Model Card of GEODE/mt5-small-coords-norm

This model is fine-tuned version of google/mt5-small for extracting and normalizing geographic coordinates from texts.

Overview

• Language model: google/mt5-small
• Language: French
• Training data:
• Online Demo: https://huggingface.co/spaces/GEODE/edda-coordinates
• Repository:

Usage

from transformers import pipeline

pipe = pipeline("text2text-generation", model="GEODE/mt5-small-coords-norm")
pipe("* AACH ou ACH, s. f. petite ville d'Allemagne dans le cercle de Souabe, près de la source de l'Aach. Long. 26. 57. lat. 47. 55.")

Evaluation

5-Fold Cross-Validation Results

Metric	Score
Mean Exact Match	0.8365
Mean Char F1	0.9675

Training hyperparameters

The following hyperparameters were used during fine-tuning:

• dataset_path:
• dataset_name: edda-coordinata
• input_types: ['encyclopedic_text_entry', 'dms_coordinates']
• output_types: 'dms_coordinates'
• model: google/mt5-small
• max_length: 512
• max_length_output: 128
• epoch: 10
• batch: 8
• lr: 0.0005
• random_seed: 42
• gradient_accumulation_steps: 1

Citation

If you use the EDDA-Coordinata dataset or the associated models, please cite our LREC 2026 paper:

@inproceedings{moncla2026edda,
  title={EDDA-Coordinata: An Annotated Dataset of Historical Geographic Coordinates},
  author={Moncla, Ludovic and Nugues, Pierre and Joliveau, Thierry and McDonough, Katherine},
  booktitle={Proceedings of the 2026 Language Resources and Evaluation Conference (LREC 2026)},
  year={2026},
  url={[https://arxiv.org/abs/2602.23941](https://arxiv.org/abs/2602.23941)}
}