Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• 8e890391f5 [Spec] Support FlashInfer CUDA graph for EAGLE draft-extend (#28782) Liangsheng Yin 2026-06-21 14:46:25 -07:00
• a4d0ff3def [misc] Make NaN-logit sanitization opt-in (default off) (#28829) Lianmin Zhengandhnyls2002 2026-06-21 14:31:35 -07:00
• 7f67965b4d [BugFix] NCCL deadlock in HiCache writing_check by making all_reduce unconditional (#26923) 4 people 2026-06-22 01:47:44 +08:00
• 5351800700 [Bugfix] Fix MTP acceptance regression on plan stream by moving int64 cast before plan stream context (#28410) iridiumine 2026-06-22 01:26:58 +08:00
• 643ee748c6 [PP] Pass DSA topk through PP warmup proxy buffers (#28785) Mohammad Miadh Angkad 2026-06-21 23:55:39 +08:00
• 7942d546d1 Revert "[Spec] Split init_backends; account draft weights in --mem-fraction-static" (#28841) Lianmin Zheng 2026-06-21 07:52:35 -07:00
• 320b231ea6 [diffusion] chore: bound DiffGenerator local cleanup (#28833) Mick 2026-06-21 22:23:49 +08:00
• a7f31a6e1b [diffusion] fix: fix SANA-WM CFG-parallel tensor devices (#28835) Mick 2026-06-21 22:01:00 +08:00
• a51d56d948 CI: Pin flash-attn-4 for diffusion CI consistency (#28838) Mick 2026-06-21 21:13:41 +08:00
• 3975ea5ac7 Fix H20 torch import reinstall fallback (#28818) Lianmin Zheng 2026-06-21 04:50:46 -07:00
• 9691a29fe0 [Spec] Split init_backends; account draft weights in --mem-fraction-static (#28683) cctry 2026-06-21 01:22:26 -07:00
• 6d4ca9bc54 Cap SWA pool sizing with chunk cache (#28755) cctry 2026-06-21 01:06:59 -07:00
• c9488241e9 [Refactor] Auto-derive CLI args from dataclass fields to eliminate duplication (#28814) Lianmin Zheng 2026-06-21 00:51:08 -07:00
• b4dda8b3ce fix(anthropic): handle mid-conversation system messages (#26773) Jairo David Campaña RoseroandXinyuan Tong 2026-06-20 23:34:47 -05:00
• c65f4ea692 [diffusion] fix: validate openai sampling dimensions (#28791) Mick 2026-06-21 09:44:38 +08:00
• 6a16573a7f [diffusion] fix: fix Qwen-Image-Layered string image paths (#28790) Mick 2026-06-21 09:43:56 +08:00
• d331fdd2ba Add project rule: prefer msgspec.Struct over dataclasses (#28816) Lianmin Zheng 2026-06-20 18:43:16 -07:00
• 54b9b9d0c9 Remove threading atexit monkey patch (#28812) Lianmin Zheng 2026-06-20 18:30:50 -07:00
• 2552b860a3 [AMD][bugfix] Place TBO cuda-graph num_token_non_padded buffer on model devices (#28337) karverma-amd 2026-06-20 20:06:22 -05:00
• 5b3eeaf504 [Fix] MM pool GPU alloc with base_gpu_id (#23377) pure water 2026-06-21 08:51:06 +08:00
• fbbf559de2 fix bench_one_batch by extending array with array not list (#28732) Polisetty V R K Jyothendra Varma 2026-06-21 05:58:41 +05:30
• 8a3d6c3403 Sort pyproject dependency lists (#28811) Lianmin Zheng 2026-06-20 17:24:57 -07:00
• f42ec350b4 [mtp] add rejection sampling for speculative decoding (#26312) 6 people 2026-06-21 06:10:42 +08:00
• 95fb1ef697 [CI] Remove deprecated test/srt legacy CI setup (#28810) Lianmin Zheng 2026-06-20 15:09:33 -07:00
• fe428dd845 Clean up startup log noise (#28807) Lianmin Zheng 2026-06-20 15:02:52 -07:00
• d6d06cdc17 [AMD] Fix no-op dtype cast in _topk_ids_logical_to_physical_dynamic on HIP (#28074) Rita BrugarolasandClaude Opus 4.6 2026-06-20 11:07:22 -07:00
