RELEASE · MODELS · #3075
Dharma releases Field Extractor built on 2B-parameter DharmaOCR-LITE v1.1
Dharma announced a Field Extractor model that takes document images and returns typed JSON fields either open-ended or schema-directed. The system is a 2-billion-parameter checkpoint based on DharmaOCR-LITE v1.1 with a LoRA adapter trained on 43,384 field-level examples and was evaluated on a 546-document benchmark covering 1,638 extraction cases.
KEY POINTS
- Dharma announced a Field Extractor model that takes document images and returns typed JSON fields either open-ended or schema-directed.
- The system is a 2-billion-parameter checkpoint based on DharmaOCR-LITE v1.1 with a LoRA adapter trained on 43,384 field-level examples and was evaluated on a 546-document benchmark covering 1,638 extraction cases.
- A compact, OCR-specialized model that returns schema-constrained, typed JSON can lower cost and latency for high-volume document-to-database workflows compared with general-purpose frontier LLM APIs.
WHY IT MATTERS
A compact, OCR-specialized model that returns schema-constrained, typed JSON can lower cost and latency for high-volume document-to-database workflows compared with general-purpose frontier LLM APIs.