Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• 0e500feae6 fix tpot by adjusting the sliding max-prefill-size window size (#34856) hanwlax 2026-08-17 09:50:48 +08:00
• 8e0499bd50 [AMD] [GLM5] Fuse shared-expert append into aiter grouped-topk (skip per-layer append kernel) (#31323) 3 people 2026-08-17 09:22:26 +08:00
• 0e178c3d22 [diffusion] chore: reuse srt siglip vision model (#34988) Mick 2026-08-17 09:16:17 +08:00
• 0aa09ab40d [diffusion] Reuse bit-exact modulation fast path for LTX-2.3 (#34930) Xiaoyu Zhang 2026-08-17 09:04:10 +08:00
• d91c3682b0 [AMD][CI] Add GPT-OSS perf benchmarks to the ROCm 7.2 nightly (#34645) 3 people 2026-08-16 16:35:09 -07:00
• f7cb328eb7 [AMD] [GLM5] Skip DSA decode indexer when kv_len <= index_topk (dense k-only fast path) (#31324) 3 people 2026-08-17 07:30:54 +08:00
• 5e73c89b34 [Spec] Simplify compute_spec_v2_logprobs signature and skip identity gathers (#35058) Liangsheng Yin 2026-08-16 16:01:26 -07:00
• a508d60295 [BCG][6/N] Allow prefill breakable CUDA graph for the Kimi archs (#34245) Yuwei AnandClaude Opus 5 2026-08-16 15:57:14 -07:00
• b7eccd642f Increase post-capture decode memory reserve (#34996) Lianmin Zheng 2026-08-16 15:31:36 -07:00
• f61f584347 Add explicit EPLB balancedness reporting modes (#34998) Lianmin Zheng 2026-08-16 15:31:11 -07:00
• 77cadf6b98 [Spec] Point multi-layer eagle's last shared-read runner at the draft runner (#35057) Liangsheng Yin 2026-08-16 15:28:12 -07:00
• c6ebcf39ee [VLM] Avoid synchronizing multimodal placeholder counts (#34995) Lianmin ZhengandJialin Ouyang 2026-08-16 15:15:49 -07:00
• 4c51248427 Support unified SWA page mapping in attention metadata (#35000) Lianmin ZhengandYonghao Zhuang 2026-08-16 15:14:50 -07:00
• 32e6fb4fdc [Frontend] Apply request header overrides to chat completions (#35001) Lianmin ZhengandYe Qi 2026-08-16 15:08:47 -07:00
• e49557b8da Support model-defined prefill input embedding width (#35002) Lianmin ZhengandLu Fang 2026-08-16 15:08:19 -07:00
• 5534380d46 [Spec] Support logprobs with DSpark speculative decoding (#34696) EthanandQAQEthan 2026-08-17 06:05:13 +08:00
• 67e12131df Build Rust extensions on demand in source checkouts (#34994) Lianmin Zheng 2026-08-16 14:58:06 -07:00
• 0e231d365a Clean up playground scripts and add PR babysitter launcher (#35018) Lianmin Zheng 2026-08-16 14:43:57 -07:00
• bae353ba55 [misc] Rename shared-read boundary to shared-read ends and fix wrapper delegation (#34982) Liangsheng Yin 2026-08-16 14:36:31 -07:00
• ace7314173 Add bit-exact guard for extra_buffer_lazy (#35030) Ke Bao 2026-08-16 23:34:46 +08:00
• d3589a7251 [diffusion] CI: tighten NVIDIA perf baselines (#35016) Mick 2026-08-16 20:54:50 +08:00
• 41abbb0d32 [diffusion] Accelerate Cosmos3 T2I QKNorm+RoPE (#34932) Xiaoyu Zhang 2026-08-16 20:15:32 +08:00
• 095ec6c997 [diffusion][kernel] Accelerate Sana BCG with bit-exact conv post-processing (#34928) Xiaoyu Zhang 2026-08-16 20:05:57 +08:00
• 3d3194f6c3 vlm: cache kimi-k3 per-image processor artifacts (#34404) Mick 2026-08-16 19:51:13 +08:00
• 968b355f12 vlm: streamline vision sdpa reshapes (#34991) Mick 2026-08-16 19:05:53 +08:00
