Docker introduced a brand new GenAI Stack in partnership with Neo4j, LangChain, and Ollama throughout its annual DockerCon developer convention keynote. This GenAI Stack is designed to assist builders rapidly and simply construct generative AI functions with out looking for and configuring varied applied sciences.
It consists of pre-configured elements like giant language fashions (LLMs) from Ollama, vector and graph databases from Neo4j, and the LangChain framework. Docker additionally launched its first AI-powered product, Docker AI.
The GenAI Stack addresses fashionable use circumstances for generative AI and is offered within the Docker Studying Middle and on GitHub. It gives pre-configured open-source LLMs, help from Ollama for organising LLMs, Neo4j because the default database for improved AI/ML mannequin efficiency, data graphs to reinforce GenAI predictions, LangChain orchestration for context-aware reasoning functions, and varied supporting instruments and sources. This initiative goals to empower builders to leverage AI/ML capabilities of their functions effectively and securely.
“Builders are excited by the chances of GenAI, however the charge of change, variety of distributors, and extensive variation in know-how stacks makes it difficult to know the place and find out how to begin,” mentioned Scott Johnston, CEO of Docker CEO Scott Johnston. “At present’s announcement eliminates this dilemma by enabling builders to get began rapidly and safely utilizing the Docker instruments, content material, and companies they already know and love along with companion applied sciences on the slicing fringe of GenAI app growth.”
Builders are supplied with simple setup choices that supply varied capabilities, together with easy information loading and vector index creation. This permits builders to import information, create vector indices, add questions and solutions, and retailer them inside the vector index.
This setup permits enhanced querying, consequence enrichment, and the creation of versatile data graphs. Builders can generate various responses in several codecs, corresponding to bulleted lists, chain of thought, GitHub points, PDFs, poems, and extra. Moreover, builders can evaluate outcomes achieved between completely different configurations, together with LLMs on their very own, LLMs with vectors, and LLMs with vector and data graph integration.