Universal-NER/UniNER-7B-definition
UniNER-7B-definition is a 7 billion parameter language model developed by Universal-NER, trained from Llama-7B on the Pile-NER-definition dataset. This model specializes in understanding short-sentence definitions of entity types and demonstrates robustness against variations in type paraphrasing, distinguishing it from models focused on common NER tags. It is designed for research purposes in open Named Entity Recognition without relying on human-labeled data.
Loading preview...
UniNER-7B-definition Overview
UniNER-7B-definition is a 7 billion parameter model, fine-tuned from Llama-7B by Universal-NER. Its core innovation lies in its training methodology, utilizing the Pile-NER-definition data, which was generated by prompting gpt-3.5-turbo-0301 to extract entities and define their types with short sentences, eliminating the need for human-labeled data.
Key Capabilities & Differentiators
- Definition-based Entity Understanding: Unlike models optimized for recognizing common, short NER tags (e.g., person, location), UniNER-7B-definition excels at interpreting and understanding entity types based on short-sentence definitions.
- Robustness to Type Paraphrasing: The model shows enhanced resilience to different phrasings of entity type definitions.
- Data Collection Innovation: It leverages a novel data collection approach using LLM prompting for synthetic data generation, detailed in their paper.
Intended Use Cases
This model is particularly suited for research in open Named Entity Recognition where the goal is to identify entities based on descriptive definitions rather than predefined, fixed categories. Its ability to handle variations in type definitions makes it valuable for flexible and adaptable NER tasks. Inference is performed by querying one entity type at a time using a specific prompting template.