It seems like the validation data might be getting included in the training process in your PyTorch model. That could explain why it's reaching such high accuracy so quickly. I recommend adding some console print statements to better track which data is being used at each step. Best regards!