Method validate_epoch

funasr/train_utils/trainer_ds.py:754–846

Defines the validation process for a single epoch. Should be implemented with the actual model validation steps. Args: epoch (int): The current epoch number.

(
        self,
        model=None,
        dataloader_val=None,
        epoch=None,
        writer=None,
        **kwargs,
    )

                optim.zero_grad(set_to_none=True)

    def validate_epoch(
        self,
        model=None,
        dataloader_val=None,
        epoch=None,
        writer=None,
        **kwargs,
    ):
        """
        Defines the validation process for a single epoch.
        Should be implemented with the actual model validation steps.

        Args:
            epoch (int): The current epoch number.
        """
        self.val_loss_avg = 0.0
        self.val_acc_avg  = 0.0

        if self.use_ddp or self.use_fsdp or self.use_deepspeed:
            dist.barrier()
        logging.info(f"Validate epoch: {epoch}, rank: {self.rank}\n")
        model.eval()

        with torch.no_grad():

            speed_stats = {}
            time_beg = time.perf_counter()
            time5 = time_beg

            dataloader_val.batch_sampler.set_epoch(epoch)
            for batch_idx, batch in enumerate(dataloader_val):

                loss_dict = {
                    "speed_stats": {},
                    "epoch": epoch,
                    "batch_idx": batch_idx,
                    "data_split_i": kwargs.get("data_split_i", 0),
                    "data_split_num": kwargs.get("data_split_num", 1),
                    "log_step": batch_idx + kwargs.get("start_step", 0),
                    "batch_total": self.batch_total,
                    "step_in_epoch": batch_idx + 1,
                    "lr": 0.0,
                }

                time1 = time.perf_counter()
                loss_dict["speed_stats"]["data_load"] = f"{time1 - time_beg:0.3f}"

                batch = to_device(batch, self.device, non_blocking=True)

                time2 = time.perf_counter()

                self.forward_step(model, batch, loss_dict=loss_dict)

                time3 = time.perf_counter()
                loss_dict["speed_stats"]["forward_time"] = f"{time3 - time2:0.3f}"

                total_time = f"{(time.perf_counter() - time5):0.3f}"
                time5 = time.perf_counter()