Train test split (70/20/10(test)) & evaluation using Trainer

Hello - hoping you can help me out here: for uni i need to fine tune BERT german for text classification.

  1. would like to know how to split my training data into 70 training, 20 validation and 10 test with the dataloader

  2. would like some tips on how to use Trainer class with arguments to monitor the loss function for each of training, validation and test phases.

  3. what are most common ways to tune hyperparameters in BERT? is # epochs enough?

Any help much appreciated.
DB

1 Like