[Perf] Integrate FlashInfer KDA kernels (#55364)

Wei Zhao committed about 2 months ago

[Kernel][Perf] Tune H20 block-FP8 MoE low-batch configs (+21%) (#54668)

leo committed about 2 months ago

[Kimi Bug] Fix kimi k3 AssertionError assert 0 <= checkpoint_idx < len(blocks) (#55747)

Wentao Ye committed about 2 months ago

[ROCm][CI] Enable HY-V4 model initialization on ROCm (#54405)

Andreas Karatzas committed about 2 months ago

[Bugfix][Quantization][MoE] Normalise an unset group_size on the compressed-tensors WNA16 MoE path (#53163)

Artur Fierka committed about 2 months ago

[Build] Remove obsolete TPU Dockerfile (#55376)

Woosuk Kwon committed about 2 months ago

[Feature] Support EAGLE3 for Sarvam (#53052)

Mohit Singla committed about 2 months ago

[Pooling] Honor request_id from request bodies (#55665)

Taneem Ibrahim committed about 2 months ago

[Bugfix] Restore Responses validation error boundary (#55761)

Taneem Ibrahim committed about 2 months ago

[CI] Recover empty multi-node Docker networks and finish partial cleanup (#55454)

Kevin H. Luu committed about 2 months ago

[CI] Synchronize shared offload unlink test before observing pathname (#55604)

Kevin H. Luu committed about 2 months ago

[CI/Build] Unskip ColQwen3 multimodal pooling tests on Transformers v5 (#55588)

Sahil Patel committed about 2 months ago

[CI] reduce npu CI use time and add timeout (#55731)

Zeyu Yang committed about 2 months ago

[Model] Add Cohere Compass model (#54774)

am-cohere committed about 2 months ago

[Tests] Update SarvamMLA transformers v5 compatibility reason to hf (#55728)

Abhi committed about 2 months ago

[Docs] Clarify admission control limits apply server-wide, not per DP rank (#55124)

Nicolò Lucchesi committed about 2 months ago

[Bugfix][Gemma] Conditionally create KV projections/norms on KV-shared layers (#54917)

Asaf Gardin committed about 2 months ago

[3/N][warmup][DSv4] Migrate FA4 MLA and shared CuTeDSL kernels (#53565)

Roberto L. Castro committed about 2 months ago

[2/N][warmup][DSv4] Migrate sequence and DCP kernels (#53564)

Roberto L. Castro committed about 2 months ago

[Pooling] Honor max_embed_len for chunked embeddings (#55551)

Taneem Ibrahim committed about 2 months ago