andreagemelli/LFM2.5-350M-IT-Extract

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.35BQuant:BF16Context Size:32kPublished:Sep 2, 2026Architecture:Transformer Featherless Exclusive Cold

andreagemelli/LFM2.5-350M-IT-Extract is a 350 million parameter language model fine-tuned by Andrea Gemelli for Key Information Extraction (KIE) from Italian document images. Based on LiquidAI/LFM2.5-350M, this model specializes in extracting structured JSON data from Italian forms and documents. It achieves an Avg F1 score of 0.5311 on the xfund-kie validation set, significantly outperforming its base model and a reference model for this specific task. This model is optimized for accurate and efficient information extraction in Italian contexts.

Loading preview...

Model Overview

andreagemelli/LFM2.5-350M-IT-Extract is a specialized language model developed by Andrea Gemelli, fine-tuned for Key Information Extraction (KIE) from Italian documents. Built upon the LiquidAI/LFM2.5-350M base model, it focuses on parsing information from Italian form/document images and outputting it in a structured JSON format.

Key Capabilities and Features

  • Specialized KIE for Italian: Specifically designed and optimized for extracting key information from Italian documents and forms.
  • Structured Output: Generates extracted data as JSON objects, facilitating downstream processing.
  • Performance Improvement: Achieves an Avg F1 score of 0.5311 on the xfund-kie Italian validation set, which is a substantial improvement over its base model (0.2138) and a previous reference model (0.2532).
  • Compact Size: With 350 million parameters, it offers efficient deployment for KIE tasks.
  • Dataset: Fine-tuned using a derived version of the XFUND Italian split (xfund-kie).

Ideal Use Cases

This model is particularly well-suited for applications requiring:

  • Automated data extraction from Italian invoices, forms, or other structured documents.
  • Processing of Italian-language visual documents to convert unstructured text into structured data.
  • Integration into workflows that need to parse specific fields from scanned or image-based Italian documents.