What are the parameters using as arguments in full fine-tuning the google flan model in lab 2?

Could you provide the detailed arguments used in the TrainingArguments for the fully fine-tuned model that was downloaded during Lab 2?

I’ve rented an A100 server for many hours and have spent much money, but I’ve still been unable to reproduce a model that had the same level of performance of the model that was available in the full fine-tuning section of Lab 2.

I tried to increase the max_steps, but still not as good as the downloaded model…

Hi Tony. Unfortunately, this info was lost. Our partner remembers though that the training ran for 5-6 hours to get that performance.