amazon-science/Glean

amazon-science/Glean

Language: Python

License: Apache-2.0

Stars: 9

Forks: 1

Open issues: 1

Created: 2024-10-11T22:22:48Z

Pushed: 2026-04-08T02:26:39Z

Default branch: main

Fork: no

Archived: no

README:

GLEAN: Generalized Category Discovery with Diverse and Quality-Enhanced LLM Feedback


This repository contains the implementation of the paper: > GLEAN: Generalized Category Discovery with Diverse and Quality-Enhanced LLM Feedback > [[Paper]](https://arxiv.org/abs/2502.18414)

> Henry Peng Zou, Siffi Singh, Yi Nian, Jianfeng He, Jason Cai, Saab Mansour, Hang Su

Setup

conda create -n glean python=3.9 -y
conda activate glean

# install pytorch
conda install pytorch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 pytorch-cuda=12.1 -c pytorch -c nvidia

# install dependency
pip install -r requirements.txt
pip install faiss-gpu==1.7.2 --no-cache-dir

To reproduce our paper results, make sure that you have the following package version installed: transformers==4.15.0, pytorch==2.1.0, 2.1.1 or 2.1.2, as we found that model performance may vary across different package versions, particularly with the transformers package.

Running

First, add you OpenAI API key in line 58 of the 'run.sh' file.

Pre-training, training and testing our model through the bash script:

sh run.sh

You can also add or change parameters in run.sh (More parameters are listed in init_parameter.py)

Bugs or Questions

If you have any questions related to the code or the project, feel free to email Henry Peng Zou ([pzou3@uic.edu](pzou3@uic.edu), [penzou@amazon.com](penzou@amazon.com)). If you encounter any problems when using the code, or want to report a bug, please also feel free to reach out to us. Please try to specify the problem with details so we can help you better and quicker!

Acknowledgement

This repo borrows some data and codes from Loop and JointMatch. We appreciate their great works!