sonofhypnos / epsilon-transformers

epsilon machines and transformers!

Epsilon Transformers

This project contains Python code for generating process data and training transformer models on it. The codebase is organized into several Python scripts and Jupyter notebooks.

Codebase Structure

Folders

• epsilon_transformers: source code for this repository.

Code

• run_sweeps.py: run sweeps on wandb

Usage

To install, alongside all dependencies, run pip install -e . from the repository folder

Persistence

This codebase provides functionality to persist models & other research artifact's to the S3 cloud provider. In order to use it, make sure that you have the AWS_ACCESS_KEY_ID AWS_SECRET_ACCESS_KEY environment variables set in a .env file.

If you are a PIBBSS affiliate, or otherwise feel like you should have access to PIBBSS' S3 instance but currently don't, reach out to Lucas.

Dev

For formatting, type checking, and testing you can run the corresponding script in scripts/