Add support for torch.set_default_device when initializing model parameters

[fshp971]

What does this PR do?

Replace torch.Tensor with torch.empty when initializing model parameters.

The reason is that torch.empty is compatible with torch.set_default_device (but torch.Tensor is not), thus enable one to efficiently initialize a large ML model with specified GPUs.

(An example is that initialize a Llama2 model with the default torch.nn.init.xavier_normal_ method is extremely slow on CPUs, but very fast on GPUs.)

Before submitting

• Did you have fun?
  • Make sure you had fun coding 🙃
• Did you read the contributor guideline?
• Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
  • N/A
• Did you make sure to update the docs?
  • N/A
• Did you write any new necessary tests?
  • N/A
• Did you update the changelog? (if needed)
  • N/A

PR review

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

[mhjacobson]

I just hit this myself, and while I am not an expert in this particular project, I think it would be a good idea.