Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• 0db2bdfec5 Fix buffer-mode HiCache load-back ownership races; add optional prefetch anchor lock (#35769) Zhiqiang Xie 2026-08-21 20:20:36 -07:00
• fbafd1b123 Add sampling observer auxiliary output hooks (#35747) Alec SandAlec Solder 2026-08-21 22:28:18 -04:00
• 5662c03363 Support CPU offload for mxfp8 KV cache (#35888) Ke Bao 2026-08-22 10:19:06 +08:00
• d90318b3e2 [MLX] Upgrade to Torch 2.13/MLX 0.32+ and redesign the Torch-MLX tensor bridge (#32984) R0CKSTARandAlex Nails 2026-08-22 09:51:42 +08:00
• 3b5909de0e [DeepSeek V4] Add W4A4 MegaMoE server flag (#35918) Baizhou Zhang 2026-08-21 18:44:18 -07:00
• 0be2a209ac [diffusion] refactor: hand out pinned host memory per layer (#35867) MickandClaude Opus 5 2026-08-22 09:37:31 +08:00
• b26695a26e [diffusion] feat: reject unsupported quantized component checkpoints (#35873) Mick 2026-08-22 09:33:43 +08:00
• 22dafbcbd9 [diffusion] feat: keep a cpu-started vae weights on the checkpoint mapping (#35862) MickandClaude Opus 5 2026-08-22 09:31:30 +08:00
• 0ee3749e07 [Fix] Read the granite sinks dtype from the exec bag, not the legacy global shim (#35921) Khoa PhamandClaude Opus 5 2026-08-21 18:13:18 -07:00
• 7fd5454335 [DSA] Route the ragged prefill top-k to the v2 kernel (#35175) DarkSharpnessandClaude Opus 5 2026-08-22 07:59:41 +08:00
• 60ff1e33a5 [DeepSeek V4] Default FP4 checkpoints to FlashInfer MXFP4 MoE (#35919) Baizhou Zhang 2026-08-21 16:48:58 -07:00
• 7d7ab4b5c6 Rainj me/rust server refactor2 (#35239) Rain Jiang 2026-08-21 16:37:02 -07:00
• fe8f9d7457 [Docs] Add --prerelease=allow to cookbook uv install commands (#35920) zijiexiaandClaude Opus 5 2026-08-21 15:51:48 -07:00
• 2440820528 docs: add website link to README header (#35210) Alison Shao 2026-08-21 15:50:52 -07:00
• 834400705f perf: overlap Qwen shared expert with DeepEP routed experts (#34938) YAMY 2026-08-21 15:39:44 -07:00
• 7d893255c3 [Spec][LoRA] Support multi-adapter LoRA with EAGLE/NEXTN/DFLASH/DSPARK speculative decoding (#34337) Yanbin Jiang 2026-08-21 14:21:53 -07:00
• 590b11a5ef [Runtime] Don't override CUDA_MODULE_LOADING (#35711) Hank HanandTRAE CLI 2026-08-22 05:09:27 +08:00
• 4d42deff0a chore: bump docs install version to 0.5.18 (#35911) sglang-botandsglang-bot 2026-08-21 13:24:03 -07:00
• 05c584c44f docs: add DSPARK speculative decoding option to Ling-3.0-flash cookbook (#35861) Xinyuan Tong 2026-08-22 02:50:55 +08:00
• 729a050ea3 refactor(disagg): extract _all_reduce_polls helper (#35886) Shangming Cai 2026-08-22 02:30:33 +08:00
• c3735625de fix(grpc): derive choice count before normalization (#35778) Connor Carpenter 2026-08-21 14:29:52 -04:00
• 61c2da42bb [Fix] Pass Anthropic thinking history as reasoning_content for custom chat encoders (#35480) Mohammad Miadh AngkadandMohammad Angkad 2026-08-22 02:22:11 +08:00
• 70983bd7db Add SGLang Granite SWA support via existing Granite models (#35794) Davis Wertheimer 2026-08-21 14:13:15 -04:00
• 0bdd28d487 [mem_cache] docs: add a layer map and placement rules (#35643) Shuwen Wangandispobock 2026-08-22 00:35:07 +08:00
