[CI] Reuse ColQwen3 models across pooling tests (#55889)

Andreas Karatzas committed about 1 month ago

[Core] Let SWA layers take the primary block size to avoid inflating the KV block LCM (#53007)

bnellnm committed about 1 month ago

[CI] Split long misc test groups by command (#52346)

Kevin H. Luu committed about 1 month ago

[Bugfix] Handle null RoPE parameters for NoPE layers (#55949)

Andreas Karatzas committed about 1 month ago

[2/N][KV Connector][NIXL] Support per-region transfer geometry (#53780)

Matthew Bonanni committed about 1 month ago

[Quant][Kernel] Remove GPTQ Group/Dynamic Activation Ordering (#54809)

Roderick Wu committed about 1 month ago

[ROCm] [Docker] Upgrade default AINIC repo to ship libionic 54.0-187-1 (#54112)

shikamd123 committed about 1 month ago

[Bugfix] speedup nvfp4 kv for FMHA (#55031)

sychen52 committed about 1 month ago

[Test] Dequantize NVFP4 KV cache scales in the layout the kernel writes (#55908)

Raya Elena Solano committed about 1 month ago

[Bugfix] Fix NVFP4 fused SiLU+mul scale allocation and global scale direction (#55643)

Jiangyun Zhu committed about 2 months ago

[Refactor] Remove utils dead code (#53941)

Wentao Ye committed about 2 months ago

[Core] Scope PCP-DP validation to GPU manager (#54523)

Qiu Chunshuo committed about 2 months ago

[Qwen3.8-Flash-Next] Tune FP8 TP2/TP4 Triton MoE on B200 (#55890)

Canlin Guo committed about 2 months ago

[LoRA] Clarify target module matching logic (#55865)

linitra24 committed about 2 months ago

[Kimi Bug] Fix kimi k3 startup cuda graph issue with recoverSSM (#55774)

Wentao Ye committed about 2 months ago

[4/N][warmup][DSv4] Migrate common attention kernels (#50176)

Roberto L. Castro committed about 2 months ago

[Bugfix] Fix Kimi K3 loading with interleaved weight streams (#53379)

Sherif Waly committed about 2 months ago

[Model] Enable torch.compile for Sarvam MLA (#55817)

Mohit Singla committed about 2 months ago

[Bugfix][Spec Decode] Cache the Mamba state at the block-grid position of EAGLE resume (#53945)

akshaver committed about 2 months ago

[Bugfix] Apply attention sinks in the Transformers backend (#52156)

Thomas Parnell committed about 2 months ago