• 0761d3f3a4 [diffusion] Accelerate lossless Ideogram norm post-processing (#34931) Xiaoyu Zhang 2026-08-16 17:20:09 +08:00
• b752f1e533 [diffusion] Enable breakable CUDA graphs for LTX-2.3 (#34929) Xiaoyu Zhang 2026-08-16 17:18:43 +08:00
• 6bb73082c8 Add skill for babysitting PR CI (#35015) Lianmin Zheng 2026-08-16 01:23:32 -07:00
• 6ab4b99bc2 [Quantization] Fix GPTQ scheme attachment broken by LinearBase.scheme default (#34962) Mohammad Miadh AngkadandMohammad Angkad 2026-08-16 15:48:02 +08:00
• 2ee0d38a85 [diffusion] chore: refresh docs, retire stale knobs, and fix nightly attribution (#34663) Mick 2026-08-16 15:41:08 +08:00
• a54de989c8 [diffusion] chore: speed up minimax-h3 vae decode on 2×h100 (#34817) Mick 2026-08-16 15:38:21 +08:00
• 8922bb98e2 refactor(hicache): flatten L2 transfer execution (#34793) cctry 2026-08-16 00:33:34 -07:00
• 56a759cffc [JIT Kernel] Migrate moe_topk_softmax from AOT to JIT (#34509) Chenzhou LiandXiaoyu Zhang 2026-08-16 15:02:57 +08:00
• 0da87024d3 [NPU] Add mxfp4-w4a8 MOE Quantization Support for NPU (#30318) LinyuanLi 2026-08-16 14:03:17 +08:00
• 24ab8f9ed9 [AMD] Qwen3.5: guard attn layers against empty DP-attention batch (#34474) Zhaoyi Liandjacky.cheng 2026-08-15 23:28:13 -05:00
• 66de161976 [Fix][AMD] MoRI EP: drop record_stream in TBO dispatch/combine (HSA out-of-resources) (#32746) 3 people 2026-08-16 12:23:48 +08:00
• 4654b927eb [HiCache] Optimize LogicalHostPool free-list release (#33998) wangwenmingaa 2026-08-16 12:17:33 +08:00
• 6314e9e4f5 [AMD][Fix] Qwen3.5: guard zero-grid launch in fused_qk_gemma_rmsnorm(_with_gate) (HIP invalid configuration on idle DP rank) (#31794) jacky.cheng 2026-08-16 12:02:54 +08:00
• 0f706c33d2 Fix swa eviction frontier for bigram keys (#34870) Ke Bao 2026-08-16 11:52:14 +08:00
• e9fe58139f [diffusion] refactor: unify component residency controls (#34736) Mick 2026-08-16 11:24:48 +08:00
• f68517f644 [AMD][Quantization][Bugfix] Fix bug related to fp8 max on gfx95x for per-token-group quant (ROCm) (#30900) Spandan Tiwari 2026-08-15 19:54:16 -07:00
• 4c0e85524d [AMD] [GLM5] Enable dense-MHA short-context prefill fallback on gfx950 (#30808) Raiden MakotoandRaiden-Makoto 2026-08-15 19:30:14 -07:00
• d269a28b47 [diffusion] refactor: route minimax h3 vae attention through native backends (#34949) Mick 2026-08-16 10:07:26 +08:00
• 4f9da62547 [diffusion] chore: use native hunyuan3d paint and delight models (#34980) Mick 2026-08-16 10:03:48 +08:00
• d106e8b23a [diffusion] chore: use native ernie prompt enhancer (#34951) Mick 2026-08-16 09:59:51 +08:00
• 19e3bd6391 [diffusion] chore: use native qwen3-vl vision encoder (#34945) Mick 2026-08-16 09:58:46 +08:00
• eb6b773149 [diffusion] chore: use native qwen2.5-vl generation (#34896) Mick 2026-08-16 09:57:39 +08:00
• 4a6dc267e1 [Spec] Support mamba-radix-cache-strategy extra_buffer_lazy with DFLASH (#34763) gilfordtingandClaude Fable 5 2026-08-15 19:36:59 -04:00
• 3802a725ac [CI] Pin the allowed_media_domains supplied-instance reads in the step-12 ratchet (#34961) Mohammad Miadh AngkadandMohammad Angkad 2026-08-16 07:35:19 +08:00
