E5-small-v2
| Property | Value |
|---|---|
| Parameter Count | 33.4M |
| Architecture | 12-layer Transformer |
| Paper | Text Embeddings by Weakly-Supervised Contrastive Pre-training |
| License | MIT |
What is e5-small-v2?
E5-small-v2 is a compact but powerful text embedding model developed for semantic similarity tasks and information retrieval. It uses weakly-supervised contrastive pre-training to generate high-quality text embeddings with a dimension size of 384. The model requires specific prefixes ("query:" or "passage:") for optimal performance and can handle sequences up to 512 tokens in length.
Implementation Details
The model implements a 12-layer transformer architecture with 33.4M parameters. It uses average pooling over the last hidden states and requires text normalization for optimal performance. The model supports both PyTorch and Sentence Transformers frameworks, making it versatile for different implementation needs.
- Embedding dimension: 384
- Maximum sequence length: 512 tokens
- Requires input prefixes: "query:" or "passage:"
- Supports multiple frameworks including PyTorch and Sentence Transformers
Core Capabilities
- Semantic text similarity assessment
- Information retrieval and passage ranking
- Text classification tasks
- Clustering applications
- Cross-lingual semantic analysis (English only)
Frequently Asked Questions
Q: What makes this model unique?
The model's unique strength lies in its efficient architecture (only 33.4M parameters) while maintaining strong performance across various tasks. It uses a distinctive prefix-based input format and weakly-supervised contrastive pre-training approach.
Q: What are the recommended use cases?
The model excels in semantic similarity tasks, passage retrieval, and information retrieval applications. It's particularly well-suited for applications requiring efficient text embeddings while maintaining high accuracy.
