Shipping on-device LLMs on iPhone: 5 lessons from benchmarking 9 models
What we learned running 9 open LLMs on an iPhone 17e: sustained GPU load, a KV-cache bug that skewed results, token limits, and choosing a model per device.
Blog
Notes from the workshop.
What we learned shipping real products: on-device AI, hardware and software, and the details that never make the launch tweet. Written by the people who did the work.
Shipping an on-device LLM in a consumer iPhone app
The model menu, RAM gates, CDN delivery, checksum verification, and crash-loop protection. Notes from shipping Ragondin Assistant.