sxjscience commented on a change in pull request #16408: Add MXNet Ops for fast multihead attention URL: https://github.com/apache/incubator-mxnet/pull/16408#discussion_r333163192
########## File path: src/operator/contrib/transformer.cu ########## @@ -22,12 +22,898 @@ * \file transformer.cu * \brief GPU implementation of the operators used in Transformer */ + +#include <cuda.h> +#include <cuda_runtime.h> +#include <cuda_fp16.h> +#include <cuda_profiler_api.h> + #include <mxnet/base.h> #include "./transformer-inl.h" +#include "../../common/cuda_utils.h" + +#include "cutlass/cutlass.h" Review comment: I guess we need to add cutlass as a third-party submodule like cub. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
