关于模型的微调。

[A-cloud-bit]

博主你好，我想问一下，在这里从验证集取数据，对模型进行微调，但是我们并没有更新元模型权重。这里的目的是什么呢？只是说单纯的验证一下模型的性能吗？ for support_images, support_labels, query_images, query_labels in val_loader:

        # Get variables
        support_images = support_images.float().to(dev)
        support_labels = support_labels.long().to(dev)
        query_images = query_images.float().to(dev)
        query_labels = query_labels.long().to(dev)

        loss, acc = maml_train(model, support_images, support_labels, query_images, query_labels,
                               3, args, optimizer, is_train=False)

        # Must use .item()  to add total loss, or will occur GPU memory leak.
        # Because dynamic graph is created during forward, collect in backward.
        val_loss.append(loss.item())
        val_acc.append(acc)