distilbert-NER
| Property | Value |
|---|---|
| Parameter Count | 65.2M |
| License | Apache 2.0 |
| Paper | DistilBERT Paper |
| Performance | F1: 0.9217, Precision: 0.9202 |
What is distilbert-NER?
distilbert-NER is a lightweight and efficient Named Entity Recognition model based on DistilBERT architecture. It's specifically designed to identify four types of entities: locations (LOC), organizations (ORG), persons (PER), and miscellaneous (MISC). As a distilled version of BERT, it maintains strong performance while requiring significantly fewer computational resources.
Implementation Details
The model was fine-tuned on the CoNLL-2003 Named Entity Recognition dataset using a single NVIDIA V100 GPU. It employs the DistilBERT architecture, which is a compact version of BERT that maintains 97% of its performance while being 40% smaller.
- Training Dataset: CoNLL-2003 English (203,621 tokens)
- Model Architecture: DistilBERT base cased
- Evaluation Metrics: F1 Score: 0.9217, Precision: 0.9202, Recall: 0.9232
Core Capabilities
- Named Entity Recognition for 4 entity types
- Efficient inference with reduced parameter count
- Support for both beginning (B-) and inside (I-) entity tags
- High accuracy (98.10%) on standard NER tasks
Frequently Asked Questions
Q: What makes this model unique?
This model provides an optimal balance between performance and efficiency, achieving 92% F1 score while using only 65.2M parameters, making it significantly lighter than BERT-based alternatives (110M-340M parameters).
Q: What are the recommended use cases?
The model is ideal for production environments where resource efficiency is crucial but high accuracy is required. It's particularly suited for tasks involving identification of persons, organizations, locations, and miscellaneous entities in English text.
