Model | Paper

BertChunker is a text chunker based on BERT with a classifier head to predict the start token of chunks (for use in RAG, etc). It is finetuned based on sentence-transformers/all-MiniLM-L6-v2, and the whole training lasted for 10 minutes on a Nvidia P40 GPU on a 50 MB synthetized dataset. This repo includes codes for model defining, generating dataset, training and testing.

See generate_dataset.ipynb

Run

bash train.sh

See test.py

If this work is helpful, please kindly cite as:

@article{BertChunker,
  title={BertChunker: Efficient and Trained Chunking for Unstructured Documents}, 
  author={Yannan Luo},
  year={2024},
  url={https://github.com/jackfsuia/BertChunker}
}