Quick Start

Follow these steps to get started quickly:

1. Clone the Repository
   Clone the repository to your local machine:

   git clone https://github.com/sunshine-JLU/deepseek-r1-distill-qwen-7B-lora.git
   
   cd deepseek-r1-distill-qwen-7B-lora
   
   

2. Enviroment

   pip install -r requirements.txt
   

3. Download the Model
   Set the Hugging Face endpoint and download the deepseek-r1-distill-llama-8b model:

   modelscope download --model deepseek-ai/DeepSeek-R1-Distill-Qwen-7B --local_dir ./DeepSeek-R1-Distill-Qwen-7B
   

4. Run the Notebook
   Open and run the DeepSeek-R1-Distill-Qwen-7B.ipynb notebook to start fine-tuning the model.

5. Run the lora-model in the lora_model_inference.ipynb
   After you successfully run over the DeepSeek-R1-Distill-Qwen-7B.ipynb, You will get a number of checkpoint files, each file is a lora weight that can be loaded independently, you can specify the lora file address in the lora_model_inference.ipynb to load and run.

Hardware requirement

GPU Memory at least 48GB would not appear OOM problem.

Test Enviroment

I deploy this program under PyTorch 2.3.0 ,Python 3.12(ubuntu22.04), Cuda 12.1