Hi, our team works on DL frameworks performance optimization on CPU.
PyTorch CPU performance can be significantly improved with MKL-DNN, an open source deep neural network library on Intel CPU platform. Currently MKL-DNN supports Conv2d, Relu, BatchNorm, Pooling, Concat, etc. The CPU performance improvement will benefit cloud users as well as HPC researchers who have access to Xeon clusters.
Will PyTorch be interested in the part?

cc @gujinghui @PenghuiCheng @XiaobingSuper @jianyuh @VitalyFedyunin