Model Deployment¶
ModelArts is capable of managing models and services. This allows mainstream framework images and models from multiple vendors to be managed in a unified manner.
Generally, AI model deployment and large-scale implementation are complex.
Figure 1 Process of deploying a model¶
The real-time inference service features high concurrency, low latency, and elastic scaling, and supports multi-model gray release and A/B testing.
ModelArts is optimized based on the high-performance AI inference chip Ascend 310. It can process PBs of inference data within a single day, publish over 1 million inference APIs on the cloud, and control inference network latency to milliseconds.