Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

main

Mono Color

• 755f97c622 [CI] Fix the DSpark dp-tier unit test fixture after #34919 (#38314) Shuwen Wang 2026-09-07 18:16:00 +08:00
• 8ae962021d Use fp32 in TRTLLM all reduce buffers (#36143) Carrie ChenandBrayden Zhong 2026-09-07 04:57:07 -04:00
• 861d40f3ee Fix DSpark CUDA graph replay with MegaMoE TP attention (#34919) BingjiaWang 2026-09-07 16:35:35 +08:00
• ba6d3df69a [diffusion] doc: document verified GB300 and derived GB200 H3 recipes (#38296) Mick 2026-09-07 16:32:29 +08:00
• b5766336d4 [Perf] Unified memory: close the DCP decode gap on Blackwell (#37926) Cheng Wan 2026-09-07 01:10:44 -07:00
• a8edafff7c [AMD][gfx95] DSV4 wo_b (dp-attention): route to tuned bpreshuffle GEMM instead of triton (#38227) lixiufei-leoandlixiufei-leo 2026-09-07 15:36:15 +08:00
• 644841c50c [AMD] Support aiter fa mha chunked kv for Kimi-K3 (#37691) billishyahaoandHAI 2026-09-07 15:31:24 +08:00
• 4d23a4fa6d [Test] Consolidate test cleanup and CI taxonomy (net -11.4K lines) (#37436) Xiaoyu ZhangandMick Qian 2026-09-07 15:13:59 +08:00
• 6a1ff90f2d [CPU] use CustomTestCase for registered CPU tests (#38244) Ma Mingfei 2026-09-07 15:09:52 +08:00
• a88e852fab [Sampling] Allow sampling-mask replay with DisallowedTokensLogitsProcessor (#38279) Byron HsuandByron Hsu 2026-09-06 23:56:16 -07:00
• 15d2cbcc90 [diffusion] CI: validate every repeated server request (#38185) Mick 2026-09-07 14:36:50 +08:00
• cfa989b7af [CI] Fix stale CP-v1 mock patches in NextN mm-embed test (#38280) Mohammad Miadh Angkad 2026-09-07 14:20:29 +08:00
• 97dcbf9410 [XPU] Re-add intel xpu on triton paths in diffusion platforms (#36654) Kotthagattu Meher Sai 2026-09-07 11:00:22 +05:30
• e24e31efd9 Fix/whisper xpu varlen encoder decoder (#36298) 5 people 2026-09-07 10:51:29 +05:30
• 503e36cbae Tiny Update CI Permission (#38278) Baizhou Zhang 2026-09-06 22:17:48 -07:00
• 7d37b86ff2 Remove obsolete CUDA graph buffer population methods (#37886) Lianmin Zheng 2026-09-06 22:09:49 -07:00
• b6c31b155c [CP V1 Deprecation 3/5] Remove generic prefill CP v1 runtime (#36228) Baizhou Zhang 2026-09-06 21:53:53 -07:00
• aaf9a95763 [Config] Round 6.5: a namespace declares what it derives, next to what it derives it from (#38113) Cheng Wan 2026-09-06 21:44:24 -07:00
• b99175dc7d [Config] Round 6.4: the runtime reads the bags, not the record (#38049) Cheng Wan 2026-09-06 21:41:46 -07:00
• 98f69ccbf3 [Config] Round 6.3: the record remembers how it was asked for, and is sealed while resolution runs (#38048) Cheng Wan 2026-09-06 21:41:14 -07:00
• ed82def55f [Config] Round 6.2: the field declarations move to their namespaces, and the record is assembled from them (#38047) Cheng Wan 2026-09-06 21:40:31 -07:00
• 45c24444b1 [Config] Round 6.1: "unset" gets its own spelling, and the declaration says what it means (#38046) Cheng Wan 2026-09-06 21:38:29 -07:00
• df2f34cca1 Drop stale test_mla_gluon_h12_fp8.py broken by aiter gluon rewrite (#38259) Mohammad Miadh Angkad 2026-09-07 11:41:42 +08:00
• a8b2f36dee [kernel] add fused silu mul quant fp8 (#37376) 5 people 2026-09-07 11:39:11 +08:00
