No update of bias in linear Regression via Pytorch

Question

I want to use PyTorch to build linear regression to solve a simple task. The 1-D output is just the average of 10-D input. What is strange is that during training, weights would