Horace He
Horace He
@ezyang i don’t think this is dynamic shapes
@alexandrudima Do you guys plan on fixing this issue at some point? Or is the workaround satisfactory as a long term solution?
Is it tricky or is it feasible as a first time contribution?
Could you remove the `model_name` modification? Nothing against it in particular, but since this repo doesn't have any CI set up and I'm currently unable to test these repos -...
It needs to be PyTorch nightly for now.
@rayleizhu To make `torch.compile` work with 3rd party ops, you need to register it. I'll put up an example of how to do this later. > This is unsurprising as...
Are you using pytorch nightly? This perf seems much worse than I would expect
cc: @drisspg
@drisspg https://github.com/pytorch-labs/gpt-fast/issues/46#issuecomment-1857615288 @merveermann says they're using the nightly I believe.
This is currently an issue we're aware of, unfortunately. In theory, it's possible to use AOTInductor https://www.youtube.com/watch?v=w7d4oWzwZ0c to completely AOT compile everything, however it's somewhat finicky to use. We also...