This repository is designed to fine tune existing GPT-2 models on lyric datasets.

To fine-tune a pre-trained GPT-2 medium model on lyrics, with all the lyrics in a file called lyrics.txt, use:

sh finetune.sh

To generate samples from the finetuned model, use:

sh generate.sh

You can spin up an inference server using:

uvicorn rest:app --reload --host 0.0.0.0