When using a deep learning model in the browser or on mobile, the model needs to be downloaded on the device. It needs to be as lightweight as possible for the following reasons:
- Users are often using their phone on a cellular connection that is sometimes metered.
- The connection can also be slow.
- Models can be frequently updated.
- Disk space on portable devices is sometimes limited.
With hundreds of millions of parameters, deep learning models are notoriously disk space-consuming. Thankfully, there are techniques to reduce their size.
