Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• 13537f8e20 Unskip Marlin NVFP4 tests (#27589) 3 people 2026-06-16 11:58:22 -07:00
• 33f205d8c5 docs(cookbook): fix GLM-5.2 thinking toggle kwarg + document reasoning effort (#28454) Xinyuan Tong 2026-06-16 19:17:34 +01:00
• 799584e173 fix: get_processor fails when --tokenizer-path lacks model config.json (#25643) 3 people 2026-06-16 22:55:05 +05:30
• 92b42c8d8a LPLB: linear-programming load balancer for MoE expert parallelism (#24515) feliang-gitandxutizhou 2026-06-17 01:19:42 +08:00
• 00081a00d5 docs(cookbook): tune GLM-5.2 MTP to 5-1-6 and simplify launch flags (#28448) Xinyuan Tong 2026-06-16 18:18:34 +01:00
• 78b6a4fabf [UnifiedTree]: Clean up some unused dead code. (#28389) Zhangheng 2026-06-16 21:53:35 +08:00
• 0cb6183432 docs(cookbook): add GLM-5.2 deployment cookbook (#28437) Xinyuan Tong 2026-06-16 14:49:25 +01:00
• 265202cda2 fix(openai): validate assistant tool call arguments before chat template (#28035) 3 people 2026-06-16 20:21:23 +08:00
• fcca4611fa [CAR] Let custom allreduce support VMM based allocation (#27593) Yinghai LuandLianmin Zheng 2026-06-16 04:49:05 -07:00
• 12ebb35439 docs: refresh README News section and add Modal to adoption list (#28330) sglang-botandClaude Opus 4.8 2026-06-16 03:45:10 -07:00
• 25e696aa8d Fix Stage B CUDA CI (#28367) Brayden ZhongandBrayden Zhong 2026-06-16 02:39:17 -07:00
• a362ba9da3 [AMD] Feat: Add prefill context parallel support for deepseek v4 unified kv attention (#27928) Wang, FangYuanandThomas Wang 2026-06-16 17:00:51 +08:00
• 149fabcca7 [AMD] Fuse sigmoid + mul into single Triton kernel for shared expert gating (#27636) jacky.cheng 2026-06-16 16:16:58 +08:00
• 102392df5b [AMD][Fix] Skip EPLB topk remap when global server args are unset (#28404) YC Yen-Ching Tseng 2026-06-16 16:13:22 +08:00
• 0fc2bc4a8b [AMD] Test DeepSeek V4 FlashMLA backend variants nightly (#28290) Bingxu Chen 2026-06-16 16:03:26 +08:00
• c5b9106c1a [perf] Use default torch compile mode for Wan2.2 T2V A14B (#28304) Xiaoyu Zhang 2026-06-16 15:45:00 +08:00
• 77f327cb6e [2/n] [CP] Add context parallel strategy abstractions (#27313) Baizhou Zhang 2026-06-16 00:20:04 -07:00
• 6c908b3a3a [UnifiedTree]: Replace anonymous tuples with NamedTuples in UnifiedRadixCache (#28375) Zhangheng 2026-06-16 14:54:04 +08:00
• 175336ff73 [Chore] update codeowner for mooncake store (#28377) Teng Ma 2026-06-16 14:51:34 +08:00
• 2a8ea70059 ✨ [llm][npu][quant] Add W8A8 MXFP8 quantization support for Qwen3 Dense on Ascend NPU (#22352) Junlin Wuandronnie_zheng 2026-06-16 14:45:18 +08:00
• 72d962be88 [AMD] Fix jit-kernel-unit-test-amd: activation.cuh ROCm build + per_token CUDA-only (R165) (#27947) Michael 2026-06-15 23:44:26 -07:00
• 556cf54d47 [Perf] Avoid per-decode-step host sync in min_new_tokens penalty (#28397) Liangsheng Yin 2026-06-15 23:32:28 -07:00
• 637c9f780b [diffusion] fix: add precision consistency layer (#27088) jy-song-hubandMick 2026-06-15 23:21:01 -07:00
• e068355831 [spec decoding] supports step 0 in adaptive spec decoding (updating draft kv cache without draft decoding) (#23994) Qiaolin Yuandshuwenn 2026-06-15 22:21:26 -07:00
