BitsyGPT - An experimental decoder only GPT transformer.

1. Clone the repo
2. Create the python virtual env at the root directory
3. pip install torch and numpy
4. Train the model using scripts/train.sh. The existing configurations as in the config directoy is for 1.2M params and would take about 35mins on Apple M1 cpu.
5. Generate the outputs using scripts/generate.sh
6. Change the hyperparameters in the config directoty and play further