Skip to content
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.

Sign up for GitHub

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

Jump to bottom

[torchax] Initialize the model weights with the right distribution for each weight #58

Open
tengyifei opened this issue Jan 28, 2025 · 0 comments
Open

[torchax] Initialize the model weights with the right distribution for each weight #58

tengyifei opened this issue Jan 28, 2025 · 0 comments

Comments

@tengyifei
Copy link
Collaborator

E.g. bias probably should be zero initialized. Linear layers should be kaiming-initialized.

We need to do this in order to actually train the model to convergence.

We could use a dict of module: initializer_fn mapping until pytorch/xla#8635 is sorted.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@tengyifei