We use BEATs model to acquire tags and utilize LLM to expand into captions

1. Download pretrain BEATs weight from BEATs

2. BEATs model to classfier

accelerate config
accelerate launch --multi_gpu classfier.py
python generate_tag.py

3. LLM(such as GPT4 or deepseek) to expand into captions

python gpt/tag_caption.py
find /path -type f > output.txt

Download the mtg dataset. You can download mtg-jamendo-dataset and get raw_30s 55,701 tracks. https://huggingface.co/datasets/wanghappy/Music-tag-generation/

This project is licensed under the MIT License.