- 스노우플레이크, 업계 최고 수준의 개방성 갖춘 엔터프라이즈급 LLM ‘아크틱’ 출시 > 뉴스 - 스타트업 커뮤니티 씬디스
- 스노우플레이크 아크틱 모델군에 동급 최고 성능·효율 갖춘 개방형 대규모언어모델 추가아파치 2.0 라이선스 기반 오픈소스… 다양한 프레임워크 지원…, 스타트업에 종사하시는 여러분들의 놀이터 씬디스는 스타트업 커뮤니티 입니다.
Snowflake adds a best-in-class, open large language model to its Snowflake Arctic model family.
Open-sourced under the Apache 2.0 license, it offers flexibility for customization with support for various frameworks.
Snowflake, the global data cloud company, has launched Snowflake Arctic, an enterprise-grade large language model (LLM) that delivers industry-leading openness and performance.
Designed with Snowflake's proprietary Mixture-of-Experts (MoE) approach, Arctic delivers best-in-class performance and productivity. It is optimized to handle complex enterprise requirements and meets the highest standards across a variety of criteria, including SQL code generation and command execution.
Specifically, Arctic is available under the Apache 2.0 license, which allows for free commercial use. Snowflake has released details on the AI training methods, establishing a new open standard for enterprise-grade AI technology. In addition, the Arctic LLM is part of the Snowflake Arctic model family, which also includes text embedding models for search applications.
“Snowflake’s AI research team has been driving innovation at the forefront of AI, marking a significant turning point for our company,” said Sridhar Ramaswamy, CEO of Snowflake. “By open-sourcing the industry’s best performance and efficiency to the AI community, Snowflake is expanding the possibilities of open-source AI. We are also enhancing Snowflake’s AI capabilities to provide customers with capable and trustworthy AI models.”
◇ Arctic: An open-source LLM that supports extensive collaboration
According to a recent Forrester report, approximately 46% of global enterprise AI decision-makers stated that they are leveraging existing open-source LLMs to introduce generative AI into their organizations as part of their AI strategy. Snowflake’s Data Cloud currently powers data-driven initiatives for over 9,400 companies and organizations worldwide. These organizations can now leverage data using an LLM with industry-leading openness.
As an open-source model under the Apache 2.0 license, Arctic also provides the flexibility to choose inference and training methods along with code templates. Users can utilize and customize Arctic using their preferred frameworks, such as NVIDIA NIM, NVIDIA TensorRT-LLM, vLLM, and Hugging Face. Arctic can be deployed immediately using serverless inference through Snowflake Cortex. Snowflake Cortex is a fully managed service that provides machine learning and AI solutions within the Data Cloud, including a diverse model catalog from Hugging Face, Lamini, Microsoft Azure, the NVIDIA API Catalog, Perplexity, and Together AI. Arctic usage will also be enabled on Amazon Web Services (AWS) in the future.
◇ Exceptional resource efficiency and top-tier performance
Snowflake’s AI research team comprises leading researchers and system engineers. The construction of Arctic took less than three months. Amazon Elastic Compute Cloud (Amazon EC2) P5 instances were used for model training, and training costs were only one-eighth of comparable models. Snowflake is setting a new benchmark for training cutting-edge open enterprise-grade models, ultimately enabling users to generate cost-effective and customized models at the optimal scale.
Arctic’s differentiated MoE design enhances both the training system and model performance through carefully designed data composition tailored to enterprise requirements. It also achieves exceptional token efficiency and industry-leading quality by activating 480 billion parameters 17 at a time, yielding optimal results. For groundbreaking efficiency improvements, Arctic activates about 50% fewer parameters than DBRX and around 75% fewer parameters than Llama 3 70B during both inference and training. It also surpasses the performance of leading existing open-source models like DBRX and Mixtral-8x7B in coding (HumanEval+, MBPP+) and SQL generation (Spider), while exhibiting top-tier performance in general language understanding (MMLU, Massive Multitask Language Understanding).
◇ AI innovation for everyone, driven by Snowflake
Snowflake provides the data foundation and cutting-edge AI building blocks needed for businesses to leverage their data and create practical AI/machine learning applications. With Snowflake Cortex, customers can readily build production-ready AI applications at the appropriate scale while remaining within the security and governance boundaries of the Data Cloud when using Arctic.
The Snowflake Arctic model family, which includes the Arctic LLM, also features Arctic embed, one of Snowflake’s recently released state-of-the-art text embedding models. This product is also open-sourced under the Apache 2.0 license and available for free use within the open-source community. This family of five models is available directly from Hugging Face and will be embedded in Snowflake Cortex through a private preview. With approximately one-third the size of comparable models, these embedding models are optimized for optimal search performance, providing a cost-effective solution for businesses when combining their unique datasets with LLMs as part of Retrieval Augmented Generation (RAG) or semantic search services.
Snowflake continues to support customers’ access to the latest, highest-performing LLMs within the Data Cloud, including recent additions of models from Reka and Mistral AI. Furthermore, Snowflake continues to drive AI innovation through its recently announced expanded partnership with NVIDIA. The Snowflake Data Cloud, combined with a full-stack NVIDIA-accelerated platform, provides a secure and powerful infrastructure and computing capacity that enables organizations across industries to fully leverage the productivity of AI. Snowflake Ventures has also invested in Landing AI, Mistral AI, and Reka to help customers unlock the value of LLMs and AI from their own enterprise data.
Comments0