• ff1fc1fbdf [mem_cache][5/N] refactor: extract host KV cache base layer into pool_host package (#27273) shuwenn 2026-06-20 20:44:08 +08:00
• 1109acc24b [diffusion] optimize: shard hunyuan text tokens under sp (#28319) Mick 2026-06-20 18:21:38 +08:00
• a38eba0f1a [diffusion] CI: remove flaky layerwise offload diffusion case (#28778) Mick 2026-06-20 17:27:27 +08:00
• 47cad39f34 [AMD] Optimize o_proj gemm and attn output rope performance (#28722) kkandwunhuang 2026-06-20 17:11:01 +08:00
• 1115373668 [AMD] Fix garbled unquantized Qwen3-30B-A3B output on ROCm/aiter where the aiter CK fused-MoE falls back to Triton with pre-shuffled weights (#28244) Zhiyao JiangandXinyu Jiang 2026-06-20 04:25:01 -04:00
• c1416bb3ee [Diffusion] Keep FastHunyuan VAE resident on high-memory GPUs (#28773) Xiaoyu Zhang 2026-06-20 15:54:06 +08:00
• 45d203fb08 Fix tokenizer state cleanup on dispatch failure (#28694) Lianmin ZhengandYinghai Lu 2026-06-19 21:55:39 -07:00
• 28e2096d1c [sgl] wire SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW on Unified Cache. (#28161) Bi Xue 2026-06-19 21:06:52 -07:00
• 7516f0db9f [cookbook] Laguna-M.1: add PD disaggregation section (#28737) Jimmy Shong 2026-06-19 19:49:49 -07:00
• 653f6735a0 [AMD] add nightly kv_canary JIT benchmark suite (#28611) Michael 2026-06-19 19:46:34 -07:00
• 2cbe1e6404 [Apple Silicon] [MLX] Fix MlxModelRunnerStub.initialize() signature desync with base (#28660) Jae B.andR0CKSTAR 2026-06-19 21:05:22 -04:00
• 871ed0dc0c Revert "ci: add 4-GPU mi35x runner and rebalance off the saturated 8-GPU pool" (#28751) Michael 2026-06-19 16:23:18 -07:00
• 6b945c16f4 [LoRA] Fix experimental fast-path multi-adapter correctness + flashinfer 0.6.12 compatibility (#28091) Yanbin Jiang 2026-06-19 16:20:19 -07:00
• 364bf976be [router] Tokenize prompt once at ingress; forward input_ids to the engine (all policies) (#28744) Kangyan-ZhouandClaude Opus 4.8 2026-06-19 16:07:17 -07:00
• c3bae61e16 [Bug] fix(DummyModelLoader): run post_load_weights before process_weights_after_loading (#28665) Shu Wang 2026-06-19 15:57:45 -07:00
• abf7011cdd [router] Raise chat body cap to 5 MiB for long contexts (#28742) Kangyan-ZhouandClaude Opus 4.8 2026-06-19 15:50:34 -07:00
• 420004827c [AMD] register 3 tests to stage-b-test-1-gpu-large-amd (batch-6) (#28736) Michael 2026-06-19 15:45:16 -07:00
• 13aab2fc06 ci: add 4-GPU mi35x runner and rebalance off the saturated 8-GPU pool (#28745) Michaelandmichaelzhang-ai 2026-06-19 15:41:19 -07:00
• 3a574846ff [CI] Publish 4-GPU nightly profiler traces (#28738) Mohammad Miadh Angkad 2026-06-20 06:27:24 +08:00
• b225f48fcc [NPU][FIX CI] Set the warm-up mode to "request" for NPU diffusion tests (#28724) Elizaveta MartirosianandElizaveta Martirosian 2026-06-20 01:14:26 +03:00
• 3ed46f599f [core] Don't force seq_lens_cpu publication under piecewise CUDA graph (#28633) 4 people 2026-06-19 15:12:07 -07:00
• d271de64fe [misc] Move bench_one_batch_server into sglang/benchmark/ with a back-compat shim (#28625) Liangsheng Yin 2026-06-19 14:19:29 -07:00
• c9c2445146 [CI] Bump actions/github-script to v8 (Node 24 runtime) (#28741) Liangsheng Yin 2026-06-19 14:18:37 -07:00
