Check out the multiple_experiements.sh to test out experiments.
Example: python main.py --cfg configs/GPS/peptides-func-GPS.yaml --repeat 1 seed 42 wandb.use True reg '0.0' loss L1
Two parameters:
- reg: float -- for the hyperparameter (eta) of your loss function. Negative Values for reg means allow full backpropagation for our regularization term. Otherwise, the regularization gradient only flows to query and key matrix transformation (e.g. -0.1).
- loss: {"L1", "CE"} -- means which loss function you want to run for the regularization term.