Enhancing Black-Box Embeddings with Model Augmented Fine-Tuning

A method to fine-tune the black-box OpenAI embedding to improve the retrieval performance 📈.

📝 See our paper Mafin: Enhancing Black-Box Embeddings with Model Augmented Fine-Tuning for a detailed introduction to this method.

Other training details and dataset construction can be found in LlamaIndex's Blog.

Install 🛠️

pip install -r requirements.txt

Fine-tune the BAAI/bge-small-en model

cd ./finetune
python train.py

Fine-tune the FAE model

cd ./finetune
python fae_train.py

The comparison of the evaluation results can be found in the jupyter notebook 📊.