• 6fdcb9934c fix(runner): size eager static buffers for prefill budget and MLP-sync autotune (#28677) Cheng Wan 2026-06-19 13:23:26 -07:00
• 2aa7b58aa7 refactor(runner): reuse a prepared static buffer for every dummy run (#28740) Cheng Wan 2026-06-19 13:16:07 -07:00
• 856b0dc74b refactor(runner): move kernel warmup into the shared runner lifecycle (warmup()) (#28739) Cheng WanandClaude Opus 4.8 2026-06-19 13:13:13 -07:00
• d705a91de1 refactor(runner): add EagerRunner, own the eager path, polymorphic dispatch (#28386) Cheng WanandClaude Opus 4.8 2026-06-19 13:05:43 -07:00
• ab0714d0ee ci: run GB300 nightly suite in the standard Nvidia nightly workflow (#28536) 3 people 2026-06-20 03:24:38 +08:00
• 7c505c2927 [AMD] fix(jit): port kv_canary write/verify/plan kernels to ROCm (#28357) Michael 2026-06-19 12:19:37 -07:00
• c436a8161a [AMD] Enable HiSparse on ROCm (#26639) 3 people 2026-06-19 13:59:45 -05:00
• ca88b7f1d2 fix: remove manual rope parameters injection in PretrainedConfig (#23910) Jaybe 2026-06-19 12:41:51 -05:00
• 2ad9a5b576 [UnifiedTree] Use dense model for HiCache+CP KL tests (#28726) Vladislav Nosivskoy 2026-06-19 19:18:01 +03:00
• cab62855f5 [router] Align sgl_router_ttft_seconds buckets with engine TTFT grid (#28717) Kangyan-ZhouandClaude Opus 4.8 2026-06-19 08:53:46 -07:00
• 88c261c3f3 Fix IndexCache PP topk handoff (#28532) Mohammad Miadh Angkad 2026-06-19 23:39:09 +08:00
• 3af991fb3e [AMD] Make breakable CUDA graph run on ROCm/HIP (#28173) Oguz Ulgen 2026-06-19 07:16:00 -07:00
• 5eaae5bacd Add CODE_OF_CONDUCT.md (#28728) sglang-botandClaude Sonnet 4.6 2026-06-19 07:08:05 -07:00
• 9bb9d17e1a [Spec] Unify speculative grammar token-accept path in decode processing (#28682) Liangsheng Yin 2026-06-19 02:49:40 -07:00
• 941a11ada3 [HiCache] refactor: remove unused transfer buffer (#26003) shuwenn 2026-06-19 17:46:04 +08:00
• 0146692cc9 [CI] Fail fast on empty install_script in Rerun Test workflow (#28721) Liangsheng Yin 2026-06-19 02:07:52 -07:00
• a6db86d535 refactor(runner): split BaseRunner (shared) from BaseCudaGraphRunner (#28385) Cheng Wan 2026-06-19 02:04:15 -07:00
• 1c6331cbd6 refactor(runner): rename runner replay/load/can_run for the shared surface (#28384) Cheng Wan 2026-06-19 01:45:04 -07:00
• 31c0a98066 [codex] Update diffusion skills for latest main (#28711) Xiaoyu Zhang 2026-06-19 16:37:59 +08:00
• b6be5dd20f [codex] Remove outdated SGLang SOTA skill (#28719) Xiaoyu Zhang 2026-06-19 16:26:44 +08:00
• af2ec2a0dd [diffusion] perf: merge LTX-2 stage-1 distilled LoRA into the base in original mode (#28594) Mick 2026-06-19 15:41:47 +08:00
• 59eb142ec2 [NPU] [Diffusion] Performance Optimization for LTX-2 Model (#22445) 4 people 2026-06-19 13:58:45 +08:00
• 3b417d3999 Revert "[Diffusion] FLUX: fuse FeedForward GELU into up-proj GEMM (cublasLt epilogue)" (#28708) Xiaoyu Zhang 2026-06-19 13:44:57 +08:00
• 547a017a77 [diffusion] CI: run nightly image comparisons on 2 GPUs (#28593) Mick 2026-06-19 13:43:02 +08:00
• 7da92e1112 [diffusion] Sana: pack self-attn q/k/v and cross-attn k/v into single GEMMs (#28393) Xiaoyu ZhangandClaude Opus 4.8 2026-06-19 13:25:52 +08:00
