What is the main purpose or outcome of this training? is it only for text completion? How do I finetune it for Question-Answering or any other downstream task? I want to use my dataset in the training. How do I do it and which format the dataset needs to be? Is there script to evaluate the model reproducing the results of the paper?