LLM to generate Taylor Swift style songs

Assignment of Outsystems

The code to generate a Taylor Swift style songs, finetuned on the GPT2 transformer LLM.

Dataset

The dataset used can be found in this link

Setting up environment

Setup the conda environment using the environment.yaml file using the following command given below.

conda env create -f environment.yml
conda activate taylorllm

Generating lyrics

Clone the repository

git clone [email protected]:Shankhanil/TaylorLLM.git
cd TaylorLLM

Download the checkpoints file from this link. Place the .pt file in the checkpoints/ folder.

Edit the config.py file. Give the specifications of the source text file and the output text file in the SOURCE_PATH and DUMP_PATH respectively. Give the specifications of CHECKPOINT_PATH. Then run the following python command

python test.py

The output will be saved in the location specified in DUMP_PATH.

Running the finetuning script

To run the finetuning script as it was, just run the run.sh file using the following command

sh run.sh

However, if you want to specify model parameters, please look for the description of the arguments using

python main.py -h

Feedbacks

Feedbacks are hugely welcome. Please register bugs, issues, feedbacks and suggestions in the Issues section. Feedbacks are also welcome through email.

License

See LICENSE.md for License details