Dharma Field Extractor: Any Document In, Structured Fields in JSON Out
Blog post from Hugging Face
Dharma-AI introduces Dharma Field Extractor, a 2-billion-parameter model that converts document images into structured JSON for tasks such as invoice, ID, medical form, and contract processing. Built on DharmaOCR-LITE v1.1 and enhanced with a LoRA adapter trained on 43,384 multilingual, field-level examples, it supports open extraction of discovered fields and schema-directed extraction using specified keys, types, nested tables, and null values for absent information. The company reports that, across a 546-document benchmark covering open, partial-schema, and nonexistent-field scenarios, the model achieved a 0.834 field-level score, exceeding tested frontier models including Claude Opus 5 and GPT-5.6 configurations while costing an estimated 6 to 147 times less when self-hosted on a single L40S GPU. Held-out evaluations on SROIE receipts and FUNSD forms are presented as evidence of generalization, including extraction of handwritten text, empty fields, tables, addresses, and boolean checkboxes. The release argues that specialized small models can provide more accurate and economical alternatives to general-purpose APIs for bounded enterprise document-extraction workflows.
| Trend | Post Mentions | Total Month Mentions | Posts | Companies | MoM |
|---|---|---|---|---|---|
| AI Model Fine-tuning | 3 | 613 | 130 | 56 | -29% |
| LLM | 3 | 3,355 | 742 | 191 | -49% |
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.