Haoyang Peng

i'll try it out, thanks

well, i passed the compilation but when running it meets cutass internal error 309. I've changed the epilogue to the below: ``` using EpilogueOutputOp = cutlass::epilogue::thread::LinearCombinationBiasElementwise< float, float, float, float,...

may be input size and kernel size not right? My kernel size is (512,3,3,512) input size is (1,7,7,512). The error happend in the line that initialize the op: ``` status...

Is that two tensors necessary? I set them to nullptr and it still fails in initial the arguments.

In fact, my purpose is to simply add a epilogue with linear combination relu operation after Implicit GEMM AX+B, i wonder is there a more simple way?

i dont know if i understand it correctly, should i add this conversion object to the LinearCombinationBiasRelu epilogues?

thank you, i passed the running test. Now I'm working on the nchw_to_nhwc kernels. I found out that the Tensor4D's n,h,w,c dimension are designed for NHWC format which means the...

thanks, i'll try it.

Hello, using jt.flags.use_cuda=1 will put all arrays after this declaration to GPU.

yes