Function single_gpu_test

modelscope/trainers/utils/inference.py:18–77

Test model in EpochBasedTrainer with a single gpu. Args: trainer (modelscope.trainers.EpochBasedTrainer): Trainer to be tested. data_loader (nn.Dataloader): Pytorch data loader. device (str | torch.device): The target device for the data. metric_classes (List): L

(trainer,
                    data_loader,
                    device,
                    metric_classes=None,
                    vis_closure=None,
                    data_loader_iters=None)

def single_gpu_test(trainer,
                    data_loader,
                    device,
                    metric_classes=None,
                    vis_closure=None,
                    data_loader_iters=None):
    """Test model in EpochBasedTrainer with a single gpu.

    Args:
        trainer (modelscope.trainers.EpochBasedTrainer): Trainer to be tested.
        data_loader (nn.Dataloader): Pytorch data loader.
        device (str | torch.device): The target device for the data.
        metric_classes (List): List of Metric class that uses to collect metrics.
        vis_closure (Callable): Collect data for TensorboardHook.
        data_loader_iters (int): Used when dataset has no attribute __len__ or only load part of dataset.

    Returns:
        list: The prediction results.
    """
    dataset = data_loader.dataset
    progress_with_iters = False
    if data_loader_iters is None:
        try:
            data_len = len(dataset)
        except Exception as e:
            logging.error(e)
            raise ValueError(
                'Please implement ``__len__`` method for your dataset, or provide ``data_loader_iters``'
            )
        desc = 'Total test samples'
    else:
        progress_with_iters = True
        data_len = data_loader_iters
        desc = 'Test iterations'

    with tqdm(total=data_len, desc=desc) as pbar:
        for i, data in enumerate(data_loader):
            data = to_device(data, device)
            evaluate_batch(trainer, data, metric_classes, vis_closure)

            if progress_with_iters:
                batch_size = 1  # iteration count
            else:
                if isinstance(data, Mapping):
                    if 'nsentences' in data:
                        batch_size = data['nsentences']
                    else:
                        try:
                            batch_size = len(next(iter(data.values())))
                        except Exception:
                            batch_size = data_loader.batch_size
                else:
                    batch_size = len(data)
            for _ in range(batch_size):
                pbar.update()

            if progress_with_iters and (i + 1) >= data_len:
                break