NuNER-v0.1
| Property | Value |
|---|---|
| License | MIT |
| Research Paper | arXiv:2402.15343 |
| Base Architecture | RoBERTa-base |
| Primary Task | Token Classification (NER) |
What is NuNER-v0.1?
NuNER-v0.1 is a specialized entity recognition model built on RoBERTa-base architecture, fine-tuned on the NuNER dataset. It provides enhanced token embeddings specifically optimized for Named Entity Recognition (NER) tasks in English. The model shows significant improvements over the base RoBERTa model, achieving better performance across various few-shot learning scenarios.
Implementation Details
The model is implemented using the transformers library and PyTorch framework. It generates high-quality token embeddings that can be used either out-of-the-box or fine-tuned for specific downstream NER tasks. The model offers two embedding extraction options: a high-quality version combining the last and seventh-to-last hidden states, and a faster version using only the last hidden state.
- Based on RoBERTa-base architecture
- Optimized for token classification tasks
- Supports both inference and fine-tuning workflows
- Achieves 34.3% accuracy on k=1 few-shot learning
Core Capabilities
- Token-level entity recognition
- Flexible embedding extraction options
- Support for both zero-shot and few-shot learning scenarios
- Improved performance over base RoBERTa model (34.3% vs 24.5% for k=1)
Frequently Asked Questions
Q: What makes this model unique?
NuNER-v0.1 offers significantly improved performance over the base RoBERTa model for entity recognition tasks, particularly in few-shot learning scenarios. It provides flexible embedding extraction options and is specifically optimized for NER tasks.
Q: What are the recommended use cases?
The model is best suited for English language entity recognition tasks, particularly when working with limited training data. It can be used either as a feature extractor or fine-tuned for specific domain applications.
