Talmud Punctuator Model A-Q
BEREL 3.0 fine-tuned exclusively on Al HaTorah Berakhot. It predicts ordinary
punctuation plus opening and closing quotation boundaries with separate heads.
The ASCII sequences ?! and !? are normalized to the question-mark class.
- Best held-out epoch: 1
- Punctuation accuracy: 0.9593
- Opening-quote F1: 0.9218
- Closing-quote F1: 0.9051
- Training source: Al HaTorah only
This is an experimental held-out model. Straight source quotes were interpreted at token edges after expanding Al HaTorah abbreviation markup; output quotes are directional curly quotes. The current supervision includes all source quotation spans, not only spans independently verified as biblical quotations.
Model tree for Joshua2/talmud-punctuator-A-quotes-question-normalized
Base model
dicta-il/BEREL_3.0