Single class training issues
#2
by
patrickjdarrow
- opened
Hello, my team has attempted using this model to train on a dataset with a single class. After much effort spent on tuning and pipeline introspection we have only received totally unpredictable training/eval curves. In our experience, this unexpected behavior is present, but avoidable, with some multiclass tasks.
If anyone has information on why this would be happening with respect to the model, theory, HF's training modules, etc., we would appreciate the discussion. Cheers.