• 800aaefc9e [AMD] Annotate ATOM source for imported v4 unified attention kernels (#28392) Thomas Wang 2026-06-16 13:12:12 +08:00
• 486ec150d0 [NPU] Add NPU fallback for fused Triton gating kernels (#28293) iridiumine 2026-06-16 11:37:05 +08:00
• ed9024e09f [AMD] Fix AITER Scout workflow permissions (#28360) 3 people 2026-06-16 11:21:39 +08:00
• 32685874f3 Reenable MNNVL backend for FlashInfer allreduce fusion (#23402) Shu Wang 2026-06-15 22:19:15 -05:00
• b23477af44 bench: infer tokenizer from serving model info (#28195) shuwenn 2026-06-16 11:03:12 +08:00
• 063ab89ac1 DeepSeek-V4 Online Compress support MTP (#26471) zhangxiaolei 2026-06-16 10:56:07 +08:00
• b5bcd76a41 [HiCache & JIT Kernel] Refactoring HiCache Write-Back Kernel (#21631) huangtingwei 2026-06-16 10:44:27 +08:00
• a4a8a614b1 [diffusion] UX: suppress noisy diffusers torchao warning (#28317) Mick 2026-06-16 10:32:21 +08:00
• c6d9d73fd6 [Spec][test] fix(kv_canary): assert draft-extend-v2 oracle tokens in token_oracle test (#28325) Michael 2026-06-15 19:14:09 -07:00
• 448af67a98 ci: run AMD and NPU PR tests on PRs not targeting main (#28368) Cheng Wan 2026-06-15 19:10:51 -07:00
• 407d3a91db docs: sync LMSYS SGLang blog cards (#28364) sglang-botandsglang-bot 2026-06-15 18:30:13 -07:00
• 01e45762ba [diffusion] feat: use srt custom allreduce for tp groups (#28324) Mick 2026-06-16 09:22:47 +08:00
• 2dd449ce5e [AMD-miles] add amd-miles daily docker build workflow (#27765) Zhiyao JiangandXinyu Jiang 2026-06-15 20:22:46 -04:00
• b3be2e7402 [dsv4] Pad MLA decode q-heads to 64 (not full n_heads) for FlashMLA head64 kernel (#27954) YAMY 2026-06-15 17:18:10 -07:00
• 14f6348524 [Fix] Demote OpenAIServingResponses init failure log to one-line WARNING (#28349) Liangsheng Yin 2026-06-15 16:58:33 -07:00
• 81dcb00673 [Spec v2] Use decode kernel for TRT-LLM MHA draft extend (#28241) Hanming Lu 2026-06-15 16:57:08 -07:00
• cad43d3212 [CI] Reclaim leaked /dev/shm segments on server startup (#28089) Kangyan-ZhouandClaude Fable 5 2026-06-15 16:14:08 -07:00
• 7221be2cec feat(cookbook): MTP --max-running-requests callout + skill sync (#28340) zijiexiaandClaude Opus 4.8 2026-06-15 15:53:01 -07:00
• 4c0457f440 [misc] Update codeowner (#28342) Baizhou Zhang 2026-06-15 14:42:01 -07:00
• 3a0dd69f8e Minor refactorings to the LFM2.5 cookbook for accuracy (#28072) Yi Zhong 2026-06-15 14:39:10 -07:00
• 30d8ee87b0 Add OrcaRouter usage example (#28004) jinhaosong-source 2026-06-16 05:15:26 +08:00
• 4ed698a491 fix(fa3): no NaN embeddings with fa_skip_kv_cache under piecewise CUDA graph (#27343) Jia Guo 2026-06-15 13:46:22 -07:00
• f870bf1ed0 [dsv4] Prewarm MHC prenorm kernel at startup (#27986) YAMY 2026-06-15 13:26:26 -07:00
• 7e629a2f8c Allow overriding tokenizer path in benchmark harness (#28280) Lianmin ZhengandIan O'Connell 2026-06-15 13:07:50 -07:00
• 33719cfb31 [PD] Optimize SWA allocation (#28085) 3 people 2026-06-15 11:01:55 -07:00
• 19e85868f6 [AMD] Point AITER scout at amd/aiter-ci (#28313) YC Yen-Ching Tsengandbingxche 2026-06-15 22:50:16 +08:00
