This project demonstrates the process of fine-tuning the Mistral 7B language model using the UltraChat dataset. It covers the entire pipeline from downloading the model to inference with the fine-tuned model.
The project uses the UltraChat dataset from Hugging Face:
- Dataset: HuggingFaceH4/ultrachat_200k
- Specific file used:
test_gen-00000-of-00001-3d4cd8309148a71f.parquet
-
Download the Model
- Clone the Mistral finetune repository
- Install requirements
- Download the Mistral 7B model
-
Prepare the Dataset
- Download and split the UltraChat dataset
- Reformat the data for training
-
Configure Training
- Set up the training configuration in a YAML file
-
Start Training
- Run the training script
-
Inference
- Load the fine-tuned model and run inference
For detailed steps and code, please refer to the Jupyter notebook in this repository.