Kuraao
commented
3 years ago Wandb provides some magic for integration with Pytorch, however using it with our code is not possible as is, and requires several modifications. So for the moment it will be used by using a logging method. To be added:
- a save method to save model parameters
- find some way to save the gradient without modifying the code of TrainableDecisionCatModel
see wandb.ai