Policy on upstreaming kernels to PyTorch

[vadimkantorov]

Hi!

I'm curious on policy of upstreaming kernels to core PyTorch: are there plans to submit some of kernels up to PyTorch?

The reason I'm asking is that many FasterTransformer kernels never made to PyTorch core and even if some of them are still relevant, they are not very accessible to wider PyTorch audience as not committed upstream

If a kernel is contributed upstream, it naturally becomes the new baseline

Thanks!