Skip to content

TaraRK/Internalize_CoT_Step_by_Step

 
 

Repository files navigation

Internalize CoT Step by Step

This is the fork of the original repository. The code structure is as follows:

  • probing experiment: our probing exeriments
  • plotting: any code used for visualizations and plots produces including the pca visualizations and the t-SNE visualizations
  • src: source code for generation, we attached hooks to cache activations in generate.py and utils.py, we also use train.py to train the model.
  • scripts: examples of scripts we use to run the code
  • data and models are directories from the original repository we use to maintain the datasets and model checkpoints.

Releases

No releases published

Packages

 
 
 

Contributors

Languages

  • Python 98.7%
  • Shell 1.3%