numind/NuExtract-2.0-8B

VISIONPricing:Input $0.4 / Output $0.8Concurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:32kPublished:May 6, 2025License:mitArchitecture:Transformer0.1K Open Weights Featherless Exclusive Cold

NuExtract 2.0-8B by NuMind is a 7 billion parameter multimodal and multilingual model specifically trained for structured information extraction tasks. Based on the Qwen2.5-VL-7B-Instruct architecture, it excels at extracting data from both text and image inputs using a JSON template to define the desired output structure. Its primary differentiator is its specialized focus on information extraction, supporting various data types and in-context learning for improved accuracy.

Loading preview...

NuExtract 2.0-8B: Specialized Multimodal Information Extraction

NuExtract 2.0-8B, developed by NuMind, is a 7 billion parameter model built upon the Qwen2.5-VL-7B-Instruct architecture. It is specifically designed and optimized for structured information extraction from both text and image inputs, supporting multilingual content.

Key Capabilities

  • Structured Data Extraction: Extracts information into a user-defined JSON template, supporting types like verbatim-string, string, integer, number, date-time, arrays, enum, and multi-label.
  • Multimodal Input Support: Processes information from both textual documents and images, allowing for flexible data sources.
  • Multilingual Processing: Capable of handling information extraction across multiple languages.
  • In-Context Learning (ICL): Improves extraction accuracy and adherence to specific formatting requirements by providing examples within the prompt.
  • Template Generation: Can automatically generate extraction templates from various formats (e.g., XML) or natural language descriptions.
  • Optimized for Extraction: Recommended to use with a temperature at or very close to 0 for precise extraction tasks.

Good for

  • Automated Data Capture: Extracting specific fields from invoices, forms, reports, or other structured/semi-structured documents.
  • Knowledge Graph Population: Converting unstructured text and images into structured data for databases or knowledge bases.
  • Document Processing Workflows: Integrating into systems that require precise and reliable information extraction from diverse inputs.
  • Custom Extraction Needs: Users who need to define custom extraction schemas and require high accuracy for specific data points.

Finetunes
4 models