Infrastructure

The models used in modern projects are the same as those used in classic projects. However, they are significantly enhanced to benefit from our upgraded model training and infrastructure service. These improvements ensure efficient resource usage and cost-effectiveness by granting the capability to load and unload models. As an additional upgrade, we now have deployments powered by GPUs, which offer noticeably superior performance than CPUs.

Modern projects also use sophisticated algorithms for balanced model training and deployments, resulting in faster training periods and more efficient resource usage. An automatic training session is triggered when required, offering quick performance feedback. Model deployments, powered by GPUs, can automatically scale according to demand, eliminating the necessity to adjust replicas or their sizes.

In terms of performance, each account can process up to 10,000 pages per hour. This limitation can be lifted on request.

There are no AI units charged when training and serving models. This means you can freely train models to improve performance, develop new capabilities, and experiment without concerns about incurring additional AI units charges. For more information on licensing, check the Metering and charging logic page.