This is a tutorial repository to show how easy it is to build a custom GPT model using Maximal and TensorFlow 2.x libraries.
For a more in depth documentation on how to build Transformer neural networks with Maximal and TensorFlow, refer to the Official documentation (Tutorials section).
You must train a model via: python train.py
. All hyperparameters can be tweaked in config.py
.
Once a model is saved and stored in /saved_models
folder, you can run: python chat.py
to load the model and generate text via command line.
In the first public version of this repository, prompt length must be at least as long as the model's input size (controllable in config.py
).
This is a constraint that will be removed in further versions.
To end the generation either use Ctr+C
or type exit
.