[Question]: Difference Between train and fine_tune Methods in ModelTrainer

[AAnirudh07]

Question

I noticed the documentation on the difference between the ModelTrainer's train() and fine_tune() methods is empty. Based on this example, it seems like train() is for FLAIR embeddings, and fine_tune() is for Transformer embeddings. Is this the only distinction, or are there other differences? Any insights would be much appreciated!

[helpmefindaname]

Hi @AAnirudh07
exactly, the only difference is the set of default parameters. Train is usually used for a feature extraction method, where you have a frozen weight and some LSTM layers afterwards while fine_tune is used for fine_tuning the whole embeddings, (e.g. using transformers with finetune=True

[AAnirudh07]

Gotcha, thank you! I'll go ahead and close this issue now.