• d22c4cc177 [AMD] perf(sgl-kernel): default block_quota=16 for MLA page_first KV gather… (#30024) 4 people 2026-08-16 07:05:40 +08:00
• 0f7aaceda5 [misc] Rename the WAR read-done fastpath to shared-read-done (#34916) Liangsheng Yin 2026-08-15 15:02:02 -07:00
• 4d0c5a89af [AMD] Add concat_and_cast_mha_k_pad_kernel to support 12-head and enable K3 aiter prefill kernel (#34837) Thomas Wang 2026-08-16 05:39:30 +08:00
• dd458f3212 Fix dsv4 kl test timeout (#34963) Ke Bao 2026-08-16 01:57:40 +08:00
• e5b3a48751 Add --http2-max-concurrent-streams server arg (#34796) cctryandYilong Zhao 2026-08-15 10:34:49 -07:00
• cef8a32b9d update codeowner (#34866) huangtingwei 2026-08-16 00:37:18 +08:00
• fb97be4359 Fix Whisper transcription for audio over 30 seconds (#33604) Shenxiu Liu 2026-08-15 08:51:05 -07:00
• 4beb157e87 [diffusion] doc: define native diffusion model integration contract (#34952) Mick 2026-08-15 23:37:34 +08:00
• 5c0ace30c0 [diffusion] model: support ltx-2.5 (#34471) 3 people 2026-08-15 08:36:02 -07:00
• e331baaaa8 [diffusion] chore: scope attention backend fallback (#34891) Mick 2026-08-15 22:00:46 +08:00
• e161bd1265 Fix Python packaging shadowing in DeepEP wheel builds (#34937) Baizhou Zhang 2026-08-15 03:27:32 -07:00
• a64791b312 test: restore GLM-4.1V nightly latency threshold (#34811) Mick 2026-08-15 18:13:19 +08:00
• 35cefd1c51 feat: add safeguards for remote media URLs (#34892) Mick 2026-08-15 18:12:15 +08:00
• 0c072235f4 [diffusion] Bound overlong weight lock filenames (#34825) Xiaoyu Zhang 2026-08-15 17:20:30 +08:00
• 8d44091326 Update sgl-deep-ep release workflow for DeepEP v2 (#34914) Baizhou Zhang 2026-08-15 02:08:28 -07:00
• 7769f54feb [Kimi-K3] Use explicit SiTU activation for MegaMoE (#34883) Baizhou Zhang 2026-08-15 01:41:59 -07:00
• c87a2ced12 test(step-12): state the bag contract as what resolution produced, and the skill rule that goes with it Cheng Wan 2026-08-15 00:40:37 -07:00
• d804b6bd98 config: pin the step-12 debt on the supplied-instance surface Cheng Wan 2026-08-15 00:40:07 -07:00
• 1ab713c334 config: the post-publish consumers of the supplied-instance surface read the bags Cheng Wan 2026-08-15 00:39:35 -07:00
• f2ab6e306b config: the alias form of the runner-side instance read Cheng Wan 2026-08-15 00:39:03 -07:00
• 61908870f6 config: spell out the one dynamic config read the census could not see Cheng Wan 2026-08-15 00:38:30 -07:00
• d13d5c03ab config: decisions keyed on the attention backend read the configured pair Cheng Wan 2026-08-15 00:38:00 -07:00
• 97279980cf config: the last runner-side instance reads read the bags Cheng Wan 2026-08-15 00:37:06 -07:00
• ab810e4052 config: each runner carries its own linear-attn kernel choice Cheng Wan 2026-08-15 00:36:04 -07:00
• 6cbfa791d6 [AMD][Spec] Accelerate Qwen3.5 verification with grouped-head shared KV (#34517) 3 people 2026-08-15 15:15:52 +08:00
• 5c9ee86d90 [CI] Move the static ratchets back to CPU unit tests (#34913) Liangsheng Yin 2026-08-14 23:32:00 -07:00
• 2a89c6823a fix: honor explicit model loader classes (#34880) raghothamandehhuang 2026-08-14 23:00:41 -07:00
