I have a question about Fine-tuning.
Can I train model only update dataset in Fine-tuning process not include pre-training(before data)?
As fine-tuning a model we are transfering the generalized knowledge acquired during pre-training process to a specified, narrow (downstream) tasks to improve performance on the task.
But we have to also be careful about the pre-training corpus. Did it had enough information in the corpus that we really needed(usefulness)? The selection of appropriate models is also crucial,
- model size
- model vocabulary
- model performance.