Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• 852b04b358 fix(xpu): read enable_deterministic_inference from the config bag (#36149) Shangming Cai 2026-08-24 13:56:41 +08:00
• ec334b17e2 xeon ci fail fast strategy change (#36146) MingxuZh 2026-08-24 13:51:22 +08:00
• 6d40b8aebf [diffusion] Fix Hunyuan QKV pack indexing at production video shapes (#36009) Xiaoyu Zhang 2026-08-24 13:39:26 +08:00
• b43931e878 [diffusion] Refresh quality and BCG benchmark skills (#36016) Xiaoyu Zhang 2026-08-24 13:37:45 +08:00
• 77940dec80 [MoE] Gather the cutlass MoE activation and its scales in one launch (#34915) Yuan Luoandluoyuan.luo 2026-08-24 13:37:14 +08:00
• 3fe18f13cd [diffusion] feat: add plain component weight overrides (#36086) Mick 2026-08-24 13:33:23 +08:00
• 8df3b9eff9 [diffusion] feat: support loading mixed w4a8 text encoders (#36037) Mick 2026-08-24 13:31:27 +08:00
• 344613c159 [diffusion] Default Hunyuan VAE to tiled decode (#36012) Xiaoyu Zhang 2026-08-24 13:18:06 +08:00
• f98b60de80 fix(xpu): enable compressed-tensors FP8 W8A8 on XPU (RedHatAI FP8-dynamic models) (#33057) vikram singh shekhawat 2026-08-24 10:23:48 +05:30
• b498efce52 chore: move cuda_vmm_utils.py under srt/utils/ (#36053) Lianmin Zheng 2026-08-23 21:47:29 -07:00
• 56834422a1 [Intel XPU] Add xpu pass for biased_topk and hash_topk (#33323) gaopengffandMa Mingfei 2026-08-24 12:18:22 +08:00
• 1daa94a069 [CPU] Fix NUMA/core binding for DP ranks (#32856) Chunyuan WU 2026-08-24 12:15:38 +08:00
• 3e30649064 [Fix] Harden FlashAttention CUDA graph metadata bounds (#35454) Aurick QiaoandYuwei An 2026-08-23 21:11:46 -07:00
• 5b5b29d4e2 [XPU] Use a fused GDN kernel from sgl-kernel for Qwen3.5 (#33354) Xia Weiwen 2026-08-23 21:07:50 -07:00
• 8dcfb3b5e7 [diffusion] Fuse LongCat-Image QKNorm and interleaved RoPE (#35995) Xiaoyu Zhang 2026-08-24 12:07:26 +08:00
• 09592f5889 [diffusion] Keep LongLive2 components resident on large GPUs (#35993) Xiaoyu Zhang 2026-08-24 12:06:52 +08:00
• acba8921bf [XPU] Support softmax_lse in sgl_kernel::fwd API (#33840) Xuan LiaoandMa Mingfei 2026-08-24 12:05:15 +08:00
• 7a7b655ddf [quantization] share bounded post-load device staging (#35180) Mick 2026-08-24 11:48:30 +08:00
• 5ce700aee8 [diffusion] feat: infer LoRA alpha from safetensors metadata (#36082) Mick 2026-08-24 11:46:18 +08:00
• 230c052ebc [diffusion] chore: reuse srt AutoRound for quantized DiTs (#36068) Mick 2026-08-24 11:21:44 +08:00
• f294d51a71 [diffusion] fix: fix a refit key error on mapped weights, and stop claiming strides the reload discards (#35832) MickandClaude Opus 5 2026-08-24 11:11:35 +08:00
• 0c1e9bda57 [OpenAI] Drop empty assistant turns for mistral_common tokenizers (#35915) Alison Shao 2026-08-23 20:07:17 -07:00
• 6ca872a11f [diffusion] chore: fetch metadata beside nested lora weights (#36057) Mick 2026-08-24 11:02:23 +08:00
• 1a368eca1c [diffusion] optimization: reuse minimax h3 prompt refinement across outputs (#36027) Mick 2026-08-24 11:01:21 +08:00
• a90d770c40 [Weight Cache] Support static DP/EP layouts (#33684) Xun Sunandsiyu 2026-08-24 10:57:52 +08:00
• 514b997e6c Register CPU CI for 17 e2e tests and partition xeon base-c suite (#35227) 5 people 2026-08-24 10:49:38 +08:00
