This work was done during one weekend by research workshop participants and does not represent the work of Apart Research.
ApartSprints
ARENA HACK
Accepted at the 
ARENA HACK
 research sprint on 
January 22, 2024

AttentionData

Note: The notebook to PDF conversion caused some issues with the cell outputs, but it is still viewable in the demo notebook: https://github.com/connor-henderson/attention-data/blob/main/demo.ipynb. Visualizing and generating data on attention patterns can be beneficial for understanding and interpreting the model's behavior. Here I've written a class with some methods for generating token and sequence-level statistics on attention patterns, viewing these stats, and passing them to OpenAI’s GPTs for interpretation. The core AttentionData class can be used with any arbitrary combination of text batch, HookedTransformer instance, and OpenAI GPT model.

By 
Connor Henderson
🏆 
4th place
3rd place
2nd place
1st place
 by peer review
Thank you! Your submission is under review.
Oops! Something went wrong while submitting the form.

This project is private