Since the `limit` flag was added, we can test on, let's say 50 samples from the full dataset, giving a small model, like: https://huggingface.co/EleutherAI/pythia-14m It can be run even on cpu and it will be fast. So add real test with real, not mocking/patching stuff.
Since the
limitflag was added, we can test on, let's say 50 samples from the full dataset, giving a small model, like: https://huggingface.co/EleutherAI/pythia-14mIt can be run even on cpu and it will be fast.
So add real test with real, not mocking/patching stuff.