gliner_medium_news-v2.1
| Property | Value |
|---|---|
| License | Apache 2.0 |
| Base Model | microsoft/deberta |
| Paper | ArXiv |
| Developer | Emergent Methods |
What is gliner_medium_news-v2.1?
gliner_medium_news-v2.1 is a specialized Named Entity Recognition (NER) model fine-tuned from GLiNER, designed specifically for improved accuracy in long-context news entity extraction. The model demonstrates up to 7.5% improvement in zero-shot accuracy across 18 benchmark datasets, making it particularly effective for diverse global news analysis.
Implementation Details
The model is built on the microsoft/deberta architecture and trained on the AskNews-NER-v0 dataset. It employs sophisticated synthetic data generation using WizardLM 13B v1.2 for translation/summarization and Llama3 70b for entity extraction. The training process emphasized diversity across countries, languages, topics, and temporal aspects.
- Supports 8 entity types: person, location, date, event, facility, vehicle, number, organization
- Trained on multi-language content including translations from 11 different languages
- Optimized for production use with compact model size
Core Capabilities
- High-accuracy entity extraction from news content
- Multi-language support through translation capabilities
- Zero-shot performance across diverse topics
- Efficient processing for production environments
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive feature is its focus on diverse global perspectives, achieved through careful dataset engineering that ensures representation across different countries, languages, and topics. It combines the power of advanced language models (WizardLM and Llama3) for synthetic data generation while maintaining production-ready efficiency.
Q: What are the recommended use cases?
The model is ideal for news analysis, content categorization, and information extraction from long-form content. It's particularly well-suited for production environments requiring high-throughput entity extraction, such as news aggregation services, content analysis platforms, and automated information processing systems.
