Skip to content

About

No description, website, or topics provided.

Resources

Stars

2 stars

Watchers

1 watching

Forks

Latest commit

 

History

3 Commits

Folders and files

Repository files navigation

CoordToken

This project implements coordinate tokenization using Finite Scalar Quantization (FSQ) for molecular data.

FSQ Implementation

The FSQ is currently implemented in the iFSQ version by default. If you want to switch to the standard FSQ:

  1. In fsq.py, comment out lines 126 and 127.
  2. Uncomment line 128.

This changes the activation from torch.sigmoid to torch.tanh in the symmetry-preserving bound.

Installation

  1. Install dependencies:

    pip install -r requirements.txt
    
  2. (Optional) Set up a conda environment or adjust paths as needed.

Usage

  • Training: Adjust and run submit_train.sh or python submit_train.py.
  • Evaluation: Adjust and run python submit_eval.py --ckpt <checkpoint_path>.

Downloads

Authenticate with Hugging Face:

hf auth login

Download the tokenizer checkpoint into the repo root:

hf download FilyaGeikyan/CoordToken tokenizer.ckpt --repo-type model --local-dir .

Download the training dataset:

hf download FilyaGeikyan/CoordToken-data merged_train.csv.zst --repo-type dataset --local-dir .

Run evaluation with the downloaded checkpoint:

python submit_eval.py --ckpt tokenizer.ckpt

Notes

  • Adjust file paths (e.g., data directories, logs) for your environment.
  • Ensure CUDA/GPU availability for training.

About

No description, website, or topics provided.

Resources

Stars

2 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages