When using "prodigy train texcat", how should I interpret a loss that is decreasing and the f-score that is also decreasing, as in the example below? Does this mean it's overfitting?
Probably yes -- that probably means it's overfitting.Your dataset is pretty small though, so you only have 100 evaluation examples. This means that the second model is only getting 7 more wrong than the first model.
I would say you should just annotate more data. It shouldn't take long to collect get to one or two thousand examples, which should let you train better models and evaluate more reliably.