distilbert-base-multilingual-cased-sentiments-student
| Property | Value |
|---|---|
| Parameter Count | 135M |
| License | Apache 2.0 |
| Supported Languages | 12 (en, ar, de, es, fr, ja, zh, id, hi, it, ms, pt) |
| Teacher Model | MoritzLaurer/mDeBERTa-v3-base-mnli-xnli |
What is distilbert-base-multilingual-cased-sentiments-student?
This is a compressed sentiment analysis model created through knowledge distillation from a larger mDeBERTa-v3 teacher model. It's designed to perform efficient multilingual sentiment classification across 12 languages while maintaining high accuracy (88.29% agreement with teacher model). The model classifies text into positive, neutral, or negative sentiments.
Implementation Details
The model uses DistilBERT architecture with zero-shot distillation techniques, trained using the hypothesis template "The sentiment of this text is {}." It processes text using F32 tensors and is optimized for both accuracy and efficiency.
- Distilled from mDeBERTa-v3-base-mnli-xnli teacher model
- Trained with 16 batch size and FP16 precision
- Achieves 88.29% agreement with teacher predictions
- Supports multiple languages without requiring translation
Core Capabilities
- Multilingual sentiment analysis across 12 languages
- Three-way classification (positive, neutral, negative)
- Efficient inference with reduced model size
- Zero-shot cross-lingual transfer
Frequently Asked Questions
Q: What makes this model unique?
This model combines broad language support with efficient architecture through knowledge distillation, making it particularly suitable for production deployments requiring multilingual sentiment analysis.
Q: What are the recommended use cases?
The model is ideal for sentiment analysis in multilingual customer feedback, social media monitoring, and content moderation systems where processing efficiency and language coverage are crucial.
