trainer.py

import torch

from util import Config





class Trainer:
    def __init__(self, config: Config):
        self.__dict__ = dict(config.__dict__)
        

    def log(self, loss: float):
        print(f"Epoch: {self.epoch} / {self.num_epochs}\t\tBatch: {self.batch} / {self.num_batches}\t\tLoss: {round(loss, 4)}")
        
        args = {'epoch': self.epoch, 'batch': self.batch, 'loss': loss}
        self.wandb(args)

        if self.inference.frequency != 0:
            if self.batch % self.inference.frequency == 0:
                print(f'{self.model.generate_text(self.inference.seed_text, self.inference.n_predict)}')


    def train(self, batches):
        self.optimizer = torch.optim.Adam(self.model.parameters(), lr=self.learning_rate)

        self.model.unfreeze()

        for self.epoch in range(self.num_epochs):
            for self.batch in range(self.num_batches):
                ids = batches[self.batch]

                loss = self.model.compute_loss(ids)

                self.optimizer.zero_grad()
                loss.backward()
                self.optimizer.step()

                self.log(loss.item())