Model Card for Qwen3.5-0.8B-SFT-name-parser-yaml
This model is a fine-tuned version of Qwen 3.5 (0.8B parameters) designed to parse unstructured, highly varied historical person names into a strict, structured YAML format. It was developed to process millions of records for cultural heritage platforms efficiently and cost-effectively.
Model Details
Model Description
This model addresses a specific entity extraction problem in the Galleries, Libraries, Archives, and Museums (GLAM) sector. Large frontier models proved too expensive and struggled with strict JSON syntax and component isolation when parsing messy historical names (achieving only ~70% accuracy).
By fine-tuning a small open-source model (0.8B parameters) on high-quality synthetic data and changing the target output format from JSON to YAML, this model achieves 94% to 96% accuracy while remaining small enough to run highly efficiently on local hardware. It extracts components such as first names, last names, middle names, life dates, flourished dates, titles, and extra information.
- Developed by: Yale University LUX Platform Team /
small-models-for-glam - Model type: Causal Language Model (Fine-tuned for specific entity extraction)
- Language(s) (NLP): English (
en), with multi-lingual naming conventions included (e.g., French, German, Italian, Arabic, Chinese). - License: Apache 2.0 (Inherited from Qwen base model)
- Finetuned from model: Qwen 3.5 0.8B
Model Sources
- Repository: https://huggingface.co/small-models-for-glam/Qwen3.5-0.8B-SFT-name-parser-yaml
- Associated Dataset: synthetic-parsed-names-yaml
Uses
Direct Use
The model is intended to be used directly for extracting structured data from raw text strings of historical names.
Given an input string like "Cynthia Ponce'Hayes (born 1633), born 1633", the model will output syntactically valid YAML identifying the parsed components:
first_name: Cynthia
last_name: Ponce'Hayes
middle_names: []
temporal:
- start: 1633
end: null
type: life_span
titles: []
extra_info: []