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.

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

position embedding is not saved #495

Closed
eghbalhosseini opened this issue Jan 27, 2022 · 3 comments
Closed

position embedding is not saved #495

eghbalhosseini opened this issue Jan 27, 2022 · 3 comments

Comments

@eghbalhosseini
Copy link

I am trying to recover position embedding for a trained model (rotary embedding) from the checkpoint folder, however its not in any of the pt files. how can I save position embedding?

@StellaAthena
Copy link
Member

StellaAthena commented Jan 27, 2022

Rotary positional embeddings are calculated in advance. They do not change during training and therefore there's no need to save them. To obtain the desired positional embeddings, you can simply calculate them using the formula.

@eghbalhosseini
Copy link
Author

thanks,
is there a tweak I can do in the code to save these during a checkpoint? or recreate them afterward. also is absolute positional embedding implemented in the pipeline ?

@StellaAthena
Copy link
Member

StellaAthena commented Feb 7, 2022

Yes absolute positional embeddings is an option. You can check out configs/neox_arguments.md details on all currently supported configurations and related information.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants