- SK C&C, 고객 맞춤형 sLLM 구현 지원 플랫폼 ‘솔루어 LLMOps’ 선보여 > 뉴스 - 스타트업 커뮤니티 씬디스
- 챗GPT, 클로바X 등 상용 LLM, 오픈소스 LLM이 제공하는 다양한 파운데이션 모델 활용데이터수집∙전처리∙학습∙테스트∙서빙 등 sLLM 제작…, 스타트업에 종사하시는 여러분들의 놀이터 씬디스는 스타트업 커뮤니티 입니다.
Leveraging various foundation models offered by commercial LLMs like ChatGPT and Clova X, as well as open-source LLMs.
Implementing hyperautomation across the entire sLLM creation process, including data collection, preprocessing, training, testing, and serving.
Providing hallucination (Hallucination) elimination and sLLM testing functionalities.
Enabling rapid sLLM generation through drag-and-drop and ensuring sLLM reliability through continuous data parallel learning.
An era has dawned where businesses can self-implement customized AI services using sLLMs (small language models) freely combined from diverse foundation models.
While smaller than LLMs (Large Language Models) like ChatGPT, HyperCLOVA X, and Gemini, sLLMs can be trained to specialize in specific domains. They consume significantly fewer computing resources, reducing development costs and enhancing security. Utilizing sLLMs allows for the efficient construction of enterprise-exclusive AI services.
SK C&C (CEO: Yoon Phong-young, skcc.co.kr) announced today that it will be providing 'Solur LLMOps' (hereinafter referred to as 'Solur LLMOps'), a platform designed to assist businesses in easily and quickly implementing customized sLLMs (small language models).
SK C&C has incorporated the accumulated knowledge from generative AI foundation model application strategies, enterprise data integration and learning expertise, and the entire process of creating customized sLLMs for enterprises into 'Solur LLMOps' based on its experience in building and operating generative AI services with various clients across diverse industries such as finance, manufacturing, telecommunications, and services.
Foundation models learn from structured and unstructured data to support a variety of AI tasks, including language comprehension, text and image generation, and natural language dialogue.
First and foremost, 'Solur LLMOps' supports the combination and utilization of various external foundation models.
In addition to commercial LLMs such as OpenAI's ChatGPT and Naver Cloud's HyperCLOVA X, a wide range of foundation models built using open-source LLMs can also be leveraged. It recommends, selects, combines, and utilizes the necessary foundation models to build sLLMs tailored to the specific AI characteristics of each enterprise.
Particularly, hyperautomation is applied across the entire process, including data collection and preprocessing, automatic learning data generation, training using external foundation models, sLLM generation and testing, to significantly enhance efficiency and reduce costs in the process of building customized sLLMs for enterprises.
In practice, 'Solur LLMOps' automatically generates training data by preprocessing unstructured data while simultaneously collecting the data held by the enterprise. Subsequently, it utilizes the selected external generative AI foundation model to conduct rapid training and completes the creation of an sLLM aligned with the enterprise's business objectives. Furthermore, it offers an architecture that optimizes limited resources through serverless learning resource management.
'Solur LLMOps' also offers hallucination (Hallucination) elimination and sLLM testing capabilities to address the issue of AI generating incorrect answers.
It provides AI automated tools for handling hallucinations at each stage of data preprocessing, model generation, evaluation, and utilization, enabling even non-experts to confidently create sLLMs.
Through the 'Solur LLMOps' chat window, users can conduct completion tests with simple questions, and the platform automatically provides questions needed for testing.
Furthermore, 'Solur LLMOps' offers a user-friendly interface (UI) and experience (UX) for rapid sLLM model generation and reliability assurance tailored to the enterprise's operational environment.
Users can repeatedly perform tasks from data refinement to model tuning and testing through a simple drag-and-drop (Drag & Drop) method by selecting data on the web screen. They can easily conduct multiple data processing and simultaneous parallel learning using multiple foundation models with just mouse operations.
This empowers on-site business professionals to easily create sLLMs tailored to their specific tasks and leverage them for various operations.
Cha Ji-won, Head of SK C&C's G.AI Group, stated, “The core features embedded within Solur LLMOps have already been deployed in generative AI development projects across various industries, including financial institutions and SK affiliates.” He further added, “We will actively endeavor to facilitate the widespread adoption of customized sLLMs for enterprises across all industries in Korea, starting with these initial implementations.”
Website: http://www.skcc.com
Contact
SK㈜ C&C
Public Relations Team
Ji Yun-jin, Manager
Comments0