Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

main

Mono Color

• 2b1c4e4c85 [NPU] Set DEEPEP_HYBRID_DEPLOYMENT=1 for collocated DeepEP test cases (#38667) pllimax 2026-09-10 00:08:07 +08:00
• ffe98a4279 [Diffusion][MiniMax-H3] Add SM90 Sage compute for SubBlock sparse attention (#37982) HuangJi 2026-09-09 20:00:03 +08:00
• dba34cc964 [NPU] Bump memfabric and sgl-kernel-npu versions in docs and pyproject_npu.toml (#38437) Even Zhou 2026-09-09 19:53:49 +08:00
• 708f51e44b [AMD][CI] Make ROCm 10 the Default for AMD PR and Nightly Tests (#38659) YC Yen-Ching TsengandChen Bingxu 2026-09-09 11:28:20 +02:00
• 72d5c5bb73 [Kimi-K3] Accept fp32 routing weights in the fused MoE finalize (#38612) Mohammad Miadh AngkadandMohammad Angkad 2026-09-09 16:33:39 +08:00
• d10ebdd0cb docker(xpu): unblock nightly build (setvars.sh + sgl-kernel rename) (#38617) ashwini rathiandarathi-hlab 2026-09-09 14:02:16 +05:30
• 07c7b2674d ci(xpu): merge stage-a+b into one job and trim main_package scope (#38014) ashwini rathiandarathi-hlab 2026-09-09 13:56:54 +05:30
• 1ad3eb09a9 [Attention] Size FlashInfer MLA indptr buffers to the padded max batch (#38590) YAMY 2026-09-09 01:25:29 -07:00
• 32d7d943d1 use private --shm-size instead of --ipc=host to stop /dev/shm leak (#38638) MingxuZh 2026-09-09 16:21:24 +08:00
• 35df2fecde [Fix] Stamp sequence-parallel state on dummy forward batches (#38564) Pranjal Shankhdharandpranjalssh 2026-09-09 01:03:03 -07:00
• 13469c16d3 Store mamba prefix-cache checkpoints at the configured SSM state dtype (#34820) Yuhao Yang 2026-09-09 15:25:57 +08:00
• daf66f6670 [Diffusion][SenseNova] support SenseNova-U1.5-8B-MoT (#36606) 4 people 2026-09-09 15:12:38 +08:00
• 78da625190 [mem_cache] skip duplicates host evict via environ (#38462) Liwansi 2026-09-09 14:31:34 +08:00
• a27be5ff62 [Diffusion] Enable lossless BCG for FLUX.1-dev (#38591) Xiaoyu Zhang 2026-09-09 14:09:21 +08:00
• 5998e9321c [AMD][Fix] Fix regression in jit build error with FLUX.2-dev on gfx1250 (#38329) sogalin_codegen 2026-09-09 13:55:13 +08:00
• 00a9028e87 [diffusion] feat: add explicit snapshot-offload component residency (#38535) MickandMick Qian 2026-09-09 13:44:54 +08:00
• 0ee8e41a4e [diffusion] CI: don't fail the nightly when a perf dump is unreadable (#38629) MickandMick Qian 2026-09-09 13:39:10 +08:00
• eb42598bdd [AMD][DSV4] Skip the paged SWA page return under the per-request ring (#38571) Ankith AverineniandHaiShaw 2026-09-08 21:15:30 -07:00
• 03d06a764e [Docs] Add measured JoyEcho H200 residency and BCG recipe (#38534) Xiaoyu Zhang 2026-09-09 11:19:27 +08:00
• 2952c8d5ea [Diffusion] Fuse LongCat Image normalization and modulation (#38530) Xiaoyu Zhang 2026-09-09 11:12:17 +08:00
• 76eea36e38 Cast fp32 routing weights to bf16 in the Kimi-K3 fused finalize (#38588) Alison Shao 2026-09-08 19:48:21 -07:00
• 7a464a7014 [Rust TreeCore] Harden runtime and CI parity (#37303) Jialin Ouyang 2026-09-08 19:40:22 -07:00
• e54ff1efb9 [CP V1 Deprecation 5/5] Update prefill CP documentation (#36230) Baizhou Zhang 2026-09-08 19:27:49 -07:00
• 8ab9982851 [NPU]support mf device urma and host rdma trans type (#38174) James 2026-09-09 09:44:19 +08:00
• db1de6ff4c [Diffusion] Keep the Wan VAE decoder channels_last and add a Triton NHWC nearest upsample (#38182) DayuxiaoshuiandXiaoyu Zhang 2026-09-09 09:31:35 +08:00
• a67a31aa81 [CPU] Support Qwen3.8 text+video: adding torchcodec, ffmpeg and removing pin_memory (#35492) YanbingJiang 2026-09-09 09:22:58 +08:00
• da821aad11 [XPU][Diffusion] Enable MiniMax H3 on XPU platforms (#33366) jianan-gu 2026-09-09 09:20:04 +08:00
• 295132c4a5 [NPU] Add sparsity-driven KV offload for DeepSeek DSA on Ascend (#33089) 6 people 2026-09-09 09:13:22 +08:00
• a8e45f16cc [diffusion] feat: support mixed INT8 embeddings and Comfy NVFP4 encoders for minimax-h3 (#38506) MickandMick Qian 2026-09-09 08:36:43 +08:00
• 65400bb420 [diffusion] model: support MiniMax-H3 singularity hybrid checkpoints (#38455) MickandMick Qian 2026-09-09 08:35:39 +08:00
• db272201a2 [Config] Retire get_global_server_args, and clear the deprecated flags that have a replacement (#38375) Cheng Wan 2026-09-08 16:42:12 -07:00
• 5177a3ec08 MiniMax-M3: Triton split-K router GEMV with in-kernel fixup (#36557) Chunan ZengandKevin Mi 2026-09-08 16:39:41 -07:00
• a25bbca8ed MiniMax-M3: share the sparse index top-k across layers and reuse the decode top-k buffer (#36527) Chunan ZengandKevin Mi 2026-09-08 16:39:01 -07:00
• ed183d45ac [CP V1 Deprecation 4/5] Canonicalize prefill CP API names (#36229) Baizhou Zhang 2026-09-08 16:03:36 -07:00
• 559c7fa75b Revert "PD disaggregation, isolated transfer, prefill OOM fixed." (#38572) Qiaolin Yu 2026-09-08 15:57:44 -07:00
• 52fecfdf09 support qwen 3.8 flash next (#37500) +313 people 2026-09-08 13:56:21 -07:00
• afe90a8bc9 Point Ling-3.0-flash-VL cookbook install section at the model image (#38539) Xinyuan Tong 2026-09-09 01:51:33 +08:00
• 30e7a3072d Keep fp32 routing weights in the fp8 block-scale and bf16 trtllm MoE (#33631) Brayden ZhongandBrayden Zhong 2026-09-08 11:56:40 -04:00
• 8a0863c728 [HiCache] Add MLA host-dedup primitives (#36800) HZYandZhangheng 2026-09-08 17:40:21 +02:00
• 482e9f257b Add Ling-3.0-flash-VL cookbook (#38434) Xinyuan Tong 2026-09-08 23:00:24 +08:00
• 325ab245a1 [DCP] Allow fi_a2a on single-node systems Blackwell without MNNVL fabric ( ex B200 B300) (#37767) 3 people 2026-09-08 21:55:36 +07:00
• beecfda314 [observability] Fix missing e2e/decode/inference latency span attributes (#37789) Liu Ziming 2026-09-08 22:50:55 +08:00
• 15ff470472 [diffusion] chore: key the VAE decode-dtype store by module layout (#38496) 3 people 2026-09-08 22:37:21 +08:00
• a6b542813f fix(glm-5.2-nvfp4): bound Mooncake synchronous transfer batches (#32758) HZY 2026-09-08 16:14:33 +02:00
• 4df5df911b [Scheduler] Add HRRN schedule policy to significantly reduce TTFT (#32911) SovietPowerandXiaoyu Zhang 2026-09-08 22:14:03 +08:00
• 5097f9ac95 Disable Hopper GLM shared-expert fusion for modelopt_fp4 Marlin (#37325) Po-Han Huang (NVIDIA) 2026-09-08 21:16:36 +08:00
• 2d339ddef1 [NPU] Phase A calibration: full GSM8K eval for dp_attention mixed-chunk (#38422) pllimax 2026-09-08 21:14:48 +08:00
• 554f817948 [Diffusion] Optimize LTX-2 QKNorm and split RoPE on Hopper (#38396) Xiaoyu Zhang 2026-09-08 19:05:02 +08:00
• 88a9bfd1ff [diffusion] CI: add per-case timing tolerance for host-I/O-bound perf guards (#38457) 3 people 2026-09-08 18:40:17 +08:00
