metadata

language:
  - en
license: cc
library_name: adapter-transformers
tags:
  - music
  - art
datasets:
  - SpartanCinder/artist-lyrics-dataset
  - SpartanCinder/song-lyrics-artist-classifier
metrics:
  - accuracy

GPT2 Pretrained Lyric Generation Model

This repository contains a pretrained GPT2 model fine-tuned for lyric generation. The model was trained using the Hugging Face's Transformers library.

Model Details

Model architecture: GPT2
Training data: The datasets were created using the Genius API and are linked in the Model's tags.
Training duration: [Mention how long the model was trained]

Usage

The model can be used to generate lyrics. It uses nucleus sampling with a probability threshold of 0.9 for generating the lyrics, which helps in generating more diverse and less repetitive text.

Here is a basic usage example:

from transformers import GPT2LMHeadModel, GPT2Tokenizer

tokenizer = GPT2Tokenizer.from_pretrained("SpartanCinder/GPT2-pretrained-lyric-generation")
model = GPT2LMHeadModel.from_pretrained("SpartanCinder/GPT2-pretrained-lyric-generation")

input_ids = tokenizer.encode("Once upon a time", return_tensors='pt')
output = model.generate(input_ids, max_length=100, num_return_sequences=5, do_sample=True, top_p=0.9)
print(tokenizer.decode(output[0], skip_special_tokens=True))