Classifiers

These are ModernBERT-large models (max length 256), fine-tuned on WildSeek with 5-fold cross-validation. Scores are macro-F1 on the held-out test set.