|
|
|
|
|
|
by josho
16 days ago
|
|
|
> It’s really disappointing to see the on-device models being limited to so few devices. At first I thought it was the usual planned obsolescence. Then I realized it may be a true technical limitation. I suspect an embedding model is required to run on device in order to make several of the features work. Embedding models are small compared to LLMs, but, depending on their capabilities, could be the memory driver. |
|
|
It's shortsightedness on Apple's part, being both the first to the game, and the last to the game. Go back any number of years in the past decade. For at ;east 10 years Apple was telling everyone how great their local models are, and how amazing desktop-class their "neural" chips are. And yet here we are with them somehow incapable of running anything except on a few recent top-tier models.