Snowflake Unveils Enterprise-Grade Generative AI Model, Arctic LLM

Snowflake's Arctic LLM is an enterprise-grade generative AI model optimized for enterprise workloads, such as generating database code. It is part of a family of generative AI models called Arctic and is available under an Apache 2.0 license. Arctic LLM is designed to help build enterprise-grade products and is free for research and commercial use.

Arctic LLM utilizes a mixture of experts (MoE) architecture, which breaks down data processing tasks into subtasks and delegates them to smaller, specialized 'expert' models. Although Arctic LLM contains 480 billion parameters, it only activates 17 billion at a time to drive the 128 separate expert models. This method allows for efficient data processing and is said to be cost-effective.

Snowflake provides various resources to assist users with Arctic LLM, including:

• Coding templates: To help users get started with the model.
• List of training sources: To guide users through the process of fine-tuning the model for specific use cases.

However, fine-tuning or running Arctic LLM requires around eight GPUs, which may be costly and complex for most developers. Snowflake plans to make Arctic LLM available across various hosts to alleviate this.

Arctic LLM will be available first on Cortex, Snowflake's platform for building AI- and machine learning-powered apps and services. Snowflake promotes Cortex as the preferred way to run Arctic LLM, highlighting benefits such as security, governance, and scalability.

Arctic LLM's context window, which refers to the amount of input data the model considers before generating output, ranges from ~8,000 to ~24,000 words, depending on the fine-tuning method. This is relatively small compared to other models like Anthropic’s Claude 3 Opus and Google’s Gemini 1.5 Pro, which may lead to limitations such as the model forgetting the content of recent conversations.