Capital Saving Stated Aim Classifier

This is a roberta-base model that is trained to classify whether an explicit set of stated aims extracted from a British historical patent includes a capital-saving objective.

Labels were generated manually and checked with Gemini 2.0 Flash using the attached prompt.

Hyperparameters: lr = 3e-5 batch size = 128

Test set results:

{'eval_loss': 0.32574835419654846,
 'eval_accuracy': 0.89,
 'eval_precision': 0.8916686674669867,
 'eval_recall': 0.89,
 'eval_f1': 0.89003300330033,
 'eval_runtime': 0.4104,
 'eval_samples_per_second': 243.688,
 'eval_steps_per_second': 2.437}