• e634ba78a4 [AMD] gfx950 assembly attention for EAGLE verify, draft extend and decode (#37465) zijiecandZijie Chen 2026-09-08 03:10:09 -07:00
• 141febf329 [AMD] fix: use the hardware fp8 e4m3 convert on gfx950 (#37140) amd-danli103 2026-09-08 18:01:53 +08:00
• 775f17b07c [diffusion] optimization: stream mapped weights on a shared host/device pool (#38441) 3 people 2026-09-08 17:52:52 +08:00
• 7edcdd5ae6 [AMD] Skip AITER FP8 ASM prefill when GQA is unsupported (#38467) vorapolsiloai 2026-09-08 12:49:47 +03:00
• ccfa120dae Fix DSA compression tail capacity for PD decode request slots (#38417) Hank Han 2026-09-08 16:50:07 +08:00
• 5ae4ccb0fd [Bench] Amortize the GDN ReplaySSM decode latency over the flush cycle (#38399) DavidLiandClaude Opus 5 2026-09-08 04:38:01 -04:00
• 4c3d47f1df [AMD] Copy MoE weight views before H2D in slow-loading nightlies (#38456) YC Yen-Ching TsengandChen 2026-09-08 10:19:28 +02:00
• f8f03910f2 【NPU】Support EAGLE when PP enabled in prefill nodes (#32207) cen121212 2026-09-08 15:45:36 +08:00
• b83a59835d sglang-server remove opaque type (#38095) Rain Jiang 2026-09-08 00:41:31 -07:00
• 5aab054ec8 [rust-server] fix p/d bootstrap across dp listeners (#36234) Sage 2026-09-08 08:22:32 +03:00
• da1c017ae8 [CI] test: initialize disable_radix_cache in scheduler stub (#38418) Shuwen Wang 2026-09-08 13:15:47 +08:00
• 61d501427d fix: make dfs weight ordering iterative (#38313) Shuwen Wang 2026-09-08 13:10:22 +08:00
• 91a45ea37e [Intel][XPU][LoRA] Enable LoRA on Intel XPU (#30345) Anupa Sajikumar 2026-09-08 10:16:41 +05:30
• 80e9a4ec74 [AMD] [Docker] Update MoRI to v1.2.3 (#38411) Duyi-Wang 2026-09-08 12:24:43 +08:00
• 92371e9887 PD disaggregation, isolated transfer, prefill OOM fixed. (#38094) Liwansiandsglang-npu-bot 2026-09-08 11:44:35 +08:00
• dfd9b5c2a4 [NPU] Enable non-greedy MTP sampling (#32495) Wuhen Duan 2026-09-08 11:18:01 +08:00
• 28ebede865 [mem_cache] Free hybrid SWA pages by one representative per page on page_size > 1 (#38159) Liangsheng Yin 2026-09-07 20:15:54 -07:00
• b23d835048 [Scheduler] Unify per-iteration request intake into ingest_requests() (#38389) Liangsheng Yin 2026-09-07 20:02:02 -07:00
• 2bf04f3a67 [Diffusion][CPU] Enable MiniMax-H3 on Xeon CPU (#35147) Xinguo Zhu 2026-09-08 10:56:42 +08:00
• 73c4cdb795 update codeowners (#38394) huangtingwei 2026-09-08 10:54:26 +08:00
• 7d2d6624b1 [NPU] Enable L1 prefix cache for Kimi-K3 W4A8 accuracy test (gpqa) (#38125) EdwardXuyandEdwardXuy 2026-09-08 10:39:58 +08:00
• cf35384fe4 [Intel GPU] Add MLA support to Intel XPU Attention backend for Prefill (#35866) Polisetty V R K Jyothendra Varma 2026-09-08 07:59:28 +05:30
• 8656901504 [Fix][DSA] Bound prefill Triton specializations for page-table stride (#37093) HZY 2026-09-08 04:23:51 +02:00
• e31e5319a0 HiCache: Reduce the number of all_reduce in check_hicache_events for PP (#37562) Chao Shiandhuangtingwei9988 2026-09-08 10:13:22 +08:00
• fba967ed9c [diffusion] Support diffusion decoder parallel tiling for LTX-2.5 (#36026) 3 people 2026-09-07 19:01:06 -07:00
• 2358916d5a [Feature][Intel XPU] Add memory saver support for Intel XPU via upstream torch_memory_saver (#29935) Siju Samuel 2026-09-08 07:08:51 +05:30
