Hi I was reading the paper for wav2vec and they mention that for the first 10k updates only the output classifier is trained, after which the Transformer is also updated, how I can add this variable on the huggingface Trainer to follow the paper suggestion.