• 11b1b4c374 [NPU] [DOC] Polish English wording in NPU docs (#36123) amote-i 2026-08-24 10:48:36 +08:00
• fbdec2855a [XPU] Support INT4 dense linear (AWQ/GPTQ) for XPU (#30236) YangKai0616andgemini-code-assist[bot] 2026-08-24 10:41:59 +08:00
• f6fff25756 [diffusion] feat: support vae weight-file overrides (#36085) Mick 2026-08-24 10:33:14 +08:00
• 2d84de5e69 [diffusion] feat: support loading serialized comfy w4a8 checkpoints (#36036) Mick 2026-08-24 10:32:21 +08:00
• fd73d4b019 [CPU] Add graph register for fused_sigmoid_mul_cpu, fused_qk_gemma_rmsnorm (#35506) YanbingJiang 2026-08-24 10:18:57 +08:00
• 4c02584773 Add intel_xpu to DETERMINISTIC_ATTENTION_BACKEND_CHOICES (#29143) Kalyan Kumarandvshekhawat-hlab 2026-08-24 07:17:34 +05:30
• fee00a41db [diffusion] feat: add composable component weight path cli (#36078) Mick 2026-08-24 09:31:08 +08:00
• 1c1c9d9b4e [diffusion] refactor: reuse srt quantization contracts and mxfp8 kernels (#36063) Mick 2026-08-24 09:26:54 +08:00
• 167c339c8e [CPU] Add check for fused_input_proj in TP=4 (#35669) YanbingJiang 2026-08-24 09:21:03 +08:00
• e129fe21e5 [diffusion] Flatten Wan VAE RMSNorm row addressing (#35981) Xiaoyu Zhang 2026-08-24 08:57:54 +08:00
• b2eb0fa51e [diffusion] Keep Cosmos3 Nano resident on high-memory GPUs (#36000) Xiaoyu Zhang 2026-08-24 08:51:58 +08:00
• 447048dba2 [diffusion] Reject unsafe quality=high BCG replay (#36008) Xiaoyu Zhang 2026-08-24 08:50:26 +08:00
• f4448e677f [diffusion] Reuse SANA fast paths in SANA-Video BCG (#35961) Xiaoyu Zhang 2026-08-24 08:47:04 +08:00
• 20064623ab [AMD][CI] Add the Qwen3.8 MXFP4 MI35x nightly (#35383) Michael 2026-08-23 17:17:10 -07:00
• fb6e3872e1 [Mamba] fix mamba index h unexpected assertion for dcp (#36005) billishyahao 2026-08-24 08:15:33 +08:00
• 95f5ecd3d2 [AMD] Update amd deepseek v4 cookbook 0822 (#35854) Thomas Wang 2026-08-24 04:26:39 +08:00
• d1af3c8923 [diffusion] feat: support loading native diffusers miniMax h3 components (#36067) Mick 2026-08-23 22:13:59 +08:00
• de6a1dbd7a [diffusion] feat: support hybrid conditioning for minimax h3 (#36080) Mick 2026-08-23 21:51:25 +08:00
• 8014d9d062 [npu] Kill evalscope session by process group and fix report score parsing (#35988) pllimaxandsglang-npu-bot 2026-08-23 21:31:42 +08:00
• 939c00a7e3 [diffusion] feat: support compact qwen3-vl conditioning for minimax h3 (#36076) Mick 2026-08-23 21:25:11 +08:00
• 9b1b06b8e6 [NPU] [DOC] Add Ascend NPU (A3) recipe to the Kimi-K3 cookbook (#35508) amote-i 2026-08-23 21:21:27 +08:00
• 886e37a649 [diffusion] CI: guard the anonymous-host budget alongside peak VRAM (#36051) MickandClaude Fable 5 2026-08-23 20:13:46 +08:00
• dd15fb57b5 [diffusion] feat: automatically infer comfy fp8 activation scaling (#36060) Mick 2026-08-23 18:43:33 +08:00
• 44db041700 [NVIDIA] Fix SM107 MXFP8 activation prep (#35405) Sahithi ChigurupatiandMohammad Miadh Angkad 2026-08-23 03:17:03 -07:00
• 27aa48bca1 [Fix] lfm2 detector: recover tool calls dropped by common model-outpu… (#34237) 3 people 2026-08-23 03:09:43 -07:00
• e3a008a9db [diffusion] UX: clean up startup and offload logs (#36034) Mick 2026-08-23 16:50:25 +08:00
