I already train my model locally and i want to commit it, but got some struggles, where I should add a training code. The training process should be in the “compute_action” function or it is used for inference and training should be performed in “init”?
Or maybe I should somehow transfer model weights?
Hey, you should train the model yourself and in the __init__
you are supposed to load your model weights. compute_action
then does the inference.
If you use the script in utils/submit.sh
then it automatically upload the model weights for you, you just need to provide the correct local path.
2 Likes