rafmacalaba/lfm2.5-350M-focus-metadata

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.35BQuant:BF16Context Size:32kPublished:Oct 1, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The rafmacalaba/lfm2.5-350M-focus-metadata model is a 350 million parameter, instruction-tuned language model based on LiquidAI/LFM2.5-350M, designed for extracting passage-grounded source attributes and data-use metadata. It processes a passage chunk and a cited data mention to output structured JSON for two tasks: identifying data attributes (tags, geography, year, data_type) or describing data usage (role, action, summary). This model is specifically optimized for structured information extraction from text, providing metadata for cited data mentions rather than general text generation.

Loading preview...

Overview

rafmacalaba/lfm2.5-350M-focus-metadata is a 350 million parameter instruction-tuned model built on LiquidAI/LFM2.5-350M. It specializes in extracting structured metadata from text passages related to cited data mentions. Given a text passage and a specific data mention, the model performs one of two tasks, outputting the results in JSON format.

Key Capabilities

  • Data Attribute Extraction: Identifies theme tags (e.g., geospatial, agriculture, health) and source attributes (e.g., geography, year, producer, data_type) for a given data mention within a passage.
  • Data Usage Description: Determines the role (primary, supporting, background) and usage action (source, analyze, curate, validate, inform) of a data mention, along with a concise English summary of its use.
  • JSON Output: Provides structured output, making it suitable for automated processing and integration into data management systems.
  • Passage-Grounded: Ensures that all extracted information is directly supported by the provided text passage, preventing hallucination.

Good for

  • Automated Metadata Generation: Ideal for systems requiring automatic extraction of structured metadata about data sources and their usage from research papers, reports, or other textual documents.
  • Information Retrieval: Enhancing search and discovery of data by providing rich, structured attributes and usage context.
  • Data Governance and Cataloging: Assisting in the creation and maintenance of data catalogs by systematically identifying how data is cited and utilized.
  • Specialized NLP Tasks: Applications focused on fine-grained information extraction from domain-specific texts where identifying data mentions and their context is crucial.

Finetunes
1 models