Hello, Can you give any clarification on how or if these kernels can be used with Automatic Mixed Precision in torch?