test_BERT.py

from sklearn.metrics import confusion_matrix

true=[]
predictions=[]
with torch.no_grad():
    model.eval()
    for data in testloader:
        tokens_tensors, segments_tensors,masks_tensors, labels = [t.to(device) for t in data]
        val_outputs = model(input_ids=tokens_tensors, 
                    token_type_ids=segments_tensors, 
                    attention_mask=masks_tensors, 
                    labels=labels)

        logits = val_outputs[1]
        _, pred = torch.max(logits.data, 1)
        true.extend(labels.cpu().tolist())
        predictions.extend(pred.cpu().tolist())


c = confusion_matrix(true, predictions)
print(c)
accuracy_score(predictions,true)