@made @firefoxwebdevs There's already lots of work for on-device ML: https://searchfox.org/firefox-main/search?q=toolkit%2Fcomponents%2Fml

Integrating models into a finalized product with the wide spectrum of end-user devices is tricky though, so it has to be done with care.

@xela @firefoxwebdevs For on-device, the power usage is on the end-user, and the text in the viewport range is translated. It's heavy CPU work that is quickly finished. So you get short bursts of heavy CPU usage while actively interacting with a translated page. All the page content is private and stays on your machine.

@xela @firefoxwebdevs Translations involve either 1 or 2 models depending on the language pairs. Each language requires a model going into, and out of English. Training a language involves on the order of hundreds of GPU hours, and the largest models probably get into the thousands range. Early models were in the thousands range probably before we optimized things.