{"id":1129,"date":"2023-03-02T14:53:32","date_gmt":"2023-03-02T14:53:32","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/startups\/blog\/?p=1129"},"modified":"2024-11-04T12:02:40","modified_gmt":"2024-11-04T20:02:40","slug":"making-generative-ai-training-simple","status":"publish","type":"post","link":"https:\/\/www.microsoft.com\/en-us\/startups\/blog\/making-generative-ai-training-simple\/","title":{"rendered":"Making generative AI training simple and cost-efficient with PeriFlow and Azure"},"content":{"rendered":"\n

Byung-Gon Chun is CEO and founder of FriendliAI, which helps businesses innovate with their generative AI models. He is also a professor at Seoul National University.<\/em><\/p>\n\n\n\n

Despite the potential of generative AI, its use involves considerable barriers for startup developers. Developers are thrown into the task of training and serving their models, setting up necessary resources and environments, and handling faults and performance problems. While all these steps are indispensable, they are also incredibly time-consuming and costly. This is where FriendliAI comes in \u2013 it helps streamline the process by allowing startups to build and serve their generative AI model using the cloud to meet their requirements quickly and efficiently.<\/p>\n\n\n\n

Building FriendliAI<\/h2>\n\n\n\n

With OpenAI\u2019s release of GPT3 in 2020, I immediately saw an immense opportunity to use my experience in highly scalable machine learning systems. With a decade of experience in distributed computing technologies, both as a professor of Computer Science and Engineering at Seoul National University and as a Principal Scientist at Microsoft, I was excited to explore new opportunities in this realm. Furthermore, as the active commercialization of generative AI began flourishing in different industries, it became evident that advancements in distributed AI computing technologies would also be necessary.<\/p>\n\n\n

\n
\"FriendliAI<\/figure><\/div>\n\n\n

<\/strong>Two years ago, I founded FriendliAI to provide startups with easy access to innovative generative AI. We first helped realize this mission by introducing PeriFlow \u2013 a solution to train and serve generative AI models effortlessly with remarkable training and inference speeds and minimal costs. Our ambition has taken us on an incredible journey. We have a unique position within the market as a provider specialized in generative AI training and serving. As a result, we are already partnering with multiple clients to develop and serve cutting-edge generative AI applications powered by PeriFlow.<\/p>\n\n\n\n

Fast, Convenient, and Cost-Efficient Ways to Serve and Train Generative AI Models<\/h2>\n\n\n\n

PeriFlow is an automated platform that handles the ML pipeline from training to serving AI models. It has successfully lowered technical barriers and high costs associated with generative AI services as it not only enables the training and inference process to run smoothly and cost-efficiently but also allows for automation in areas such as resolving faults and performance problems – which would usually require tedious manual monitoring. With PeriFlow, companies, especially , can rest assured that their AI workflow will run at peak efficiency on the cloud; their generative AI inference becomes up to tens of times faster than the conventional inference .<\/p>\n\n\n

\n
\"Byung-Gon<\/figure><\/div>\n\n\n

FriendliAI is a leading example of how Azure can assist startups in their growth and success. Through Azure\u2019s AI infrastructure and the PeriFlow platform, users can effortlessly transition cloud services without sacrificing performance or reliability when scaling generative AI models from the laptop to the cloud. Countless organizations rely on the services that Azure offers – proving its immense value within the Microsoft for Startups Founders Hub.<\/p>\n\n\n\n

On top of that, PeriFlow helps users find the most optimal combination of generative AI models and cloud resources in a much simpler way allowing users to scale their models in a hassle-free manner. With these features, startups can confidently leverage PeriFlow’s training and serving capabilities while taking advantage of significantly higher request throughput at minimum costs.<\/p>\n\n\n\n

Enabling Human-like Communication at Minimal Costs<\/h2>\n\n\n\n

One customer success story includes ScatterLab\u2019s Luda Lee 2.0, an Open-domain Conversational AI service that has made a major splash in the technology industry, rising to the top spot of both Korea\u2019s Google Play and Apple App Store. What makes Luda Lee 2.0 extraordinary is its impressive contextual prowess\u2014this chatbot can recognize nuanced semantic differences and respond naturally to buzzwords or jokes, prompting even philosophical replies like “Everyone is precious regardless of race.” With the ScatterLab team\u2019s innovation and the PeriFlow platform\u2019s features, the impressive chatbot now services a 17X larger model with remarkable speeds and minimal costs. ScatterLab has set a new gold standard for Open-domain Conversational AI services in Korea.<\/p>\n\n\n

\n
\"PeriFlow<\/figure><\/div>\n\n\n

Choosing the Perfect Cloud Environment for an AI Platform\u00a0 <\/h2>\n\n\n\n

Cloud computing is driving innovation and transforming the industry. With the PeriFlow Platform leading generative AI innovation through Azure cloud, users can retain performance, connectivity, and flexibility guaranteed on a scalable basis. These are some of the Azure services that PeriFlow has leveraged to offer secure, reliable results:<\/p>\n\n\n\n