• 503511c963 [AMD] Cherry-pick aiter commit for dsv4 a8w8 bpreshuffle gemm config (#38256) Thomas Wang 2026-09-07 11:29:31 +08:00
• 6e312af8c2 fix: collect prefix hash values iteratively (#38204) Shuwen Wang 2026-09-07 11:10:32 +08:00
• f25848913d fix: preserve SWA host lock on node split (#38138) Shuwen Wang 2026-09-07 11:04:22 +08:00
• c8207e32b6 [Intel][XPU] Add NUMA node binding support for Intel XPU (#31113) Siju Samuel 2026-09-07 08:09:21 +05:30
• 1c992bbd94 [CPU] Fix shm allreduce collision and sglang-router import (#37179) Chunyuan WU 2026-09-07 10:28:19 +08:00
• 214313ee79 Fuse Nemotron latent MoE projection and shared add (#30430) 4 people 2026-09-07 04:26:19 +02:00
• 1d5d85260c [AMD] support qlen>1 for aiter gluon path for Kimi K3 (#37601) billishyahao 2026-09-07 10:25:58 +08:00
• b9b39f222d [AMD] Update ROCm AITER pin to 4ad9983 (#37784) kangwangamd 2026-09-07 10:22:37 +08:00
• ff08bcdda9 [diffusion] UX: quiet internal warmup frame searches (#38226) Mick 2026-09-07 09:49:13 +08:00
• e3140fb9d4 [diffusion] CI: rebalance 2-gpu shards and cut the job timeout to 45m (#38239) Cheng Wan 2026-09-06 18:39:03 -07:00
• b83f1bdd21 [diffusion] fix: stabilize H3 reference audio across repeated requests (#38225) Mick 2026-09-07 09:32:57 +08:00
• 0afba909e7 [XPU][CI] Fix empty nightly dashboard (#37800) ashwini rathi 2026-09-07 06:59:51 +05:30
• e0a83a2215 test(xpu): pin --mem-fraction-static=0.7 for DeepSeek-OCR test (#38221) ashwini rathiandArathi Rajanra 2026-09-07 06:55:11 +05:30
• c4e52a1051 XPU: Enable GLM5.1 (GlmMoeDsaForCausalLM) DSA Attention (#24959) 3 people 2026-09-07 09:24:45 +08:00
• 39a80354aa [MUSA] Add installation guide and Dockerfile (#36709) 3 people 2026-09-07 09:13:53 +08:00
• 707da81e84 [CPU] Add native CPU kernel for MurmurHash32 (#35604) Yuxingwang-intelandMa Mingfei 2026-09-07 09:10:35 +08:00
• 30d0eb2ca9 [NPU] Adapt DFlash2 speculative decoding to Ascend NPUs (#35629) faceless voidandgithub-actions[bot] 2026-09-07 09:08:39 +08:00
• 6252993afe chore: update CI test est_time values (#38238) sglang-botandsglang-bot 2026-09-06 17:49:41 -07:00
• 15aa2fb843 [ROCm] Take the fused DSA metadata kernels and drop redundant work from the absorb path (#37124) 5 people 2026-09-07 08:39:37 +08:00
• 30705c004c [Deepseek V4] Keep fp32 routing weights in the mxfp4 trtllm MoE (#33608) Brayden ZhongandBrayden Zhong 2026-09-06 20:12:55 -04:00
• 2e8c03e2c7 Fix inflated row pitch when a CP round-robin shard has a single row (#34142) AMD-yanfeiwang 2026-09-07 06:40:14 +08:00
• 2c05ed4e77 [ROCm] Stage large pageable H2D copies instead of pinning them in place (#37720) JohnQinAMD 2026-09-06 12:28:58 -07:00
• 31d28a2961 [NPU] Fix failed test cases in pr‑test‑npu and improve execution efficiency (#38112) 3 people 2026-09-07 01:48:27 +08:00
• 28457f0dca fix(gpt-oss): avoid duplicate MoE reduction with DP attention (#37199) s 2026-09-07 00:33:04 +08:00
• f3d05644db [diffusion] docs+skill: document which components to stream under layerwise offload (#35674) Mick 2026-09-06 23:12:36 +08:00
• ade1da017f [diffusion] docs: verify the DGX Spark H3 recipe (#37456) MickandClaude Fable 5 2026-09-06 23:10:27 +08:00
