Skip to content

Code and Data Repo for the CoNLL Paper -- Future Lens: Anticipating Subsequent Tokens from a Single Hidden State

License

Notifications You must be signed in to change notification settings

KoyenaPal/future-lens

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

42 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Future-Lens

Code and Data Repo for the CoNLL 2023 Paper -- Future Lens: Anticipating Subsequent Tokens from a Single Hidden State

alt text

In this repo, we include two distinctive way to reveal the extent to which individual hidden states may directly encode subsequent tokens: 1) Linear Model Approximation and 2) Causal Intervention Methods. With the best result, we propose Future Lens as the tool to extract information about future (beyond subsequent) tokens from a single hidden token representation.

Set-up

Run the following code to install relevant packages to your virtual environment

pip install -r scripts/colab-reqs/future-env.txt

Linear Model Approximation

To run the linear modoel approximation, you may run the following command:

python linear_methods/linear_hs.py

Causal Method

To train a soft prompt for GPT-J, you may run the following command:

python causal_methods/train.py

And then test by running the following script:

python causal_methods/test.py

Note: To load the default training set, you can access the file with this link.

Future Lens

We provide an online demo notebook for the Future Lens. You may also run the code locally from

demo/FutureLensDemonstration.ipynb

Updates

Oct.6th 2023

Our paper is accepted by CoNLL 2023!

Nov.8th 2023

We release the code and preprint version of the paper!

Contact

If any questions about the code, please contact Koyena at [email protected].

About

Code and Data Repo for the CoNLL Paper -- Future Lens: Anticipating Subsequent Tokens from a Single Hidden State

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published