• 378e66d248 [PD] Remove outdated backend whitelist for decode radix cache (#28238) Shangming Cai 2026-06-15 22:36:05 +08:00
• 20f4272109 fix: Fix DSR1 perf regression due to unnecessarily falling back to triton gemm (#28073) Trevor Morris 2026-06-15 06:45:09 -07:00
• d5899b95c4 fix(qwen3.5): keep CUDA dual-stream overlap (regressed by #25885) (#27868) YAMY 2026-06-15 06:44:21 -07:00
• 09e9c4fde3 【bugfix】The NPU's forward_dsa_prepare_npu also needs special handling for is_nextn (#28118) littleyellowbicycle 2026-06-15 21:31:39 +08:00
• e985422b2b [Fix][MTP][MM] Fix EAGLE v2 chunked-prefill next-token chain crash on multimodal models due to placeholder tokens (#27863) Kurkur 2026-06-15 21:30:45 +08:00
• 81166f382d Fix inaccuracies and add NPU constraints in ascend_npu_profiling.mdx. (#28283) qinsir5522 2026-06-15 21:12:18 +08:00
• f8d1d397b6 [NPU] fix ascend_docs (#28279) jianzhao-xu 2026-06-15 21:11:55 +08:00
• 8bdb007e58 [NPU] Docs op performance optimize (#28277) McZyWu 2026-06-15 20:28:18 +08:00
• 818808d152 [diffusion] optimize: optimize causal conv3d vae padding (#28204) Mick 2026-06-15 20:18:08 +08:00
• f768344b1a [DOCS][NPU]Supplementary Notes (#28295) loading66 2026-06-15 20:06:05 +08:00
• 7bd1a9d163 Update documentation for Ascend NPU Guide (#28284) longxin9715 2026-06-15 20:02:50 +08:00
• edd5eff519 [NPU] [DOC] fix issues in ascend_npu_support_new_models (#28296) amote-i 2026-06-15 19:58:33 +08:00
• 3df6e2f968 [NPU] Add MiMo-V2-Flash manual testcases (#28223) iridiumine 2026-06-15 19:57:49 +08:00
• eb349efb14 [EPD][BugFix] Fix encode_with_global_cache_mooncake (#28031) 3 people 2026-06-15 19:47:52 +08:00
• c4ec39a785 [AMD] refactor sparse MLA decode kernel for Deepseek V4 triton backend (#28265) 3 people 2026-06-15 18:26:59 +08:00
• da12f36629 [AMD] Refactor unified_kv attention metadata to data class and fuse c4/128 out_loc (#28275) Thomas Wang 2026-06-15 17:59:01 +08:00
• 3b419f66da [JIT] Track angle-bracket includes in source hash (#28273) Lianmin Zhengandgemini-code-assist[bot] 2026-06-15 02:58:14 -07:00
• 2a33724c9b [perf] Reuse a pooled HTTP session for multimodal URL downloads (#28056) Zhonghua Deng 2026-06-15 17:47:47 +08:00
• bf186cf8fc bugfix revise interface get cpu copy for npu mem pool to align with gpu (#27802) McZyWu 2026-06-15 17:24:10 +08:00
• 1180b70440 triton-ascend update (#27624) monkeyLoveding 2026-06-15 17:20:19 +08:00
• 9864059e2b [AMD] Update AITER commit (#28249) Bingxu Chen 2026-06-15 17:04:34 +08:00
• 19c78552dc [AMD] Restrict CI image fallback to versioned tags (#28263) Bingxu ChenandCursor 2026-06-15 16:08:48 +08:00
• 33f99831f8 docs(minimax-m3): refresh B200 benchmarks (tp8, piecewise) + add GPQA (#28207) Xinyuan Tongandzijiexia 2026-06-15 08:15:38 +01:00
• a88ba6cc0b [Fix] Reduce power of two to constant time (#28228) Jia-Wei Jiang 2026-06-15 14:25:22 +08:00
• 63df86f5e7 [AMD] Skip eplb bookkeeping and topk remap when EPLB is not in use on mori-ep / HIP (#22985) (#28188) Duyi-Wang 2026-06-15 14:00:23 +08:00
• 578e936d8d [diffusion] feat: persist torch.compile inductor/triton cache across restarts (#28205) Mick 2026-06-15 13:34:19 +08:00
