PyTorch F1 score monitoring and re-training

Question

I have a pre trained model that I load and test with testing data using PyTorch. Within the testing loop, I am calculating the f1 scores of each batch and monitor their f1 scores through a simple IF condition:

if f1_scores <= 0.5:
        anomaly_store_x.append(test_inputs)
        anomaly_store_y.append(test_label)

anomaly_store_x and anomaly_store_y stores the data and labels of the batches that have triggered the f1_score condition. After gathering them, I would like to retrain the data with the a singular data loader with both the anomaly lists and the original training data.

My questions is how can I combine both these different data into one DataLoader?

A simplified code of the testing loop can be seen below:

for i, data in enumerate(test_dl):           

    test_inputs, test_label = data
    
    if CUDA == True:
        test_inputs, test_label = torch.tensor(test_inputs).cuda(), torch.tensor(test_label).cuda()    

    ''' Feed input to model '''
    outputs = model_load(test_inputs)     
    test_label = test_label.long()    
    _, y_pred_tags = torch.max(outputs, dim = 1)                   
                               
    ''' F1 Scores '''                    
    f1_scores = f1_score(y_test_label_flat2, y_test_pred_list_flat2, average=None)        
    
    ''' Anomaly Detection '''
    if f1_scores >= 0.5:
        anomaly_store_x.append(test_inputs)
        anomaly_store_x.append(test_label)