-
Notifications
You must be signed in to change notification settings - Fork 1.1k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Enable MPS support for LitGPT #1724
Conversation
result = x.new_zeros(*x.shape[:-1], self.linear.out_features) # (64, 64, 384) | ||
return result.index_copy_(dim=-1, index=self.lora_ind, source=x) # (64, 64, 384) | ||
else: | ||
if all(self.enable_lora): |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That shouldn't be in the compatibility mode.
In fact, only index copy should be altered with slicing in the compatibility mode.
To my mind, the implementation should not wire a mps compatibility mode through the model but just check the tensor device in index copy. |
Agree.
M1 runner for public repos should be free of charge: actions/runner-images#9254 |
Yeah, we need to switch to M1 anyways because x86 macs don't get PyTorch support anymore (I think the last version was 2.2). |
Re-enables MPS support for LitGPT with an alternative
index_copy_
implementation.