Snowflake introduces its own flagship generative AI model

Snowflake, the cloud computing company, has introduced a new generative AI model called Arctic LLM, specifically designed for enterprise use. This model, available under an Apache 2.0 license, is optimized for generating database code and is free for both research and commercial purposes.

Arctic LLM is part of Snowflake’s Arctic family of generative AI models and is positioned as a competitor to Databricks’ DBRX in the enterprise space. Snowflake claims that Arctic LLM outperforms DBRX in tasks such as coding and SQL generation, as well as surpasses other models like Meta’s Llama 2 70B and Mistral’s Mixtral-8x7B.

The model, built using a mixture of experts (MoE) architecture, contains 480 billion parameters but only activates 17 billion at a time to drive 128 separate expert models. This design allowed Snowflake to train Arctic LLM on open public web datasets at a significantly lower cost compared to similar models.

While Snowflake is providing resources and support for users to fine-tune and run Arctic LLM, the model will initially be available on Cortex, Snowflake’s platform for AI and machine learning applications. The company aims to make Arctic LLM accessible across various hosting services, including Hugging Face, Microsoft Azure, and Lamini.

Despite Snowflake’s enthusiasm for Arctic LLM, some experts question its significance in a market saturated with generative AI models. Concerns about the model’s relatively small context window and potential for hallucinations raise doubts about its practicality for enterprise use.

As the generative AI landscape continues to evolve, Snowflake’s Arctic LLM represents the company’s commitment to innovation and providing advanced AI solutions for enterprise customers. However, the true impact and adoption of this model remain to be seen in a competitive and rapidly changing market.