Embeddings Drive the Quality of RAG: Voyage AI in Chat LangChain

The choice of embedding models significantly impacts the overall quality of a chatbot based on Retrieval-Augmented Generation (RAG). In this post, we focus on the case of Chat LangChain, which uses fine-tuned Voyage embeddings in production. We demonstrate that Voyage's generalist model voyage-01 and its enhanced version fine-tuned on LangChain docs, voyage-langchain-01, improve both retrieval quality and response quality compared to OpenAI's text-embedding-ada-002. The results suggest that the quality of the final response is highly correlated with the retrieval quality, and Voyage's embeddings can enable more accurate responses by improving the retrieval quality. As of LangChain 0.0.327, Voyage is integrated into the LangChain Python package, allowing anyone to access the voyage-01 model for their own applications.