def train(model,
    optimizer,
    criterion=nn.BCELoss(), ------ 영화 리뷰는 좋고 나쁨만 있으므로 BinaryCrossEntropy(BCELoss)를 사용
    num_epochs=5, ------ 에포크를 5로 사용
    eval_every=len(train_loader)//2,
    best_valid_loss=float("Inf")):

total_correct = 0.0
total_len = 0.0
running_loss = 0.0
valid_running_loss = 0.0
global_step = 0
train_loss_list = []
valid_loss_list = []
global_steps_list = []

model.train() ------ 모델 훈련
for epoch in range(num_epochs):
    for text, label in train_loader:
        optimizer.zero_grad()
        encoded_list = [tokenizer.encode(t, add_special_tokens=True) for t in text]
        padded_list = [e + [0] * (512-len(e)) for e in encoded_list] ------ 인코딩 결과에 제로패딩(zero-padding)을 적용
        sample = torch.tensor(padded_list)
        sample, label = sample.to(device), label.to(device)
        labels = torch.tensor(label)
        outputs = model(sample, labels=labels)
        loss, logits = outputs

        pred = torch.argmax(F.softmax(logits), dim=1) ------ ①
        correct = pred.eq(labels)
        total_correct += correct.sum().item()
        total_len += len(labels)
        running_loss += loss.item()
        loss.backward()
        optimizer.step()
        global_step += 1

        if global_step % eval_every == 0: ------ 모델 평가
            model.eval()
            with torch.no_grad():
                for text, label in valid_loader:
                    encoded_list = [tokenizer.encode(t, add_special_tokens=True) for t in text]
                    padded_list = [e + [0] * (512-len(e)) for e in encoded_list]
                    sample = torch.tensor(padded_list)
                    sample, label = sample.to(device), label.to(device)
                    labels = torch.tensor(label)
                    outputs = model(sample, labels=labels)
                    loss, logits = outputs
                    valid_running_loss += loss.item()

            average_train_loss = running_loss / eval_every
            average_valid_loss = valid_running_loss / len(valid_loader)
            train_loss_list.append(average_train_loss)
            valid_loss_list.append(average_valid_loss)
            global_steps_list.append(global_step)

            running_loss = 0.0
            valid_running_loss = 0.0
            model.train()

            print('Epoch [{}/{}], Step [{}/{}], Train Loss: {:.4f}, Valid Loss: {:.4f}'
.format(epoch+1, num_epochs, global_step, num_epochs*len(train_loader), average_train_loss, average_valid_loss))

            if best_valid_loss > average_valid_loss: ------ ②
                best_valid_loss = average_valid_loss
                save_checkpoint('../chap10/data/model.pt', model, best_valid_loss) ------ 오차가 작아지면 모델 저장
                save_metrics('../chap10/data/metrics.pt', train_loss_list, valid_loss_list, global_steps_list) ------ 평가에 사용된 훈련 오차, 검증 오차, 에포크(스텝)를 저장

    save_metrics('../chap10/data/metrics.pt', train_loss_list, valid_loss_list, global_steps_list) ------ 최종으로 사용된 훈련 오차, 검증 오차, 에포크(스텝)를 저장
    print('훈련 종료!')