According to the developers, the platform allows creating "any" trained ML models, large language models (LLM), and computer vision (CV) models. With Inference Valve, they can be implemented on the existing infrastructure, connected to the company's IT systems via standard APIs, scaled, as well as updated and monitored.
CV models can automatically analyze video, finding specific objects, scenes, and actions, or classify medical images. ML and LLM models will be able to generate and structure texts, analyze data, create descriptions, predict marketing and sales metrics, and answer frequently asked questions in HR portals. The platform can also be used to deploy voice models for speech synthesis and analysis, as well as to implement them in call centers.
Companies can deploy both their own trained AI models and use ready-to-use open-source models. The platform is available in a private cloud on the MWS Cloud infrastructure, on-premises on the customer's servers, as well as as part of software and hardware complexes (PAK) in a closed loop, including modes with limited access to external networks.
The platform supports simultaneous operation with multiple models with allocation of computing resource quotas, version control, traffic routing between versions, and scaling under load on both GPU and CPU.