Dharma Field Extractor: Any Document In, Structured Fields in JSON Out

Dharma Field Extractor is a 2-billion-parameter model designed to extract structured data from any document image and output it as JSON. The model operates in two modes: an "open" mode that finds all key-value pairs, and a "schema-directed" mode that fills a user-provided data structure. Built upon the DharmaOCR-LITE model, it was trained using a LoRA adapter on a multi-domain and multilingual dataset of over 43,000 examples. Benchmark results show it outperforms general-purpose frontier LLMs like Claude and GPT on this specific task, offering higher quality at a lower cost.