Gemma-Embeddings-v1.0
| Property | Value |
|---|---|
| Author | |
| Base Model | Gemma2 9B |
| Training Data | BGE-EN-ICL |
| Model URL | Hugging Face |
| MTEB Score | 72.72 |
What is Gemma-Embeddings-v1.0?
Gemma-Embeddings-v1.0 is a state-of-the-art dense vector embedding model developed by Google, currently ranking #1 on the MTEB leaderboard. This research project specializes in generating high-quality embeddings optimized for retrieval tasks, demonstrating superior performance across various benchmark categories.
Implementation Details
Built upon the powerful Gemma2 9B architecture and trained on the BGE-EN-ICL dataset, this model represents a significant advancement in embedding technology. It achieves remarkable scores across different tasks, particularly excelling in Classification (90.00%), Retrieval (63.71%), and Reranking (62.14%).
- Achieves state-of-the-art performance on MTEB with a score of 72.72
- Outperforms previous leaders like BGE-EN-ICL and NV-Embed-v2
- Particularly strong in classification and retrieval tasks
- Significant improvement in summary task performance (40.52%)
Core Capabilities
- Dense vector embedding generation
- Superior performance in classification tasks
- Enhanced retrieval capabilities
- Improved text similarity matching
- Effective clustering and reranking
Frequently Asked Questions
Q: What makes this model unique?
The model's exceptional performance across the MTEB benchmark, particularly its leading position with a 72.72 score, sets it apart. It shows balanced excellence across various tasks while significantly improving summary task performance compared to competitors.
Q: What are the recommended use cases?
The model is particularly well-suited for text retrieval, classification tasks, and reranking applications. It excels in scenarios requiring high-quality text embeddings for similarity matching and information retrieval.
