Allow users to fine-tune models without specifying a metric (Python) #580
Description
The fine-tuning recipes (Python notebooks) currently require users to specify a metric to fine-tune against. We should enable them to skip that step (e.g. there is no metric), which would simply use all available inferences. (This is helpful to collect data with an expensive/slow model and use it to fine-tune a cheaper/faster model.)
#579 is similar but for the dashboard