bge-m3-zeroshot-v2.0
| Property | Value |
|---|---|
| Parameter Count | 568M |
| License | MIT |
| Paper | View Paper |
| Base Model | BAAI/bge-m3-retromae |
| Context Length | 8192 tokens |
What is bge-m3-zeroshot-v2.0?
bge-m3-zeroshot-v2.0 is a powerful multilingual zero-shot text classifier designed for efficient classification tasks without requiring training data. Built on the BGE-M3 architecture, this model can process texts in multiple languages and handle documents up to 8,192 tokens in length.
Implementation Details
The model uses Natural Language Inference (NLI) as its foundational task, determining whether a hypothesis is "true" or "not true" given a text input. It's trained on commercially-friendly data, including synthetic data generated by Mixtral-8x7B-Instruct and established NLI datasets.
- Trained on synthetic data generated with Mixtral-8x7B-Instruct-v0.1
- Incorporates MNLI and FEVER-NLI datasets
- Supports both single-label and multi-label classification
- Implements FP16 tensor format for efficient computation
Core Capabilities
- Zero-shot classification across multiple languages
- Extended context window of 8,192 tokens
- Flexible hypothesis template customization
- Strong performance across 28 different classification tasks
- Commercial-friendly licensing and training data
Frequently Asked Questions
Q: What makes this model unique?
The model combines multilingual capabilities with an extensive context window and commercial-friendly training data, making it suitable for production environments while maintaining strong performance across diverse classification tasks.
Q: What are the recommended use cases?
The model excels in multilingual text classification scenarios, particularly when working with longer documents. It's ideal for users who need a production-ready solution with commercial licensing compliance and don't require the full capabilities of larger generative models.
