Current Position:fig. beginning " AI Answers

How do I train a model using the Open R1 project?

2025-09-10

2.2 K

The following steps need to be followed to train a model using the Open R1 project:

Environment Configuration: First create a Python virtual environment and activate the
```
conda create -n openr1 python=3.11
conda activate openr1
```
Installation of dependencies: Install vLLM and project dependencies
```
pip install vllm==0.6.6.post1
pip install -e ".[dev]"
```
Account Login: Login to Hugging Face and Weights and Biases accounts
```
huggingface-cli login
wandb login
```

training model: Training using the provided scripts

GRPO Training:

python src/open_r1/grpo.py --dataset <dataset_path>

SFT Training:

python src/open_r1/sft.py --dataset <dataset_path>

Notably, the project supports multi-stage training, which can start with a base model and gradually transition to a reinforcement learning tuning model.

This answer comes from the articleOpen R1: Hugging Face Replicates the Training Process of DeepSeek-R1The

May not be reproduced without permission:AI productivity tools " How do I train a model using the Open R1 project?