onnxruntime
8d379964 - Split DeepSeek V4 attention into contrib ops

Commit

86 days ago

Split DeepSeek V4 attention into contrib ops

References

#31570 - Add DeepSeek V4 GQA trailing RoPE/de-rotation and MoE/QMoE router_weights support

Author

Copilot

Copilot

Parents

Loading