AmlNarrate is a small, on-device banking 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: answerdotai/ModernBERT-base
Entity types (4): AMOUNT, ENTITY, JURISDICTION, TXN_TYPE
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.

Usage

python

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

License & attribution

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