I need to fine-tune a model on the iPhone after deployment (in an online manner). However, it seems like both Torch Mobile, and the recently released ExecuTorch are for inference purposes only. My questions are:
1- Is it possible to train a model on edge device (e.g. with LibTorch Lite)?
2- If yes, what is the most straightforward way to do it?
I recently succeeded in building android pytorch to enable learning. There should be a USE_DISTRIBUTE setting, but it was forcibly turned off in case of mobile. That part needs to be corrected. In addition, many source modifications such as sleef and protobuf must be made. In the future, I plan to distribute it to GitHub using a shell script. The iPhone is expected to be similar.