NuExtract
| Property | Value |
|---|---|
| Parameter Count | 3.82B |
| Base Model | Phi-3-mini-4k-instruct |
| License | MIT |
| Tensor Type | F32 |
What is NuExtract?
NuExtract is an advanced fine-tuned version of Microsoft's Phi-3-mini model, specifically designed for structured information extraction tasks. Built on a private high-quality synthetic dataset, this 3.82B parameter model excels at extracting structured information from texts using JSON templates.
Implementation Details
The model is implemented using the Transformers architecture and provides both F32 tensor support. It's designed to handle input texts up to 2000 tokens and can process structured extraction requests through JSON templates.
- Built on Phi-3-mini-4k-instruct architecture
- Supports purely extractive information retrieval
- Includes custom code for efficient processing
- Available in three versions: tiny (0.5B), standard (3.82B), and large (7B)
Core Capabilities
- Zero-shot information extraction
- JSON template-based structured data extraction
- Support for example-based formatting
- Efficient processing of long sequences
- Purely extractive text processing
Frequently Asked Questions
Q: What makes this model unique?
NuExtract's uniqueness lies in its specialized fine-tuning for structured information extraction tasks, with the ability to process JSON templates and maintain purely extractive outputs, ensuring all generated content comes directly from the source text.
Q: What are the recommended use cases?
The model is ideal for automated information extraction tasks, structured data parsing, and scenarios requiring precise extraction of specific information from text documents according to predefined templates.
