After we kind of fine-tune the model with some trainings params we asked to download the really fine-tuned model from AWS. But that real model was likely fine tuned with the different parameters.
I would be very interested to see with what parameters both full and PEFT fine-tuning was done.
Actually I found that AWS Sagemaker lab, which is free, offers more performant environment than a lab so I could try to make a real training there.
Also this course does not explain anything about Trainer API and how to configure and use it. Is there any more advanced courses?