The service will allow developers to switch between different types of virtual machines without stopping calculations and pay only the calculation time.
Yandex.Cloud has opened access to the Yandex DataSphere service, a cloud environment for using machine learning tools, the company told vc.ru.
The service uses serverless computing technology: when editing and viewing the code, the CPU (conventional processors) or GPU (graphic processors) capacities are not used, and the virtual machine of the desired type is connected only for the duration of direct calculations (training models, launching and other calculations).
This allows the user to pay only the time of real use of computing power, the company notes. While editing and viewing code, random idle machines are not counted.
Yandex DataSphere also supports seamless switching between different types of capacities. Users can use different types of virtual machines without stopping calculations and maintaining progress: economical with a CPU and fast with a GPU.
Yandex notes that this will speed up development, save and increase the efficiency of using more expensive computing power.
Later, the company plans to add a function to save versions of model calculations in three dimensions – data, code, and laptop status. This will simplify the teamwork of data specialists and make ML development a more manageable process for corporate users. Other expected features include code quality checks and recommendations for the optimal use of machines.
While Yandex DataSphere is available in test mode by pre-registration, the service is free until the end of June 2020. After some of the functions may become paid, specified in Yandex. The company will determine the final tariffs by the time of the commercial launch; the company did not specify the date.