Dharma AI · Document Field Extraction

Dharma Field Extractor

A 2-billion-parameter model, built as a LoRA adapter on DharmaOCR-LITE v1.1, that takes a document image and returns structured JSON, either from an open request or an exact schema with typed fields and nulls for missing values. On a 1,638-case benchmark it scored 0.834 field-level quality, above every frontier LLM configuration tested (Claude Opus 5 at 0.816, the lowest-scoring at 0.741), at a lower self-hosted cost per extraction on a single L40S. Licence and weights availability are not stated on the page.