Spiced - Semantic Sentence Embedding Model
| Property | Value |
|---|---|
| Paper | Research Paper |
| Framework | PyTorch (sentence-transformers) |
| Vector Dimension | 768 |
| Base Architecture | MPNet |
What is spiced?
Spiced is a specialized sentence transformer model developed by copenlu that converts text into high-dimensional vector representations. Built on the MPNet architecture, it maps sentences and paragraphs into a 768-dimensional vector space, making it particularly effective for semantic similarity tasks, clustering, and information retrieval applications.
Implementation Details
The model utilizes the sentence-transformers framework with a sophisticated architecture combining an MPNet transformer model, mean pooling layer, and normalization. It was trained using cosine similarity loss with AdamW optimizer and implements a warmup linear scheduling strategy over 5 epochs.
- Maximum sequence length: 384 tokens
- Training batch size: 8
- Learning rate: 1e-05
- Warmup steps: 10000
Core Capabilities
- Sentence and paragraph embedding generation
- Semantic similarity computation
- Text clustering
- Information retrieval
- Cross-lingual text comparison
Frequently Asked Questions
Q: What makes this model unique?
The model combines MPNet's powerful architecture with specialized training for semantic similarity tasks, offering a balance between performance and efficiency. Its 768-dimensional output space provides rich semantic representations while maintaining practical usability.
Q: What are the recommended use cases?
The model excels in applications requiring semantic understanding, such as document similarity comparison, semantic search systems, text clustering, and information retrieval tasks. It's particularly suitable for production environments requiring robust sentence embeddings.
