https://github.com/opencsgs/opencsg-r1
https://github.com/opencsgs/opencsg-r1
Last synced: 10 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/opencsgs/opencsg-r1
- Owner: OpenCSGs
- Created: 2025-02-16T08:39:40.000Z (about 1 year ago)
- Default Branch: main
- Last Pushed: 2025-03-03T02:24:29.000Z (about 1 year ago)
- Last Synced: 2025-07-04T22:38:40.677Z (10 months ago)
- Language: Python
- Size: 17.6 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# OPENCSG R1
**Table of Contents**
1. [Overview](#overview)
2. [Installation](#installation)
3. [Training models](#training-models)
## Overview
The goal of this repo is to build the different dataset or methods for trainning r1-like models.The project is simple by design and mostly consists of:
- `src`: contains the scripts to train and evaluate models on different datasets and trainning methods:
- `full_train_grpo.py`: trains a model with GRPO by using full-parameters training.
- `lora_train_grpo`: performs a simple SFT of a model on a dataset.
- `scripts`: contains easy-to-run commands for each step in the R1 pipeline leveraging the scripts above.
- `inference`: contains some code for model tests.
## Installation
```shell
pip install -r requirements.txt
```
## Training models
```shell
# full parameter trainning method with grpo
bash scripts/full_train_grpo.sh
# lora trainning method with grpo
bash scripts/lora_train_grpo.sh
```