• a7ec6b97f7 Restructure mem_cache auto-labels by layer (#25122) Shuwen Wangandispobock 2026-08-21 23:29:59 +08:00
• 8658d00764 [diffusion] feat: support loading peft lora (#35868) Mick 2026-08-21 22:58:59 +08:00
• 0447ade326 [diffusion] fix: fall back to a component's default attention backend (#35796) li_maxandMick 2026-08-21 22:56:44 +08:00
• 932f632158 [diffusion] fix: do not warn that the recommended short edge is unverified (#35745) 王鹤男 2026-08-21 22:56:03 +08:00
• a41da991c8 refactor(disagg): collapse duplicated branches in get_kv_class (#35847) Shangming Cai 2026-08-21 22:11:31 +08:00
• 5ecd6d794d [diffusion] fix: fix quantized qkv scales and missing-param policy for minimax-h3 (#35740) 王鹤男 2026-08-21 22:10:50 +08:00
• 4f343abc13 refactor(disagg): remove unreferenced dead code (#35838) Shangming Cai 2026-08-21 22:09:25 +08:00
• 5a46d657b7 [diffusion] refactor: resolve lora weight sources deterministically (#35774) Mick 2026-08-21 21:05:40 +08:00
• 5206f11543 [diffusion] fix: stop the mapped-weight store from holding the parameter itself (#35813) Mick 2026-08-21 21:02:21 +08:00
• dad6fd0f04 refactor(disagg): remove dead get_embedding_port (#35844) Shangming CaiandClaude Opus 5 2026-08-21 18:12:23 +08:00
• 39d4d65a51 [diffusion] Accelerate SANA-Video linear attention in quality=high (#35728) Xiaoyu Zhang 2026-08-21 18:05:43 +08:00
• a5c52a9358 [diffusion] Enable LongCat breakable CUDA graphs (#35724) Xiaoyu Zhang 2026-08-21 17:59:16 +08:00
• e7a37c8550 refactor(disagg): remove dead build_and_send_encode_request (#35843) Shangming Cai 2026-08-21 17:58:35 +08:00
• 0db2c53dec Fix overlap prebuilt row reuse race (#35748) jasonjk-park 2026-08-21 18:00:13 +09:00
• 896acc8860 [Fix] Clear full-to-SWA mapping with index_fill_ to avoid a blocking H2D copy (#35773) Liangsheng Yin 2026-08-21 01:23:26 -07:00
• 4c98759c73 [AMD] fix(rocm): support flydsl 0.3.0 in the FlyDSL fused norm kernel (#34536) 3 people 2026-08-21 16:13:21 +08:00
• 8ff9c2b227 [mem_cache][9/N] refactor: move DSAIndexerPoolHost to pool_host.dsa (#35306) Shuwen Wang 2026-08-21 15:59:15 +08:00
• 8a123cbd0e [Refactor] New EPD (#30398) 4 people 2026-08-21 15:22:48 +08:00
• 6a12583679 [AMD] Update ROCm AITER pin to c16d44b (#35810) Bingxu Chen 2026-08-21 15:02:17 +08:00
• f3fe81583e add py env activate in xpu kernel release workflow (#35726) Zaili Wang 2026-08-21 14:27:00 +08:00
• 44c90c6282 [AMD] DSv4: fuse the qk-norm-rope pair on the MTP target-verify path (#34973) 3 people 2026-08-21 01:08:22 -05:00
• a688682f4b [AMD][CI] Fix ROCm 7.0's dead apt index fail the MORI dependency install (#35764) 3 people 2026-08-20 22:13:09 -07:00
• 73a2c117c6 Support mxfp8 KV cache in PD transfer (#35718) Ke Bao 2026-08-21 13:05:06 +08:00
• f64080fbaf [AMD] CI: cut two setup cycles from the AMD multimodal-gen lanes (#34483) Michaelandquitenode 2026-08-20 22:01:51 -07:00
• 3efa057449 [docs] Retune the Qwen3.8-27B RTX 5090 DFLASH2 cells against 1cf2b8c (#35786) Jimmy Shong 2026-08-20 21:54:51 -07:00
• 78c964d9d7 [AMD] Retry transient network failures in ROCm Dockerfile curl fetches (#35654) Bingxu Chen 2026-08-21 12:54:24 +08:00
• 34180a0d35 [AMD] Improve K3 dspark draft attn kernel perf (#35499) Thomas Wang 2026-08-21 12:48:59 +08:00
• bda9952377 [AMD] DeepSeek-V4 MI355X: eliminate bpreshuffle fp8-scale copies at producer sites (MoE down, MLA o_proj bmm) (#33166) 3 people 2026-08-20 23:46:23 -05:00
• 6127d1daee [diffusion] feat: allow offloaded weights stay on the checkpoint mapping (#35701) Mick 2026-08-21 10:54:28 +08:00
• 44806dc507 Using unified radix tree by default for all case (#35081) Zhangheng 2026-08-21 10:45:46 +08:00
• e0cf75d9bd [doc] standardize diffusion cookbook model pages (#34247) MickandClaude Opus 5 2026-08-21 10:25:40 +08:00
• 7e80e889a2 [diffusion] Fuse LTX-2.5 decoder 3D RoPE (#35698) Xiaoyu Zhang 2026-08-21 10:13:09 +08:00
• 978244d671 [P/D disagg] Decode-side radix cache for SWA hybrid models (unified radix tree) (#27770) ishandhananiandShangming Cai 2026-08-20 20:25:06 -05:00
• aa3f766799 [CI] Temporarily disable B300 jobs (#35627) Mohammad Miadh Angkad 2026-08-21 08:54:53 +08:00
• 5bb981dce1 [diffusion] chore: read the cgroup this process is actually in (#35707) Mick 2026-08-21 08:48:50 +08:00
• f825d72936 [Sampling] Restore finite top-k requirement for sampling masks (#35205) Nan Jiang 2026-08-20 16:55:25 -07:00
• 14795dcb1a [docs] Point the Qwen3.8-27B DFLASH2 note back at the rolling dev image tag (#35767) Jimmy Shong 2026-08-20 16:43:55 -07:00
• 67f6ad61d9 fix(kernel) Fix Helion small-token prefill bug (#35197) ethcheandEthan Che 2026-08-20 19:41:59 -04:00
• ba8e601358 [docs] fix note formatting in sglang-d documentation (#35761) Yihao Wang 2026-08-20 16:39:39 -07:00
• 1a138e13b9 [docs] Tell Qwen3.8-27B DFLASH2 users to build from main (#35753) Jimmy Shong 2026-08-20 16:34:44 -07:00
• 779e593bd1 Fix _GenerationStreamAccumulator logprob_end off-by-one under retract (#26510) Shenxiu LiuandQiaolin Yu 2026-08-20 16:03:13 -07:00
• 5a7b26c636 [AMD] [sgl-kernel] Bypass caches for peer traffic in ROCm custom all-reduce (#32832) Wenkai DuandHubert Lu 2026-08-20 15:24:18 -07:00
• a4ef828207 fix(openai): avoid duplicate routed expert in response when return_meta_info = True (#35323) Jiajun Li 2026-08-20 15:21:58 -07:00
• 94907f05c4 Add CI permissions for four contributors (#35600) Richard Gong 2026-08-20 15:18:18 -07:00
• 0149f56e84 [CI] Gate /rerun-test on commenter trust and remove /rerun-stage (#35750) Liangsheng Yin 2026-08-20 15:05:52 -07:00
• 92eeed41d7 [Docker] Defer CUDA 13 NCCL override until after dependency resolution (#35756) Baizhou Zhang 2026-08-20 14:48:07 -07:00
• 0f744b6848 feat: make mm_inputs msgpack-native (#29656) ishandhananiandAlex Nails 2026-08-20 16:30:07 -05:00
• 5a100d9086 [misc] Trim restating comments and docstrings in srt/managers (#35622) Liangsheng Yin 2026-08-20 14:18:40 -07:00
• ad367d72b0 [Kimi K3] Select FlashInfer MXFP4 for SM107 auto MoE (#35554) Lee Nau 2026-08-20 14:09:31 -07:00
• d9f6861359 [docs] Add DFlash2 speculative cells to the Qwen3.8-27B cookbook (#35663) Jimmy Shong 2026-08-20 13:26:55 -07:00
• eac91ac362 [Fix] Land the decode mamba checkpoint depth on the tree page under DCP (#35412) 3 people 2026-08-20 12:15:42 -07:00