• 07b9108348 [Diffusion] FLUX: fuse FeedForward GELU into up-proj GEMM (cublasLt epilogue) (#28166) 3 people 2026-06-15 13:00:26 +08:00
• 441b75ee69 [quantization] NVFP4 MoE: split fused w13 gate/up global scales (#27588) Prajjandprajjwal1 2026-06-15 00:18:36 -04:00
• ce9fad7196 [Bugfix][DeepSeek-V4] Fix Spec V2 Draft Input ID Dtype for DP Collectives (#28043) Ryan Zzzandzhujunyu 2026-06-15 12:11:53 +08:00
• bf38a0b03d Fix disaggregated decode load token accounting (#25736) weireweire 2026-06-15 11:37:41 +08:00
• 0417951a86 [Bug Fix] Validate tokenizer-dependent features with skip_tokenizer_init (#27882) 3 people 2026-06-14 20:07:52 -07:00
• 37505eca27 feat: report multimodal (image/audio/video) token counts in usage.prompt_tokens_details (#27122) Zyann 2026-06-15 11:04:10 +08:00
• c0dfe4c8ec [router] Reconcile workers that registered without resolving model_ids (#27980) Kangyan-ZhouandClaude Opus 4.8 2026-06-14 18:47:50 -07:00
• 69b02ea68a [Distributed] Guard torch symm mem all-reduce sizes (#24548) Mohammad Miadh Angkad 2026-06-15 09:28:57 +08:00
• 1a66059c4e [Spec] Restore index_share_for_mtp_iteration in EAGLE V2 draft worker (#28192) Xinyuan Tong 2026-06-15 02:01:11 +01:00
• c127ba6483 [AMD] ci: fix scheduled AMD runs startup failure when calling extra-a suite (#28214) Michael 2026-06-14 17:43:47 -07:00
• 000fc975c7 ci(docker): support layered overlay images in release-docker-dev (#28206) Xinyuan Tong 2026-06-15 00:34:02 +01:00
• f18d38d040 Revert "[AMD][Quantization] Online MXFP4 quantization 2/N - FP8 to MXFP4 requantization on AMD GPUs" (#28213) Lianmin Zheng 2026-06-14 13:34:06 -07:00
• 3cb29f6747 [diffusion] feat: use regional torch.compile (compile_repeated_blocks) for DiT of diffusers backend (#28193) Mick 2026-06-15 00:34:10 +08:00
• ec36dde580 [diffusion] feat: add --warmup-mode enum server arg (#28184) Mick 2026-06-14 23:09:04 +08:00
• 582bd23f71 [diffusion] feat: enable spatial-shard vae decode across GPUs (#28071) Mick 2026-06-14 20:19:44 +08:00
• d72314808f [JIT Kernel] Multi-GPU test/bench framework for custom all-reduce + TP QKNorm (#26706) 3 people 2026-06-14 17:20:36 +08:00
• 5331de0f8c [diffusion] chore: resolve model_index.json Hub-first with local-cache fallback (#28177) Mick 2026-06-14 16:48:38 +08:00
• 8c334e2224 fix(io_struct): index extra_key per sub-request in batched GenerateReqInput (#26971) Humphrey 2026-06-14 02:50:38 -05:00
• 1456eb612d [diffusion] CI: tighten perf baselines (#28123) Mick 2026-06-14 15:50:21 +08:00
• 54acffc864 Eval accuracy gpqa aime25 mixins (#27102) Jimmy Shongandgithub-actions[bot] 2026-06-14 00:49:12 -07:00
• f293ddf3ce [perf] reduce overhead of fill_ids list reconstruction and decref (#27965) Qiaolin Yu 2026-06-14 00:41:11 -07:00
• f2d7d67603 numa: bind within allowed CPUs when affinity is already constrained (#26983) Yongji Wu 2026-06-14 00:38:40 -07:00
• b796338271 Fix prefill delayer wait histograms always observing 0 (#25975) 4 people 2026-06-14 15:35:29 +08:00
• 8c5320b37e dflash add sliding window attention draft layer support (#27469) David Wang 2026-06-14 03:32:02 -04:00