Selectel Inference platform

Main article: IT Service Management, ITSM

2024: Inference Platform Beta Launch

[[Selectel] October 10, 2024 announced the start of beta testing of its own Inference platform, designed to simplify the launch]]ML-models for commercial operation and acceleration of their operation. This service will allow you to deploy a ready-made ML model on computing power in a few minutes provider without involving developers.

The solution is relevant for companies in business processes whose products or services part of the tasks is solved by ML models: voice robots for automating work, call center for chat boats consulting employees and customers, image recognition services and other tasks. With the Inference platform, you can load the desired ML models into the Selectel object store, and as a result, get ready-made - API interfaces without development. The solution also makes it possible to regularly update versions of ML models without stopping processing incoming user requests, which avoids downtime and ensures the efficiency and stability of services necessary for large companies.

The use of the Inference platform makes it possible to reduce the time for bringing AI projects to market several times, taking into account specific load profiles and performance requirements. This customization is achieved by involving Selectel ML team specialists in the selection of configurations and infrastructure adjustment.

Selectel has helped many companies build infrastructure for a wide variety of AI and ML tasks. The expertise accumulated in recent years helps to develop existing products, as well as launch new ones. In 2023, the line of our products was replenished with a customizable ML platform for model training, the client experience of which helped to better understand current expectations from platform ML products, - said Anton Chunaev, manager of ML products at Selectel. - The maturity of the Russian AI market is increasing, a large number of ready-made ML models appear, which do not require significant resources for refinement for the specifics of a specific client task - all this leads to an increase in the number of companies that include ML services in their IT landscape. We strive to provide our customers with a large selection of effective solutions that will allow the application of ML models even without experience. Launching the Inference platform is a logical step in this direction.

The platform is based on Managed Kubernetes Selectel GPU clusters and is complemented by the popular open source tools NVIDIA Triton Inference Server, Ray Serve Cluster and Istio Ingress Controller. Thus, the solution is completely independent of the vendor lock (binding to the supplier) and provides high performance with built-in capabilities to increase throughput and maximize resource utilization.