from datasets import load_metric
import numpy as np
 
# define metrics and metrics function
accuracy_metric = load_metric( "accuracy")
 
def compute_metrics(eval_pred):
    predictions, labels = eval_pred
    predictions = np.argmax(predictions, axis=1)
    acc = accuracy_metric.compute(predictions=predictions, references=labels)
    return {
        "accuracy": acc["accuracy"],
    }