Harnessing the Power of Voyage-Lite-02-Instruct: A Specialized Text Embedding Model
The voyage-lite-02-instruct model is an advanced text embedding solution offered by Voyage AI, specifically designed to excel in tasks like classification, clustering, and sentence textual similarity. This instruction-tuned model brings a unique set of features tailored for these specialized applications.
Model Overview
The voyage-lite-02-instruct model offers a generous context length of 4,000 tokens and operates with an embedding dimension of 1,024. This model maintains a batch size limit of 128 to ensure consistency across all Voyage AI models, making it a robust choice for structured and large-scale data processing.
Key Features
- Context Length: 4,000 tokens ensuring extensive data coverage.
- Embedding Dimension: 1,024, providing a detailed representation of input data.
- Batch Size Limit: 128, supporting efficient data processing.
Use Cases
While the voyage-lite-02-instruct model shines in its intended roles, it's important to note that it's not the default choice for general-purpose embedding tasks. For broader applications, Voyage AI recommends models like voyage-large-2 or voyage-2.
Performance
Despite its specialized nature, the voyage-lite-02-instruct model performs exceptionally well in its targeted use cases. However, for more generalized tasks, the voyage-large-2 model may offer superior performance.
Configuration and Usage
To maximize the model's capabilities, users can specify the input_type as either query or document, enhancing retrieval quality. Additionally, the model can be configured to truncate input texts exceeding the context length, ensuring optimal performance.
def get_embeddings(docs: List[str], input_type: str, model: str = "voyage-lite-02-instruct") -> List[List[float]]:
response = voyage_client.embed(docs, model=model, input_type=input_type)
return response.embeddings
Evaluation and Benchmarks
While this model is optimized for specific tasks, it is worth noting that Voyage AI's voyage-large-2-instruct model leads the MTEB leaderboard, outperforming many commercial models in several benchmarked tasks.
Additional Parameters
Additional configuration options include setting a base_url for a custom Voyage AI endpoint and enabling truncate to manage text input effectively.
