Embedding Models
Embedding models convert text and potentially other data types into high-dimensional vector representations that capture semantic meaning. These vectors are crucial for effective retrieval systems as they enable similarity matching between queries and your data sources.
Supported Embedding Models​
Teammately provides access to high-quality embedding models from OpenAI:
| Model | Description | Dimensions | Context Window | Use Case |
|---|---|---|---|---|
| text-embedding-3-large | OpenAI's most powerful embedding model | 3072 | 8191 tokens | High-accuracy retrieval for complex or specialized content |
| text-embedding-3-small | Efficient embedding model with strong performance | 1024 | 8191 tokens | General-purpose retrieval with good performance-cost balance |
Using Your Own Embedding Models​
You can also use your own embedding models through our platform connection system. This requires:
- Setting up a Platform Connection with your preferred model provider
- Configuring your retrieval pipeline to use the specified embedding model
This approach offers flexibility when your use case demands specific embedding models or when you have existing investments in other embedding technologies.
Embedding Model Selection Considerations​
When choosing an embedding model for your retrieval system, consider:
- Quality vs. Speed: Larger models like text-embedding-3-large offer higher accuracy but may have slightly slower processing times
- Cost Efficiency: Smaller models like text-embedding-3-small provide good performance at lower cost for large-scale deployments
- Domain Specificity: Consider domain-specialized models for technical or industry-specific content
Integration with Retrieval Pipeline​
Teammately's Retrieval system integrates embedding models seamlessly within the RAG pipeline. The embedding process occurs automatically during document processing and when handling queries, ensuring optimal vector representations for accurate retrieval.
This integration is crucial for addressing the hallucination prevention capabilities that form the core of Teammately's approach to reliable, accurate AI systems.