PyTorch on Mobile Devices
In the previous chapter, we learned extensively about operationalizing PyTorch models as services in production systems. While productionizing machine learning (ML) models as services in the cloud remains the most popular form of ML deployment, several use cases require models to be deployed on mobile devices, such as:
- User data protection – mobile models do not require third-party data transfer, as the processing is done where the data is first acquired
- Reduced latency – mobile models save us the cloud network I/O time
- Better user experience – mobile models can provide real-time user interaction with lower latency compared to running models remotely from the cloud
- Leveraging dedicated mobile hardware and software for ML (eg., coreML) that mobile phone makers are increasingly adding to their products
In this chapter, we will learn how to deploy PyTorch models on mobile devices using PyTorch Mobile...