• b8a81f055d [CI] Replace the Lark queue-digest card with a daily queue-timeline chart (#38380) Liangsheng Yin 2026-09-07 18:28:34 -07:00
• c72cae201e [NPU] Fix ViT graph key layout handling (#37758) qyb233 2026-09-08 09:21:02 +08:00
• 5aa913e156 [AMD][GLM-5.2] Keep GlmMoeDsa MoE e_score_correction_bias in fp32 (#37133) 4 people 2026-09-08 09:17:54 +08:00
• f4bbf12423 docs(cookbook): Qwen3.5 FP8 on B200/B300 — trtllm-gen MoE + symm mem (#38374) Kedar PotdarandPo-Han Huang 2026-09-07 18:12:49 -07:00
• 792543f98c [Scheduler] Make request-timeout aborts rank-consistent to fix TP collective hangs (#37143) Kamil 2026-09-08 04:01:25 +03:00
• 4dcecc7891 Revert "[kernel] add fused silu mul quant fp8" (#38381) Liangsheng Yin 2026-09-07 17:36:10 -07:00
• 20ca564bf7 Add zianglih as online NVFP4 and DSA Top-K code owner (#33624) Ziang Li 2026-09-07 16:08:11 -07:00
• 85d39401c8 [CP V1 Deprecation 3.5/5] Deprecate HIP/NPU/MUSA prefill CP and remove legacy implementation (#38293) Baizhou Zhang 2026-09-07 15:39:34 -07:00
• f4b75b5c36 docs(cookbook): Qwen3.8-Flash-Next NVFP4 recipes for DGX Spark (1x, 2x) and RTX PRO 6000 (#37995) 3 people 2026-09-07 15:33:03 -07:00
• 5a5d8e47c5 [AMD][CI] Remove obsolete split-dim check from Kimi-K3 prefill test (#38288) Bingxu Chen 2026-09-08 05:42:53 +08:00
• e9e9e37ddc [AMD] Restore SWA reprefill-tail on UnifiedRadixCache when HiCache is off (#32759) amd-danli103andThomas Wang 2026-09-08 05:27:53 +08:00
• 570087ceda [AMD][DSV4] Reland unified-KV pool sizing and SWA ring accounting, fully gated (#38192) 3 people 2026-09-08 04:13:04 +08:00
• 6287ebf43a [AMD] Fix EAGLE crash when no kv_index_translator is bound on the DSA fp8 read door (#38318) jiaryang 2026-09-08 04:03:10 +08:00
• a711785475 [Kernel] Drop the vendored dense BF16 GEMM port in favor of FlashInfer 0.6.18 (#38124) Mohammad Miadh AngkadandMohammad Angkad 2026-09-08 03:59:33 +08:00
• bf68369a18 [ray] Support Ray metric backend for engine metrics (#31415) 3 people 2026-09-07 12:43:19 -07:00
• 8392c36bce [Bugfix][Mamba] Clear deferred init metadata before speculative decode (#37165) Aurick Qiao 2026-09-07 12:41:03 -07:00
• dcebe8c473 [Kernel] Add fused MoE Triton configs for Qwen3.8-Flash-Next FP8 on NVIDIA H200 NVL (TP2+EP2) (#38116) Chan ahn 2026-09-08 02:17:55 +09:00
• c99d906eff Drop the routing bias casts in flashinfer trtllm MoE (#33591) Brayden ZhongandBrayden Zhong 2026-09-07 12:04:33 -04:00
• 62bca081a3 [CI] Fix request receiver EP scale joiner test patch (#38342) Shuwen Wang 2026-09-07 23:52:32 +08:00
• f3ccd1c0e4 [diffusion] CI: baseline the e2e of ten unguarded perf cases (#38335) 3 people 2026-09-07 23:13:14 +08:00
• b5c9b68f03 [Kimi-K3] Recover the reply when the model skips the think channel (#37743) zijiexiaandXinyuan Tong 2026-09-07 07:08:42 -07:00
• e4008de757 Add MiniCPM5-2B cookbook (#38295) zijiexiaandClaude Opus 5 2026-09-07 06:30:17 -07:00
• 62a4a6ea0e [NPU] Add NPU arch35 support and enhance DSV4 processing in DeepSeek-V4 (#37373) 10 people 2026-09-07 21:08:06 +08:00
• df623d3cbd fix: keep queued Mooncake linker loads after abort (#38195) Shuwen Wangandhuangtingwei 2026-09-07 20:25:13 +08:00
• c5367fa964 [CI] Fix stale ServerArgs fake in chunked-SGMV LoRA test (#38315) Mohammad Miadh Angkad 2026-09-07 19:25:12 +08:00