New LLM: Snowflake Arctic Model for SQL and Code Generation

Table of Contents

1. Introduction
2. Arctic Arctic LLM
3. The LLM for Enterprise Use Cases
4. NVIDIA NIM microservices
5. Experience the model

---

Introduction

Large language models (LLMs) have made significant advancements in natural language processing (NLP). Arctic, developed by Snowflake, is a new open LLM optimized for high performance and cost-efficiency in various NLP tasks. It employs a Dense-MoE Hybrid transformer architecture to effectively handle communication overhead and utilize resources efficiently during training and inference.

Arctic Arctic LLM

Arctic Arctic LLM is based on a Dense-MoE Hybrid transformer architecture, combining a 10B parameter dense transformer model with a residual 128×3.66B MoE MLP. This architecture consists of 480B total parameters with top-2 gating for experts, allowing for top-tier intelligence while engaging a moderate number of active parameters during training and inference.

The LLM for Enterprise Use Cases

Arctic is tailored for SQL, coding, and instruction following in enterprise applications. It excels in semantic parsing, SQL translation, code generation, and instruction-following tasks, outperforming other open models in benchmarks like Spider, HumanEval+, MBPP+, and IFEval.

NVIDIA NIM microservices

Arctic is optimized for latency and throughput and is now supported by NVIDIA NIM microservices, simplifying the deployment of performance-optimized AI models for enterprise developers. The NVIDIA API catalog offers free cloud credits for testing at scale, POC development, and seamless deployment with frameworks like Langchain and LlamaIndex.

Experience the model

Developers can experience Arctic through NVIDIA's API catalog, testing and deploying the model with NVIDIA cloud credits and NIM microservices. The flexibility to run anywhere ensures data security, platform independence, and optimal resource utilization across on-premises, cloud, or workstation deployments.

For more information and getting started, visit build.nvidia.com.