gte-tiny
| Property | Value |
|---|---|
| Parameter Count | 22.7M |
| Model Type | Sentence Transformer |
| Embedding Dimension | 384 |
| Author | TaylorAI |
What is gte-tiny?
gte-tiny is a compact and efficient sentence transformer model that maps sentences and paragraphs to 384-dimensional dense vector space. It is a distilled version of thenlper/gte-small, designed to maintain comparable performance while significantly reducing the model size.
Implementation Details
The model utilizes a BERT-based architecture with mean pooling and achieves impressive performance across various natural language processing tasks. It supports both sentence-transformers and HuggingFace Transformers implementations, making it versatile for different development environments.
- Optimized for semantic similarity tasks
- Supports maximum sequence length of 512 tokens
- Implements mean pooling strategy for sentence embeddings
- Available in both F32 and FP16 precision formats
Core Capabilities
- Semantic Search: Achieves strong performance on retrieval tasks (MAP@100: 83.68 on Quora dataset)
- Text Classification: Shows robust performance across various classification tasks (80.54% accuracy on IMDB)
- Sentence Similarity: Demonstrates strong correlation with human judgments (87.16% correlation on BIOSSES dataset)
- Clustering: Effective for document clustering tasks (57.54 v-measure on StackExchange)
Frequently Asked Questions
Q: What makes this model unique?
The model's key strength lies in its efficient architecture that maintains strong performance while being significantly smaller than its parent model. With only 22.7M parameters, it offers a great balance between computational efficiency and accuracy.
Q: What are the recommended use cases?
The model is particularly well-suited for semantic search, document clustering, and similarity comparison tasks. It's ideal for applications requiring efficient processing of text embeddings while maintaining high accuracy.
