Selectel ML platform

The main articles are:

• PaaS - Platform As A Service
• Machine Learning

2023: Presentation of the ML platform

Selectel on September 22, 2023 introduced the ML platform, a cloud-based solution with pre-configured hardware and software components designed to train and deploy ML models.

The platform is deployed individually for each client and can include popular open source ClearML or Kubeflow tools that allow you to organize a full cycle of training and testing ML models. The infrastructure solution is deployed on the basis of the Managed service Kubernetes with GPU, data sets and experimental results are stored in the S3 object storage, and images in the Container Registry. In this solution, Selectel is used video cards Nvidia with a large amount of memory - T4, A2, A30, A100, A2000, A5000.

The platform allows customers to significantly simplify the process of training and testing ML models, as well as speed up the time-to-market of their solutions. This is possible by delegating infrastructure tasks to the provider. In addition, the use of the platform allows customers to reduce the cost of training ML models by dividing one GPU between different tasks and paying only for consumed resources. The solution will help users standardize and unify workflows with a single infrastructure, platform entry point, artifact storage, and experiment management tools.

The Russian AI market is actively developing: according to MIPT, in 2022 it grew by 17%. This also reflects the growth in demand for specially prepared infrastructure for training and testing ML models, said Anton Chunaev, manager of ML products at Selectel. - The platform was tested during the year - our goal was to take into account feedback from users and bring to the market a solution that will allow customers to think less about infrastructure issues and concentrate on their business tasks.