Can you provide the training script for the large_44k_v2 model, or what modifications need to be made to the current script in order to train the large_44k_v2 model?