Open ghost opened 6 years ago
NNPACK has a fast implementation of strided convolution in nnp_convolution_inference wich is very good. But for training this feature is not yet implemented.
+100
any planning to implement the Winograd’s algorithm for training?
NNPACK has a fast implementation of strided convolution in nnp_convolution_inference wich is very good. But for training this feature is not yet implemented.