Where will the model live: in the cloud or on the device?πŸ“± ☁️

Nadia Zueva
2 replies
πŸ“± Device No internet needed, real-time support, no server costs, input data is not sent to the cloud (which often involves a third-party). However, keep in mind that if you have a complex, tricky model architecture, which also maybe includes preprocessing, it will likely not be easy to port on a device. In such a case, all the logic with conditions will need to be rewritten in tensor arithmetic, which is not just difficult, but not always possible. Here, you can get stuck and waste weeks. But if you manage it, you can show off how you rewrote conditions with tensors πŸ’ͺ ☁️ Cloud You can use a larger model, easily update it for all users, scale it. Today there are many convenient services that are privacy-certified and bill on-demand. If 4 years ago you had to set up everything by yourself, now deployment is quite easy and less costly! I also published a decision tree whether cloud or device is better for your case here: https://shorturl.at/BFHU0 What's your thoughts on where to host a model?

Replies

α—°α—©α™­ ᒍ.
Nice, good post! And great treeπŸ‘
Share