is_correct = []
predicted_class = []
original_class = []

for i in tqdm(range(0, test_pos.shape[0], batch_size)):

    if i+batch_size>=test_pos.shape[0]:
            break

    #getting batch
    test_pos_batch = test_pos[i:i+batch_size]
    test_tok_batch = test_tok[i:i+batch_size]
    test_targ_batch = test_targ[i:i+batch_size]

    #making prediction, not masking anything
    clsf_logits, _ = model(test_tok_batch, test_pos_batch, torch.zeros(test_pos_batch.shape))

    #converting logits to probabilities then rounding to classifications
    res = torch.sigmoid(clsf_logits).round().squeeze()

    #keeping track of the original class (positive or negative) and if the model was correct
    original_class.extend(np.array(test_targ_batch.to('cpu')))
    is_correct.extend(np.array((res == test_targ_batch).to('cpu')))
    predicted_class.extend(np.array(res.detach().to('cpu')))

#accuracy
sum(list(is_correct))/len(is_correct)