Community guidelines
Be specific and constructive. No vendor spam — promoting your own product belongs in a listing. Anyone can read; posting needs a free account.
product wants text summarization to run on device so user notes never leave the phone, using Apple's Foundation Models framework on ios and Gemini Nano through AICore on android. The model fits on newer devices, but startup time, memory and battery are rough on older ones. How are apps deciding when to run locally and when to fall back to cloud?
we use a capability check and offer local processing only on devices that meet memory and acceleration requirements. Cloud remains an opt in fallback. The worst experience is pretending every supported phone can run the same model
Measure energy, not only latency. A task that finishes in two seconds can still spike power and heat.
do you download the model after install? Bundling it makes the app enormous, but a one gigabyte download on first use isn't friendly either
we download on demand over Wi Fi with clear size and storage information, then allow deletion from settings. Version the model separately from the app. That lets you patch it without a full store release and avoids shipping a huge asset to users who never touch the feature
Also test on a device that's already warm from something else - camera, navigation, whatever. The thermal throttling numbers from a cold, idle test device are optimistic compared to what a user's phone actually looks like mid afternoon.