Function evaluate

DeepSpeedExample/tasks/zeroshot_gpt2/evaluate.py:109–131

Evaluation.

(data_loader, model, eval_metric)

def evaluate(data_loader, model, eval_metric):
    """Evaluation."""
    args = get_args()

    # Turn on evaluation mode which disables dropout.
    model.eval()

    total_output = 0.0
    with torch.no_grad():
        # For all the batches in the dataset.
        for iteration, batch in enumerate(data_loader):
            if iteration % args.log_interval == 0:
                print_rank_0('> working on iteration: {}'.format(iteration))
            # Forward evaluation.
            output = forward_step(batch, model, eval_metric)

            # Reduce across processes.
            torch.distributed.all_reduce(output,
                                         group=mpu.get_data_parallel_group())

            total_output += output

    return total_output


def evaluate_and_print_results(task, data_loader, model, eval_metric):