Unable to save checkpoints, when use manual optimization

Automatic opt SGD version works well, but manual opt version does not save any checkpoint even save_last or save_top_1

Manual optimization version differs from automatic optimization SGD version:

    def training_step(
        self, batch: Tuple[torch.Tensor, torch.Tensor], batch_idx: int
    ) -> torch.Tensor:
        """Perform a single training step on a batch of data from the training set.

        :param batch: A batch of data (a tuple) containing the input tensor of images and target
        :param batch_idx: The index of the current batch.
        :return: A tensor of losses between model predictions and targets.
        optimizer = self.optimizers()

        # first forward-backward pass
        loss_1, preds_1, targets_1 = self.model_step(batch)

        # update and log metrics
        self.train_acc(preds_1, targets_1)
        self.log("train/loss", self.train_loss, on_step=False, on_epoch=True, prog_bar=True)
        self.log("train/acc", self.train_acc, on_step=False, on_epoch=True, prog_bar=True)
        self.log("learning rate", self.get_lr(optimizer), on_step=True, on_epoch=False, prog_bar=False)

        # second forward-backward pass
        loss_2, preds_2, targets_2 = self.model_step(batch)

        # return loss or backpropagation will fail
        return loss_1