• e99ecb6eee [CI] Path-gate Rust workspace tests in lint (#34864) Xinyuan Tong 2026-08-15 13:41:54 +08:00
• 7216a44dd8 [AMD] Broadcast the EAGLE greedy verify decision across TP ranks on ROCm (#34238) Zhiyao Jiang 2026-08-15 01:37:45 -04:00
• aeee1562e6 [Kernel] Enable Helion backend for Kimi Delta-Attention (#32593) ethcheandEthan Che 2026-08-15 01:22:46 -04:00
• 3adbbec2fd [MoE] Route every trtllm-gen MoE call site through one PDL guard (#34789) Khoa PhamandClaude Opus 5 2026-08-14 22:09:03 -07:00
• bc7e3ba66c [AMD][Quantization] Online MXFP4 quantization 4/N - NVFP4 to MXFP4 Online Requantization on AMD GPUs (#29328) Colin Z 2026-08-14 21:59:39 -07:00
• 5afdb1caea [AMD CI] follow the miles nightly-prefixed MI350 suite names (#34877) 4 people 2026-08-15 00:57:12 -04:00
• 2012b4b196 [AMD][CI] Fix stage-b: AttributeError on multimodal embedding requests (#34769) Michael 2026-08-14 21:35:26 -07:00
• 8720a72814 fix(scheduler): track max prefill batch size over recent real admissions (#34284) hanwlax 2026-08-15 10:48:03 +08:00
• 6eb941a34c [Spec] Support MegaMoE for DSpark under dp attention (#34844) weireweireandweireweire 2026-08-15 08:20:01 +08:00
• a5ba081fbb [Spec] Wire DFLASH aux-hidden capture into the Qwen3.5 text-only wrapper (#34771) gilfordtingandClaude Fable 5 2026-08-14 20:10:09 -04:00
• 6f005e4da1 [Fix] Require JSON booleans for response_format json_schema.strict (#34777) gilfordtingandJames Liu 2026-08-14 19:46:27 -04:00
• a1844709f1 [CI] Trim Qwen3.5 FP8 GB300 performance batches (#34882) Baizhou Zhang 2026-08-14 16:45:11 -07:00
• 3f64f14360 [Fix] Work around xgrammar 0.2.1 negative integer minimum in Kimi-K3 structural tags (#34778) gilfordtingandJames Liu 2026-08-14 19:41:06 -04:00
• be804c1b83 [RL] Expose top-p-only sampling masks (#33593) Nan Jiang 2026-08-14 16:40:27 -07:00
• 90b3db6dd8 [Fix: RL] Snapshot async state-capture outputs before overlap (#34319) Jason Mancuso 2026-08-14 19:39:35 -04:00
• 8b4faa3336 [Docs] Update Kimi-K3 installation options (#34886) Baizhou Zhang 2026-08-14 16:19:23 -07:00
• 17efe37428 Remove 11 dead CODEOWNERS rules (#34875) DanielSwift1992 2026-08-14 18:38:19 -04:00
• a9654eacc1 fix(dsa): use FlashInfer fused top-k for packed PAGED rows (#33006) Ziang Li 2026-08-14 15:22:18 -07:00
• bfb224ff01 Add Reasoning-Aware Compression (RAC) pruning recipe for reasoning models (#32414) 4 people 2026-08-14 15:13:45 -07:00
• 03c1d58112 perf: add H200 Triton MoE configs for E256 N512 (#34150) 3 people 2026-08-14 14:37:58 -07:00
• 22dde1dd5b [Docs] Fill GLM-5.2 H200 FP8 speed cells (low-latency, balanced); fix MTP notation (#31554) Han-Yin ChangandClaude Fable 5 2026-08-15 04:56:55 +08:00
• b676793e5e docs: sync LMSYS SGLang blog cards (#32982) sglang-botandsglang-bot 2026-08-14 13:55:52 -07:00
• 42e8718d3d fix(muse-glimmer): parse required/named tool calls natively (#34781) Sidhartha Reddy PotuandXinyuan Tong 2026-08-14 13:43:48 -07:00
• d7207be156 Fix startup weight load after TorchAO removal (#34869) Mohammad Miadh Angkad 2026-08-15 04:18:53 +08:00