mDeBERTa-v3-base-mnli-xnli
| Property | Value |
|---|---|
| Parameter Count | 279M |
| License | MIT |
| Languages Supported | 100+ (15 primary) |
| Primary Paper | DeBERTa-v3 Paper |
What is mDeBERTa-v3-base-mnli-xnli?
mDeBERTa-v3-base-mnli-xnli is a state-of-the-art multilingual model designed for natural language inference (NLI) and zero-shot classification tasks. Developed by Microsoft and fine-tuned by Moritz Laurer, this model represents a significant advancement in multilingual NLP, supporting 100 languages with particular optimization for 15 core languages.
Implementation Details
The model is built on Microsoft's mDeBERTa architecture, pre-trained on the CC100 multilingual dataset and fine-tuned on both XNLI and MNLI datasets. It achieves an impressive average accuracy of 80.8% across 15 languages, with English performance reaching 88.3%.
- Pre-trained on CC100 multilingual dataset
- Fine-tuned on professionally translated XNLI development set (37,350 texts)
- Incorporates English MNLI training set (392,702 texts)
- Supports zero-shot classification across languages
Core Capabilities
- Multilingual Natural Language Inference
- Zero-shot Classification
- Cross-lingual Transfer Learning
- Text Classification across 100+ languages
Frequently Asked Questions
Q: What makes this model unique?
The model's uniqueness lies in its combination of broad language coverage (100+ languages) while maintaining state-of-the-art performance for a base-sized model. It specifically avoids using machine-translated training data, focusing instead on high-quality professional translations to prevent overfitting.
Q: What are the recommended use cases?
The model excels in zero-shot classification tasks and natural language inference across multiple languages. It's particularly useful for applications requiring multilingual text understanding without the need for language-specific training data.
