What information is contained in each row of the MPRC dataset?
How does the tokenizer tell the model which part of the input is the first sentence vs second sentence?
Why do we need to pad the the inputs?
Section 3
What does a Trainer do?
What information do you need to pass when constructing a Trainer?
What information do you need to pass when computeing a metric? What information is given in the results?
note: f1 summarizes a model’s accuracy in a way that balances precision and recall. Technically, it is the harmonic mean of precision and recall. It’s not perfect, but it’s very commonly used.
Section 4
Note: look at the for - break. That’s a useful Python trick for debugging iterable things (like data loaders) in notebooks.
What does model(**batch) give us? (Note: the ** means to pass everything in batch as keyword arguments (“kwargs”) to the function. So gets parameters like input_ids=SOMETHING, attention_mask=SOMETHING, labels=SOMETHING.)
Be able to explain what each line of code in the code chunk right before “The evaluation loop” does.