• 308bc1228b 📝 [NPU] Clean up quantization comments (#34829) Junlin Wuandronnie_zheng 2026-08-21 03:03:58 +08:00
• 61fa64ae7e feat(grpc): expose KV event discovery metadata (#35714) Connor Carpenter 2026-08-20 14:39:24 -04:00
• 2ef0fe4669 TP/PP Consensus checker (#34406) Chao Shi 2026-08-21 01:36:03 +08:00
• 23cb04093c fix(multimodal): keep LLaVA image fetch off the CPU-preprocess timeout budget (flaky test_mixed_batch) (#35700) Shangming Cai 2026-08-21 01:26:30 +08:00
• ba97cc6397 Skip empty linear-attention state buffers in PD transfer (#35689) Ke Bao 2026-08-21 01:00:50 +08:00
• 81df6f2c57 [MUSA] Harden CI dependencies and diffusion warmup (#35610) R0CKSTAR 2026-08-21 00:40:24 +08:00
• be373395b4 [diffusion] feat: support out-of-tree models and pipelines (#35713) Mick 2026-08-21 00:33:34 +08:00
• 7f8f030000 [diffusion] feat: let every layerwise component be configurable (#35688) Mick 2026-08-20 22:38:05 +08:00
• 04444ee352 [diffusion] Refresh eager optimization skills and benchmark safeguards (#35679) Xiaoyu Zhang 2026-08-20 22:03:57 +08:00
• 9b249a25a1 test: switch the Inkling-Small NVFP4 deterministic suite to DSPARK (#35293) Shuwen Wang 2026-08-20 21:54:52 +08:00
• b03ac355e7 [NPU] [FIX] Fix non-contiguous parameter issue in FIA operator (#34936) silencejade 2026-08-20 20:43:40 +08:00
• c98f1ccedb [NPU]Ensure tensors allocated by empty_like are contiguous (#34935) Estrella-xx 2026-08-20 20:40:57 +08:00
• a4ffb996db [Fix] Keep deterministic GDN prefill on Triton (#35632) Mohammad Miadh Angkad 2026-08-20 19:38:53 +08:00
• 82c6fc2db9 [diffusion] quant: support pruned safetensors checkpoints for minimax-h3 (#35418) Mick 2026-08-20 19:34:14 +08:00
• 97efc0507c [diffusion] feat: plan pinned host memory against the cgroup cap not the machine (#35641) Mick 2026-08-20 19:32:29 +08:00
• 710267dc4c [Quant] Load compressed-tensors kv_cache_scheme scales (#35455) Jimmy Shong 2026-08-20 04:17:59 -07:00
• cf3813f4ce [diffusion] feat: add weight source reader (#35668) Mick 2026-08-20 18:37:29 +08:00
• 17313cf4b2 [diffusion] CI: add minimax-h3 ref2va audio consistency coverage and guard peak vram (#35511) Mick 2026-08-20 17:22:05 +08:00
• f1b9a1f42a [diffusion] feat: support unverified short edge instead of rejecting it for minimax-h3 (#35664) Mick 2026-08-20 16:52:44 +08:00
• 06ad7b2b0d [AMD][CI] Run Both ROCm 7.2.4 and ROCm 7.2.0 Images on Nightly Test AMD (#35603) Bingxu ChenandCursor 2026-08-20 16:37:23 +08:00
• f386e2a471 [AMD][CI] Default the ROCm 7.2 PR gate to ROCm 7.2.4 Image (#35602) Bingxu Chen 2026-08-20 16:36:26 +08:00
• 21c88f8625 [diffusion] quant: support gguf (#35370) 3 people 2026-08-20 00:46:34 -07:00
• ae23423b46 Split TRTLLM MHA decode batches by KV sequence length (#34888) YAMY 2026-08-20 00:44:26 -07:00
• b8996a5ab2 [diffusion] fix: keep large vocab tables in host memory under layerwise offload (#35626) Mick 2026-08-20 15:20:06 +08:00
• 7ba3430365 Fix Grok-2 nightly: derive image-understanding capability from is_multimodal (#33730) YC Yen-Ching Tseng 2026-08-20 08:56:51 +02:00