• 7e20e25848 [docs] Add B300 cookbook deployment options (#28697) Xiaoyu Zhang 2026-06-19 12:52:45 +08:00
• 24d15dd92e [AMD][DSV4] fix nonetype issue when enabling hicache (#28541) billishyahao 2026-06-19 12:26:48 +08:00
• fac11f3bc1 [AMD] Document Mori XGMI for Single-Node PD Disaggregation (#25094) Clintandclintg6 2026-06-18 21:25:49 -05:00
• d962d18f15 docs: add --trust-remote-code to Laguna-M.1 / XS.2 cookbook configs (#28693) Jimmy Shong 2026-06-18 19:21:39 -07:00
• c7397de571 Use Marlin for SM120 MXFP4 MoE (#28231) Mohammad Miadh Angkad 2026-06-19 10:19:41 +08:00
• 4d94e9471a [AMD] Relax allreduce-fusion residual accuracy tolerance to 1 bf16 ULP (#28226) 4 people 2026-06-19 10:18:53 +08:00
• b36360dc5b [AMD][Perf] Split-KV flash-decode attention for EAGLE target-verify (Triton backend) (#27382) giang_ng_tr 2026-06-19 09:11:10 +07:00
• 1c8551169d Add opt-in CUDA-graph capture-trace export (#28551) Lucia Fang 2026-06-18 18:51:14 -07:00
• 6b7ecca663 [HiCache]Support hybrid pool staged H2D kernel (#28434) huangtingweiandhzh0425 2026-06-19 09:48:03 +08:00
• 05ee93c44f Remove redundant cast and copy in calling trtllm_fp8_block_scale_moe (#28555) Brayden ZhongandBrayden Zhong 2026-06-18 18:11:29 -07:00
• ef01618dfb support MPServer and embedded server for granian to enable muti tokenizer worker (#28573) Rain Jiang 2026-06-18 17:59:12 -07:00
• ea407df4b0 Use Flashinfer allreduce fusion for MNNVL allreduce for Nemotron (#28346) Brayden ZhongandBrayden Zhong 2026-06-18 17:30:12 -07:00
• 62ab09a478 [AMD] register 2 spec tests to stage-b-test-1-gpu-large-amd (batch-5) (#28558) Michael 2026-06-18 16:33:23 -07:00
• c1067f88d6 [AMD][Perf] Tune extend attention block sizes for gfx950 (head_dim > 128) (#27793) giang_ng_tr 2026-06-19 05:16:52 +07:00
• e3026ef016 [3/N][CP] Implement zigzag CP strategy (#28421) Baizhou Zhang 2026-06-18 15:10:30 -07:00
• 9fc9d37f6d Fix spec decoding with grammar in disagg (#24082) 6 people 2026-06-18 14:58:32 -07:00
• bea282cede [DeepSeek-V4] Fuse UE8M0 scale rounding into FP8 group quantization (#26766) 3 people 2026-06-19 05:41:56 +08:00
• 27a374eaef [AMD][CI] Use non-gated Qwen3-8B for MI35x disaggregation tests (#28678) Zhaoyi Li 2026-06-18 16:29:35 -05:00
• f83e4d5968 refactor(runner): unify eager-forward DP/MLP-sync padding into one helper (#28383) Cheng Wan 2026-06-18 13:41:30 -07:00
• 2411737244 [spec decoding] fully overlap spec decoding for hybrid linear attention backend (#28579) Qiaolin Yu 2026-06-18 13:31:12 -07:00
• cf0afe3223 [FA]Add lost params in fa varlen func (#28642) DovLin 2026-06-19 04:30:20 +08:00
• 66a7fd5c0b refactor: mamba radix cache server args initialize (#28151) Yi ZhangandKe Bao 2026-06-19 04:02:36 +08:00
• 8f6d9ef9a5 [misc] Drop redundant req_pool_indices_cpu guards; fold hisparse into GLM-5.1 e2e (#28607) Liangsheng Yin 2026-06-18 12:51:23 -07:00
• 792cb3a5d0 fix: add missing guard for use_jit_ep_activation (#27377) Lawrence Wu 2026-06-18 12:51:09 -07:00
• 61a8b42c00 docs(minimax-m3): add MMMU-Pro accuracy to B200 benchmark card (#28668) Xinyuan Tong 2026-06-18 19:40:56 +01:00
• f7632ef860 [Cookbook] Laguna-M.1: enable FP8 on Blackwell + drop provisional AIME numbers (#28664) Jimmy Shong 2026-06-18 09:23:53 -07:00