 23:51:27 -07:00
• ca3bc05fea Deepseek-v4-Pro share expert tp1 (#24949) zhangxiaolei 2026-05-12 14:19:35 +08:00
• a72d3aef17 [Spec] Multi-layer mamba scatter cleanup; fix positional call bug (#25030) Liangsheng Yin 2026-05-11 22:42:30 -07:00
• 71285335f7 Revert "Migrate Intel CPU cases to the test/registered." (#25044) Ma Mingfei 2026-05-12 13:32:47 +08:00
• 1f985c50db [Spec] Rename accepted_indices -> accept_indices; drop _token_id suffix per Rule 5 (#25038) Liangsheng Yin 2026-05-11 22:29:02 -07:00
• ecf5d844f5 Migrate Intel CPU cases to the test/registered. (#22670) jundu 2026-05-12 13:27:51 +08:00
• d7f4761a48 [PD] Refactor hybrid state transfer (#24932) Ke Bao 2026-05-12 13:16:54 +08:00
• 91907b7b93 [UnifiedTree]: Fix Unified HiCache tombstone lock release replay (#24972) Zhangheng 2026-05-12 13:16:06 +08:00
• 4ad63ad02f [Spec] Rename accepted_drafts -> correct_drafts for unambiguous naming (#24081) Liangsheng Yin 2026-05-11 22:12:40 -07:00
• 6bfb365c07 [PD] Rate limit prefill inflight polling warnings (#24967) tangcy98andzhangzhang 2026-05-12 12:50:57 +08:00
• 6bb79c155b [Linear Attn] Add CUSTOM enum and plugin extensibility for kernel backends (#24937) Ming Yang 2026-05-11 21:46:53 -07:00
• cfc41d5b15 Fix kimi k2.5 mla eagle + dp attention (#25033) Qiaolin Yu 2026-05-11 20:58:00 -07:00
• 0f3932c055 [Fix] Qwen3-ASR config: set thinker_config before super().__init__ (#24187) Yihao Wang 2026-05-11 20:52:15 -07:00
• f526e3fa27 [Spec] Mamba scatter cleanup; fix multi-layer positional bug; dflash naming (#25029) Liangsheng Yin 2026-05-11 20:36:50 -07:00
• 10375a1037 [NIXL][XPU] Fix uint64 overflow for mismatched P/D TP sizes (e.g. prefill_tp=1, decode_tp=2) (#24648) Jianhong Zhang 2026-05-11 20:02:24 -07:00
• 0a37d24e62 [diffusion] hardware: support sage attention backend on MUSA (attn backend, 21/N) (#24752) R0CKSTAR 2026-05-12 10:50:52 +08:00
• 5495026a3b [HiCache] feat: default storage prefetch timeout (#23309) shuwenn 2026-05-12 09:49:35 +08:00
• 186eb42459 Feat: Support SWA (Sliding Window Attention) for EAGLE-3 drafter (#24664) Doğaç Eldenk 2026-05-11 20:40:10 -05:00
• a75b79e03b Feat: Support newer EAGLE-3 drafters (#24663) Doğaç Eldenkandhnyls2002 2026-05-11 20:17:20 -05:00
• f3a8189e20 [Spec] Internal rename per N2 v2 naming rule (#25014) Liangsheng Yin 2026-05-11 18:16:26 -07:00
• bfc2eda42d [MUSA] Use MUSA-optimized operators in piecewise CUDA graph (#23633) popsiclexu 2026-05-12 08:55:27 +08:00
• 74d70af09a [Apple Silicon] Add Metal kernel support in sgl-kernel (#23449) R0CKSTAR 2026-05-12 08:54:27 +08:00
• de098f4f4d rerun-test: show running state in reply comment (#25018) Liangsheng Yin 2026-05-11 17:04:29 -07:00
• 34555aee9c Add minosfuture to CI_PERMISSIONS.json (#25019) Lianmin Zheng 2026-05-11 16:54:32 -07:00
• e9dea79755 (3/n - prefill optimize)[LoRA][MoE] Optimize virtual experts: remove CPU-GPU sync & multi-block CUDA JIT histogram (#24262) 3 people 2026-05-12 01:36:57 +02:00
• 6c3541a914 spec: route idle hidden_size via EagleDraft{,Extend}Input classmethods (#25013) Liangsheng Yin 2026-05-11 15:59:51 -07:00