cmboulanger/tei-bibl-annotator-citext-exparser

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 3, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

The cmboulanger/tei-bibl-annotator-citext-exparser is a 3.1 billion parameter model, fine-tuned from Qwen/Qwen2.5-3B-Instruct, designed for annotating plain-text bibliographic citations with inline TEI XML markup. It precisely reproduces the input text, inserting tags for bibliographic elements like author, title, date, and publisher. This model specializes in processing complex citations from legal-sociology scholarship, including short-form and anaphoric references, without paraphrasing or normalization.

Loading preview...

TEI Bibliographic Annotator Overview

This model, cmboulanger/tei-bibl-annotator-citext-exparser, is a 3.1 billion parameter LoRA fine-tune of the Qwen/Qwen2.5-3B-Instruct base model. Its core function is to tag plain-text bibliographic citations with inline TEI XML markup. A key differentiator is its ability to exactly reproduce the input text, inserting XML tags around structural elements such as author, title, date, and publisher, without any paraphrasing or normalization of the citation content.

Key Capabilities

  • Precise TEI XML Annotation: Accurately identifies and tags bibliographic components within citations.
  • Preserves Original Text: Ensures the output citation is byte-identical to the input, with only XML tags added.
  • Handles Complex Citations: Trained on footnotes from legal-sociology scholarship (German and English), including short-form ("Id.", "ibid.") and anaphoric citations, which are often mixed with prose.
  • Specialized Training Data: Utilizes citext-tei-enriched and exparser-tei corpora, comprising 16,279 training records from 418 documents.

Performance and Usage

During development, a training pilot achieved a 99.83% parse success rate (valid TEI XML) and a 42.50% exact match rate on a 1,207-record development set. The model expects a fixed system instruction and the raw citation as user input, with greedy generation (do_sample=False) and max_new_tokens=1024.

License

This model is intended for research use only. It inherits the Qwen Research License from its base model, which restricts commercial use.

Finetunes
1 models