• be00a543a7 perf: use Gumbel-max trick in the main sampler to cut decode CPU dispatch (#38117) Xiaoyu Zhang 2026-09-06 22:40:27 +08:00
• 5bebe7a033 [Router] Add bucket-aware policy domains and native cache indexing (#38108) 3 people 2026-09-06 19:47:51 +08:00
• a176ba2f7b [diffusion] feat: measure warmup memory and layer usage per phase for residency calibration (1/4) (#37916) MickandClaude Fable 5.1 2026-09-06 19:42:42 +08:00
• 938dc5621d [diffusion] refactor: reuse plain state-dict loading without per-model classes (#38127) Mick 2026-09-06 18:39:21 +08:00
• 97c6978369 GLM-5.3-Flash support (#36507) +212 people 2026-09-06 17:27:59 +08:00
• a9944aec01 [diffusion] CI: guard the allocated vram peak with reporting the reserved one (#38172) Mick 2026-09-06 17:22:16 +08:00
• 8ef646a5c6 fix(vlm): contain EPD request lifecycle failures (#36944) Mickandmickqian 2026-09-06 16:05:10 +08:00
• d61378af77 docs(diffusion): add per-model tuning decision table to performance guide (#38148) Xiaoyu Zhang 2026-09-06 15:26:40 +08:00
• ae54ccb25d [Router] Publish cache-aware load state (#38139) Vincent Gao 2026-09-06 15:22:25 +08:00
• 6cee9285a3 [ROCm] Make DSA indexer top-k exact with cooperative selection (#37591) Zhang, Jiejing 2026-09-05 23:55:56 -07:00
• e3f7097591 [diffusion] refactor: consolidate plain state-dict component loaders (#38128) Mick 2026-09-06 13:52:54 +08:00
• 67e3ccda97 [diffusion] fix: restore non-layer placeholders before releasing host copies (#38171) Depend 2026-09-06 13:41:30 +08:00
• febb360519 [VLM] retire aborted disaggregated prefill results (#36988) Mick 2026-09-06 10:15:33 +08:00
• f5819b09bf Revert "[AMD][DSV4] Fix unified-KV pool sizing and SWA ring accounting" (#38163) Liangsheng Yin 2026-09-05 17:28:46 -07:00
• 09daea94ac Support NoPE layers in the tokenspeed_mla FP8 prefill hook (#38152) Mohammad Miadh AngkadandMohammad Angkad 2026-09-06 07:51:04 +08:00
• 514b45fd34 [AMD][DSV4] Fix unified-KV pool sizing and SWA ring accounting (#30315) yuttian1 2026-09-06 07:39:40 +08:00
• 6a0c55fd6c [CI] Pin the Rust TreeCore build to the resolved libtorch instead of interpreter discovery (#37696) Alex NailsandClaude Opus 5 2026-09-05 15:30:32 -07:00
• 77aee20259 [Model] Add support for Nanbeige4.2 (#32151) 3 people 2026-09-06 03:57:27 +08:00
• ccf9fe6590 [Kernel] Add KDA FP8 skinny GEMM for SM120 (#38082) Xiaoyu ZhangandWaterpine 2026-09-05 22:27:06 +08:00
• 9acbf75159 [CI][NPU] Fix pr-test-npu failing at Install dependencies with set: Illegal option -o pipefail (#38132) Cherry_ming 2026-09-05 22:26:47 +08:00
• dc2843801d perf(lfm2): fuse gating and short convolution on SM90 (#37622) Xiaoyu Zhang 2026-09-05 21:52:16 +08:00
• 1e6f18bfeb [MoE Refactor] Migrate SM100 trtllm-gen mxfp4 MoE onto MoeRunner (#32405) Beihao Zhou 2026-09-05 06:48:10 -07:00
• eda10c3678 [Diffusion] Enable breakable CUDA graph for JoyEcho (#38110) Xiaoyu Zhang 2026-09-05 21:45:53 +08:00
• 5df60a21cd fix(vlm): harden EPD receiver validation and liveness (#36945) Mick 2026-09-05 21:22:37 +08:00
• 4b802c052b test(npu): remove obsolete npu pr nightly cases, move accuracy cases to full (#37990) 3 people 2026-09-05 20:58:52 +08:00
