A ModernBERT regression model that scores local-ordinance text along the Problem Salience axis of the LOCUS (Local Ordinances Corpus, United States) dataset.

Fine-tuned from answerdotai/ModernBERT-base. The target is a TrueSkill mu distilled from pairwise LLM comparisons on the problem salience axis, then z-score normalized across the training corpus.

Usage

python

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tok = AutoTokenizer.from_pretrained("LocalLaws/LOCUS-Problem-Salience")
model = AutoModelForSequenceClassification.from_pretrained("LocalLaws/LOCUS-Problem-Salience")
model.eval()

text = "No person shall keep any swine within the city limits."
enc = tok(text, return_tensors="pt", truncation=True, max_length=2048)
with torch.no_grad():
    score = model(**enc).logits.squeeze(-1).item()
print(score)