davanstrien/ultrabert-380m-curation
Zero-Shot Classification • 0.4B • Updated • 22 • 4
A 380M curation decision model, its training data (source, labeller, licence per row) and a 1M-page FineWeb example.
Note The model: topic, format, educational value, difficulty; System One API
Note Training data with source, labeller and licence per row
Note Worked example: 1,020,000 FineWeb pages scored on HF Jobs in ~19 min for ~$5
Note Base encoder
Note Decision data used (licence-filtered) in the mix
Note General decision model compared in the evaluation
Note General decision model compared in the evaluation