• a18106bbc3 fix(vlm): make EPD cache publication transactional (#36949) Mick 2026-09-05 20:33:23 +08:00
• bd16c22a04 [diffusion] fuse LingBot MoE group-limited top-k index selection (#38044) 3 people 2026-09-05 18:12:30 +08:00
• 50c1bf0db0 [Diffusion] Port the Wan VAE decoder fast paths to the Qwen-Image VAE (#38020) DayuxiaoshuiandXiaoyu Zhang 2026-09-05 18:02:33 +08:00
• 0948e6ebed [CI] Remove metrics artifact mechanism from nightly NPU workflows (#35489) hhhh1252023 2026-09-05 17:16:52 +08:00
• d49180019b fix(moe): cast filtered-activation expert_ids to int32 for torch.compile (#38085) Xiaoyu ZhangandBBuf 2026-09-05 17:13:07 +08:00
• a74470e904 fix(mamba): unify causal_conv1d col* dtype to x (MiniCPM-V-4.6 GDN prefill bf16/fp16 mismatch) (#38039) Xiaoyu Zhang 2026-09-05 17:08:20 +08:00
• 0bdc15d20f [AMD] Fix ROCm VAE Conv2D fast path breaking spatial-parallel decode (#34424) jacky.cheng 2026-09-05 16:00:25 +08:00
• da76fa073f [diffusion] fix: fix host-resident vocab tables loaded on GPU (#38012) Xiaoyu Zhang 2026-09-05 14:03:21 +08:00
• 0ea8378085 [diffusion] feat: support request-scoped skip-softmax attention (#37959) Mick 2026-09-05 13:50:17 +08:00
• 4b44a1cde2 [Refactor] Let eviction policies take construction parameters (#37795) Shuwen WangandClaude Opus 5 2026-09-05 13:48:30 +08:00
• 32a1d55431 fix(modelopt_fp4): skip NVFP4 swiglu-fusion interleave for shared experts with swiglu_limit (#37378) Xinyuan Tong 2026-09-05 13:42:29 +08:00
• ae3205ba28 Fail fast on undersized swa pool (#37610) Ke Bao 2026-09-05 13:40:37 +08:00
• 09f542b23a [CI] Add /rerun-test --changed to rerun every test file a PR modifies (#37618) 3 people 2026-09-05 13:38:11 +08:00
• 756d0e0a85 [Model] Add K2 Horizon FP8 checkpoint support (#38033) Yash AkhauriandXiaoyu Zhang 2026-09-04 22:18:26 -07:00
• e980c1a2f1 fix(glm4v): disambiguate mixed image video offsets (#37971) 3 people 2026-09-05 13:16:16 +08:00
• 0454c074b4 [mem_cache] Clean up unified allocator leftovers (#38103) Liangsheng Yin 2026-09-04 22:02:50 -07:00
• 3a770da756 [Unified Tree] Support Branching-Point Caching for the SWA Component (#34565) 3 people 2026-09-05 12:29:16 +08:00
• 991368d880 Fix Muse Glimmer ModelOpt mixed weight mapping (#37510) Po-Han Huang (NVIDIA) 2026-09-05 11:53:55 +08:00
• ecd97de1fc [Router] Add load-aware prefill admission and bounded policy proposals (#37843) 4 people 2026-09-05 11:53:26 +08:00
• d50e9a9756 [Test] Prune redundant unified-memory allocator and pool tests (#38093) Liangsheng Yin 2026-09-04 20:47:01 -07:00
• 92a4d8b5ee Clean logging under --weight-loader-prefetch-checkpoints (#33930) 3 people 2026-09-04 23:05:53 -04:00
• 0645398a32 [mem_cache] Move the unified-memory allocators into allocator/ and split the composites out (#38072) Liangsheng Yin 2026-09-04 19:53:15 -07:00
• d6e0a8cbf4 [diffusion] add Helios per-token gated-residual fusion (quality-gated) (#38042) Xiaoyu ZhangandBBuf 2026-09-05 10:29:09 +08:00
• f1f2380d2b [Unified Cache][6/N]: Add UMBP external linker (#37578) Niko MaandZhangheng 2026-09-05 09:21:43 +08:00
• bc727bc4ee [FP8] SM120: route FP8 linear to per-tensor (cudnn/nvjet) instead of channelwise cutlass (#38006) Xiaoyu ZhangandBBuf 2026-09-05 09:07:13 +08:00