voyage-code-3 Embedding Model
Text embedding model for code retrieval and AI applications. 32K context length. Multiple output dimensions and embedding quantization.
Product Description
Overview
Text embedding models are neural networks that transform texts into numerical vectors. They are a crucial building block for semantic search/retrieval systems and retrieval-augmented generation (RAG) and are responsible for the retrieval quality. voyage-code-3 is optimized for code retrieval, outperforming OpenAI-v3-large and CodeSage-large by an average of 13.80% and 16.81% on a suite of 238 code retrieval datasets, respectively. By supporting smaller dimensions with Matryoshka learning and quantized formats like int8 and binary, voyage-code-3 can also dramatically reduce storage and search costs with minimal impact on retrieval quality. Latency is 90 ms for a single query with at most 100 tokens, and throughput is 12.6M tokens per hour at $0.22 per 1M tokens on an ml.g6.xlarge. Learn more about voyage-code-3 here: https://blog.voyageai.com/2024/12/04/voyage-code-3/
Highlights
Optimized for code retrieval. Outperforms OpenAI-v3-large and CodeSage-large by an average of 13.80% and 16.81% on a suite of 238 code retrieval datasets, respectively.
Supports embeddings of 2048, 1024, 512, and 256 dimensions and offers multiple embedding quantization, including float (32-bit floating point), int8 (8-bit signed integer), uint8 (8-bit unsigned integer), binary (bit-packed int8), and ubinary (bit-packed uint8).
32K token context length. Latency is 90 ms for a single query with at most 100 tokens. 12.6M tokens per hour at $0.22 per 1M tokens on an ml.g6.xlarge.
