As stated, the maximum sequence length for the model is 512. Based on my experience, 512 tokens usually contain enough information to be able to classify with good accuracy, so it won’t affect the training too much :)