You can find the unit theory here.
First check that you have the requirements listed here.
IMPORTANT: Before runing the following commands ensure that you are inside the unit folder.
You only need to run this command once
docker-compose build
In this unit the train script also pushes the model to the hub after training. Use the following command to run it:
docker-compose run train
If you want to modify some hyperparameters you can do it in the main function of the train.py script. You MUST change the hf_username variable to use your hugging face account. It is also recomended to check that the n_workers variable is equal or lower to the number of cores of your CPU.
The results of the experiments are stored in the runs folder, which is automatically created.