• 340391a297 config: publish before the launcher reads effective configuration (#35910) Cheng WanandClaude Opus 5 2026-08-23 01:20:20 -07:00
• a43592dce5 config: pin two orderings resolution relies on (#35909) Cheng WanandClaude Opus 5 2026-08-23 01:19:44 -07:00
• 362c2ee849 config: borrowed-record reads follow the config bags (#35908) Cheng WanandClaude Opus 5 2026-08-23 01:19:20 -07:00
• 64aa859da2 config: constructing a config no longer resolves it (#35907) Cheng WanandClaude Opus 5 2026-08-23 01:18:53 -07:00
• 4bc79a1b49 config: project the config bags from the resolution result (#35906) Cheng WanandClaude Opus 5 2026-08-23 01:18:24 -07:00
• 0e22777572 config: record resolution writes in a declaration stash (#35905) Cheng WanandClaude Opus 5 2026-08-23 01:17:27 -07:00
• 6218d6ce3f config: a defensive publish must not re-project over a live process (#35904) Cheng Wan 2026-08-23 01:16:39 -07:00
• bd3cc97e7e [diffusion] CI: let the 5090 consumer case runs two warm requests on the full recipe (#36032) MickandClaude Fable 5 2026-08-23 15:39:32 +08:00
• 155aa26c19 [AMD][DSV4] perf: use full 1024-thread block for indexer top-k on ROCm (#36004) karverma-amd 2026-08-23 02:22:30 -05:00
• edd675cecf [AMD] Add Radix-4 MoE top-k router kernel for Kimi-K3 routing (#34490) Xinyi Song 2026-08-22 23:57:35 -07:00
• 8270ac4621 refactor(disagg): move _is_watermark_ready into StagingManagerMixin (#36030) Shangming Cai 2026-08-23 14:18:41 +08:00
• 849ce71976 refactor(disagg): dedupe mooncake failure_exception into a mixin (#36031) Shangming Cai 2026-08-23 14:17:44 +08:00
• c9f6b9ba25 [NPU] [DOC] Refresh supported features and models on Ascend NPU (#35836) amote-i 2026-08-23 13:05:03 +08:00
• bd72190728 refactor(disagg): register SGLANG_ENCODER_MM_LOAD_WORKERS in Envs (#36006) Shangming Cai 2026-08-23 12:47:38 +08:00
• 97ae27893a [diffusion] feat: release a layerwise component's non-layer weights between uses (#35734) Mick 2026-08-23 12:05:24 +08:00
• 70319a0881 [diffusion] feat: support loading serialized comfy convrot int8 native encoders (#36023) Mick 2026-08-23 12:00:38 +08:00
• bbbcbf9418 [diffusion] fix: stabilize ltx-2.3 two-stage cold requests (#35997) Mick 2026-08-23 10:04:00 +08:00
• 59cdc9de7d [diffusion] chore: re-home decode-dtype vae weights to a file-backed store (#35986) MickandClaude Fable 5 2026-08-23 09:32:04 +08:00
• a36c0746b9 [diffusion] feat: support serialized comfy convrot int8 dits (#35994) Mick 2026-08-23 09:29:59 +08:00
• 7f30d66045 [Kimi-K3] Fix "wrong grids" crash in DP-sharded vision preprocessing (#35305) Chungman LeeandMohammad Miadh Angkad 2026-08-23 10:20:51 +09:00
• eec794bce0 [AMD][Spec] Fix aiter GQA packing + split-KV routing in NEXTN spec attention (verify & draft_extend) (#30105) Hồ Sỹ Thế 2026-08-23 04:22:43 +07:00
• cce0a1244b refactor(disagg): hoist staging helper imports out of the bootstrap loops (#35980) Shangming Cai 2026-08-23 01:24:58 +08:00
• 7d22b7a875 [diffusion] docs: add tuning guide for h3 on consumer-level gpu (#35816) MickandClaude Opus 5 2026-08-22 23:48:40 +08:00
• b98d472158 [NPU] [Diffusion] Fix critical Ascend NPU Diffusion regression/bugs & restore 2-NPU CI testcase (#34855) 5 people 2026-08-22 18:32:20 +03:00
