[Perf] Avoid more h2d copies from non-pinned tensors (#54660)

Nick Hill committed about 2 months ago

[Kimi-K3] Overlap low-M TP8 KDA projections (#54697)

Yongye Zhu committed about 2 months ago

[Bugfix][KV Offloading] Fix eager SimpleCPUOffload cache registration and final flush (#53532)

maithilijoshi20 committed about 2 months ago

[Bugfix] Fix ColQwen3.5 pooler projector initialization (#54847)

Divakar Verma committed about 2 months ago

[Bugfix] Handle padded routes in CUTLASS MoE permutations (#54747)

Kevin H. Luu committed about 2 months ago

[Qwen3.8-Flash-Next] Separate prefill and decode paths for QSA indexer (#54513)

Thien Tran committed about 2 months ago

[CI/Build] Gate PR title check on ready PRs & use slim runners (#54827)

Bugen Zhao committed about 2 months ago

[ROCm][CI] Handle tied experts in softplus sqrt top-k test (#52679)

Andreas Karatzas committed about 2 months ago

[CI] Shard long kernel test groups (#54754)

Kevin H. Luu committed about 2 months ago

[CI] Shard distributed model jobs above the 24h P90 threshold (#54752)

Kevin H. Luu committed about 2 months ago

[CI] Split nightly MTP acceptance tests (#52353)

Kevin H. Luu committed about 2 months ago

[Rust Frontend] Attribute decoded text to tokens (#52910)

Bugen Zhao committed about 2 months ago

[Kimi-K3][Perf] Prefetch ll_bf16 router weights for M=1 (#53524)

mingg26 committed about 2 months ago

[CI] Remove MRV2-specific tests (#54823)

Nick Hill committed about 2 months ago

[Kimi Bug] Fix `cannot access local variable 'active_non_spec_mask_cpu'` (#54781)

Wentao Ye committed about 2 months ago

[Bugfix] Restore `weight_dtype` in `QuarkW8A8Fp8MoEMethod` to fix GPT-OSS FP8 MoE weight loading (#54824)

Micah Williamson committed about 2 months ago

[Bugfix][KV Connector] Fix DecodeBench DCP block selection (#54679)

Jeff (Junze) Ma committed about 2 months ago

[CI] Read the CRCR report token from a Buildkite secret (#54605)

Andrey Talman committed about 2 months ago

[ModelOpt] Redesign the LinearMethod classes using the generic QuantKey-driven method (#49381)

Juhi Mittal committed about 2 months ago

[CI] Disable CUDA graphs for GLM PCP evals (#54745)

Kevin H. Luu committed about 2 months ago