🔍 Read the full analysis: Any Document In, JSON Fields Out: Meet The Dharma Field Extractor on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Hugging Face has introduced Dharma Field Extractor, a 2-billion-parameter model that turns document images into JSON, either by finding fields itself or following a supplied schema. The company reports better benchmark results and lower estimated costs than four tested frontier model configurations, but full scores, independent validation and real-world workflow results were not provided.
Hugging Face has introduced Dharma Field Extractor, a 2-billion-parameter model that converts document images into structured JSON, either by identifying fields on its own or filling a schema provided by a user, as detailed in the original analysis. The company says it outperformed four tested frontier model configurations on its benchmark and had substantially lower estimated costs, though the supplied announcement does not include full scores or independent validation.
The model supports two request types. In open extraction, it identifies information from labels and page layout and returns key-value pairs. In schema-directed extraction, a user specifies field names and types; the model is intended to return those keys, put null for information missing from the document, and arrange table entries row by row.
Hugging Face says a single checkpoint can handle different document types without per-type fine-tuning. Examples it gives include extracting supplier names, invoice numbers, due dates and totals, or passing requisition details into scheduling systems. The company positions the model as a way to turn document content into records other software can use.
The extractor builds on DharmaOCR-LITE v1.1, a 2-billion-parameter OCR model, with a LoRA adapter trained on 43,384 examples, according to Hugging Face. The examples included English documents from the public web, internally sourced Brazilian Portuguese documents and smaller collections in other European and Asian languages. Document types included receipts, invoices, forms, bank statements, identification documents and synthetic examples.
From Scanned Pages to Business Records
Document processing often requires more than reading text: a system must identify which value is an invoice number, which is a total, and how table entries map to columns. Dharma Field Extractor targets that handoff from page content to structured data. If it returns accurate fields in the format a business system expects, it could reduce manual entry or custom mapping work in repetitive workflows.
The two request modes address different needs. A fixed schema can fit software that expects specific keys and data types; open extraction may be useful when a document’s fields are not known ahead of time. Hugging Face’s reported cost advantage could be relevant for organizations processing many pages, but it is an estimate rather than a demonstrated saving in a live operation.
Practical value will depend on factors the benchmark summary does not settle: the mix and quality of documents, the cost of handling mistakes, and how much human review remains necessary. A low model cost per page would not by itself establish a lower total cost for a complete workflow.
Products Worth Considering
EcoFlow Delta 3 Max Portable Power Station – 2048 Wh LiFePO4 Battery, 2400 W Output (3400 W X‑Boost)
Bose SoundLink Flex Bluetooth Speaker (2nd Gen) – Portable Outdoor Speaker with Hi‑Fi Audio, Waterproof and Dustproof, USB‑C, Up to 12 Hours
EcoFlow Delta 3 Portable Power Station – 1024 Wh, 1800 W
As an affiliate, we earn on qualifying purchases.
A Field Layer Above OCR
Optical character recognition, or OCR, turns writing in an image into text. Field extraction adds another step: it interprets the text and layout, assigns values to named fields and can organize table content. Hugging Face presents the new model as a task-specific extension of its work on DharmaOCR-LITE, which the company says it released in April for reading varied documents. The supplied material does not give the year of that release.
For its reported evaluation, Hugging Face says it used 546 documents and 1,638 cases covering open extraction, partial schemas and schemas that included fields absent from the source documents. Its scoring combined field-level F1 by path, including table position, value similarity and the share of outputs that were valid JSON. The summary reports a quality gap of 1.8 to 9.3 points against four frontier configurations, but does not specify the score scale or provide individual results.
The company also reports costs roughly six to 147 times lower than those configurations. Its estimate uses average input and output tokens per page, including reasoning tokens, and providers’ published list prices. The supplied information does not state the Field Extractor’s own price or provide enough detail to independently reproduce the comparison.
“The Field Extractor takes a document image and either a request for everything it can find, or an exact schema to fill, and returns structured JSON.”
— Hugging Face
Benchmark Results Need More Detail
The available benchmark summary leaves key questions unanswered. It does not provide full results, individual model scores, or the scale behind the reported 1.8-to-9.3-point quality gap. It also does not give the Field Extractor’s own price, a public endpoint, deployment requirements or a publication date for the announcement.
The reported comparisons have not been independently validated in the supplied material. It is also unclear how the model performs on unseen document sets, how errors vary across languages and document categories, and what level of human checking businesses would need. The cost estimates use token averages and published list prices; actual spending and total workflow costs may vary with document mix, volume and error handling.
What Real-World Tests Must Show
Hugging Face says users can try the model, but the supplied material does not identify a release date or explain how to access or deploy it. Further evidence would include complete benchmark tables and evaluation details that other researchers or organizations can reproduce.
For prospective users, the next practical step is testing the model against their own documents and required fields. Those trials would need to measure not only JSON validity and extraction accuracy, but also correction rates, review workload, data handling and total cost. Until results from varied, real-world workflows are available, the benchmark supports a model announcement rather than proof of reliable automation across business settings.
Key Questions
What does Dharma Field Extractor do?
It takes a document image and returns information as structured JSON. Users can ask it to identify fields or supply a schema of required field names and types.
How is it different from OCR?
OCR transcribes text from a page. Field extraction adds an interpretation step, assigning values to named fields and organizing information such as table rows.
Did Hugging Face show that it beats other models?
Hugging Face reports that the extractor scored 1.8 to 9.3 points higher on its benchmark than four tested frontier configurations. The supplied summary does not provide the score scale or full results, and independent validation was not included.
How much cheaper is the model?
Hugging Face estimates costs at roughly six to 147 times lower than the tested configurations, based on average tokens per page and published list prices. The material does not state the extractor’s own price, and actual workflow costs may differ.
Is it ready to automate document processing without review?
The supplied material does not establish that. Performance on unseen business documents, error rates and the amount of human review required remain unclear.
Primary source: Hugging Face · via ThorstenMeyerAI.com
