Snowflake's New Arctic AI Outperforms Meta's Llama 3 with Half the Training

Arctic is the new open-source AI model announced by Snowflake. It is designed to be more efficient than Meta's Llama 3 at enterprise tasks, such as SQL coding for database retrieval.

According to Snowflake's claims on its GitHub repository, Arctic is on par or better than both Llama 3 8B and Llama 2 70B on enterprise metrics, while using less than half of the training compute budget.

Arctic can be accessed immediately as an inference service through Snowflake's Cortex offering. It will also be available on Amazon AWS, Hugging Face, and other venues.

Snowflake's Arctic uses a Dense - MoE Hybrid Transformer model, which combines a 10B dense transformer model with a residual 128x3.66B MoE MLP, resulting in 480B total and 17B active parameters chosen using a top-2 gating. This approach is designed to conserve computing and memory needs.

Snowflake has released all of Arctic's parameters and model code under the Apache 2.0 open-source license, along with the data recipe for training proposed by the company, a collection of research insights, and a Hugging Face repository.