DocFormNER

DocFormNER is a small, on-device cross NER model from the FlowX OpenNER family. Developed by FlowX.AI. Runs 100% on-premise / air-gapped, so no data leaves your boundary.

What it does

• Task: token-classification
• Base model: microsoft/layoutlmv3-base
• Entity types (9): APP_NO, BORROWER, INCOME, INTEREST_RATE, LENDER, LOAN_AMOUNT, LTV, PROPERTY_ADDR, TERM_YEARS
• Held-out F1: 1.0000
• Runtime: CPU, Apple Silicon, one GPU, or browser/edge via ONNX (INT8). ~100-160 ms/doc on CPU.

Why a small model

Fine-tuned encoders match or beat frontier LLMs on structured, convention-bound extraction, at a fraction of the latency and cost, with zero data egress. Identifiers are validated by checksum (IBAN mod-97, card Luhn, ISIN/LEI, container ISO-6346, VIN, national IDs), a correctness guarantee general LLMs lack. See the FlowX OpenNER benchmark for measured results.

Document-AI preview. DocFormNER is a LayoutLMv3 model that reads text + 2D layout + the page image, so it works on scanned / photographed forms (mortgage packets, bills of lading, ACORD forms). This checkpoint is trained on synthetic rendered forms as a walking skeleton; for production, fine-tune on real OCR'd documents (FUNSD / DocLayNet / CORD or your own scans) with the same {words, bboxes, labels, image} shape.

Usage

from transformers import AutoTokenizer, AutoModelForTokenClassification
tok = AutoTokenizer.from_pretrained("flowxai/docformner")
model = AutoModelForTokenClassification.from_pretrained("flowxai/docformner")

License & attribution

Licensed under the Apache License 2.0. Copyright 2026 FlowX.AI (https://flowx.ai). See the NOTICE file. Trained on synthetic, checksum-validated data.

Part of the FlowX OpenNER model family. Synthetic-data F1 reflects an in-distribution synthetic distribution; validate on real documents before production use.