• 0b064e3739 [diffusion] comfyui: add a minimax-h3 node and a generic extra-fields passthrough (#35352) Mick 2026-08-22 23:11:49 +08:00
• 453b98c490 [diffusion] feat: support single-file component weight overrides (#35979) Mick 2026-08-22 23:09:35 +08:00
• 46cb12ab45 [diffusion] fix: fix hunyuan3d stale extension lock hangs (#35989) Mick 2026-08-22 23:08:44 +08:00
• d315eb7250 [AMD] DeepSeek-V4: add aiter fused mHC post+pre with cross-layer boundary dispatch (#32577) 3 people 2026-08-22 09:34:57 -05:00
• db570fe619 [diffusion] chore: let the auto policy select h3's dit for layerwise offload (#35812) MickandClaude Opus 5 2026-08-22 21:36:46 +08:00
• 61981e1fcd [diffusion] optimization: keep vae decoder weights in their decode dtype from load (#35967) MickandClaude Fable 5 2026-08-22 21:31:06 +08:00
• 3c69a4c744 fix(test): unbreak test_kv_transfer_replica_metric after #35950 (#35974) Shangming Cai 2026-08-22 19:23:46 +08:00
• a8c16b2e55 [diffusion] feature: use the directory for the vae mapping gate (#35946) MickandClaude Opus 5 2026-08-22 19:12:48 +08:00
• 489e605b35 [diffusion] feat: admit compatible quantized native encoders (#35962) Mick 2026-08-22 18:51:01 +08:00
• 382343f860 [diffusion] optimization: transfer mapped layers through a courier thread (#35882) MickandClaude Opus 5 2026-08-22 17:17:14 +08:00
• cb10ca16dd [FEAT] Weight Daemon abstraction (#33279) siyuandliusy58 2026-08-22 17:13:46 +08:00
• 5c03069d4b refactor(disagg): drop dead placeholder overrides in Common KV sender/receiver (#35950) Shangming Cai 2026-08-22 15:45:18 +08:00
• 15a4398320 refactor(disagg): hoist duplicated _handle_staging_req into a mixin (#35948) Shangming Cai 2026-08-22 15:35:55 +08:00
• b391ef171f [diffusion] feat: load serialized bnb4 components with transformers (#35945) Mick 2026-08-22 15:33:53 +08:00
• 90354326c7 [VLM] Split Pixtral multi-image features before the CUDA IPC wrap (#35463) Mohammad Miadh Angkad 2026-08-22 15:17:55 +08:00
• 6fd0384d42 Make draft attention backends extensible (#35932) Lianmin ZhengandYichao Fu 2026-08-21 23:54:44 -07:00
• af39ad9349 [Model] Complete dots.note.omni support with native encoders, video preprocessing, and MTP decoding (#33829) Jianfei Wangandmiraclezqc 2026-08-22 14:19:14 +08:00
• c35683fda0 [HiCache] Clamp tombstoned SWA locs in UnifiedSWAKVPool translation (#35933) Zhiqiang Xie 2026-08-21 23:08:56 -07:00
• 4cb5aebfe0 [docs] Re-measure the Qwen3.8-27B RTX 5090, RTX PRO 6000 and DGX Spark grids on 1cf2b8c (#35825) Jimmy ShongandClaude Fable 5 2026-08-21 22:03:18 -07:00
• 96bfd2476c [diffusion] Enable SANA-Video breakable CUDA graphs (#35729) Xiaoyu Zhang 2026-08-22 12:57:09 +08:00
• 83e9ece672 [diffusion] Fuse SANA-Video interleaved RoPE (#35695) Xiaoyu Zhang 2026-08-22 12:56:45 +08:00
• 3e096629cf [CI] Re-enable B300 jobs (#35607) Mohammad Miadh AngkadandMohammad Angkad 2026-08-22 12:34:21 +08:00
• 5290327025 [diffusion] feat: resolve hub component subfolders (#35939) Mick 2026-08-22 12:01:44 +08:00
• ac179eec11 fix(disagg): PD transfer-failure injection was silently inert (#35890) Shangming Cai 2026-08-22 11:51:16 +08:00