We persistently hear from our prospects that one of many headwinds to transitioning Generative AI purposes from pilot to manufacturing is the accuracy — or lack thereof — of the outcomes generated by off-the-shelf massive language fashions (LLMs). One of many ways in which organizations are addressing this accuracy hole is by constructing what are referred to as compound AI techniques, which may embrace a Retrieval Augmented Technology (RAG) structure. RAG architectures and compound AI techniques improve the standard of responses from off-the-shelf LLMs by incorporating related company- or domain-specific knowledge as a part of the immediate and response. At Databricks, we view this as a strategy to shift from LLM common intelligence to what we name knowledge intelligence. Even a small enchancment within the high quality and effectivity of retrieval could make an outsized influence on the tip consumer expertise.
A top-quality embedding mannequin is a cornerstone to an correct RAG system. And with the explosion of RAG purposes developed on Databricks Mosaic AI this yr, it’s crucial that Databricks is ready to provide industry-leading embedding and rerank fashions. That’s why we’re excited to share that Databricks Ventures has invested within the Collection A funding spherical of Voyage AI, a startup co-founded by CEO Tengyu Ma, a pioneer within the discipline of deep studying.
We’re additionally asserting a brand new partnership that brings Voyage rerank and embedding fashions to Mosaic AI Mannequin Serving for Databricks prospects. All of our work with Voyage AI is in service of serving to our prospects construct knowledge intelligence and high-quality manufacturing AI techniques.
Why Voyage AI
Voyage AI gives a few of the world’s greatest embedding fashions out there in the present day. Their high-quality embedding and rerank fashions enable firms to boost RAG search and retrieval accuracy and effectivity, leading to extra correct RAG options and compound AI techniques. Their fashions are additionally optimized for particular domains and tailor-made to an organization’s knowledge. Because of this, Voyage AI’s fashions are being utilized by many main AI firms to ship best-in-class consumer experiences.
What’s Subsequent for Voyage AI and Databricks
We’ll provide Voyage AI’s newest technology of embedding and rerank fashions natively throughout the Mosaic AI Mannequin Serving resolution. By providing these fashions natively, we’ll allow our prospects to securely and cost-effectively construct production-quality RAG purposes, they usually can accomplish that confidently, realizing that Databricks gives the long-term help and stability that enterprises require.
We anticipate to announce integration of Voyage AI’s fashions inside Mosaic AI within the coming months. Keep tuned for extra info!