AI Judge changed title after evaluation, original title: "Apple's new AFM Cloud Pro server model is built on Google's Gemini foundation and training data"
The models dynamically activate 1-4B parameters per inference
"AFM Cloud Pro seems to be based on Gemini foundation and data, but Apple did their own pre-training, post-training, RL, etc"
Google shared its training data and training recipes DAYUM
The technical details behind Apple Foundation Models are worth a read:
https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models
Parallel Track MoE Apple released it last year in July
It's very cool that Apple shipped a 20B parameter on-device.
You can't put 20B parameters in RAM at any reasonable precision. To make it work they are using pretty exotic architecture by today's standards.
A small model predicts from the query (or prompt) which experts to load from Nand into RAM. The key distinction from a typical MoE is that you do this once per query and then generate all the tokens with the same experts (instead of switching the experts for every token).
Blog post: https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models
Original publication: https://machinelearning.apple.com/research/pruning-large-language
It's very cool that Apple shipped a 20B parameter on-device.
You can't put 20B parameters in RAM at any reasonable precision. To make it work they are using pretty exotic architecture by today's standards.
A small model predicts from the query (or prompt) which experts to load from Nand into RAM. The key distinction from a typical MoE is that you do this once per query and then generate all the tokens with the same experts (instead of switching the experts for every token).
perfect reporting, 5 grafs i can skim to quickly understand what the hell is going on https://www.macrumors.com/2026/06/08/apple-reveals-new-ai-architecture/
This is cool... I wonder if they will open up developers access to this

@zephyr_z9 My Internal Plan!📈
⬇️

@zephyr_z9 My strategy plan.
🔻↩️↩️

@zephyr_z9 My strategy plan.
🔻↩️↩️

@zephyr_z9 I share my real-time TRADE alert (entry & exit points) on WhatsApp, free to join ✅!!! 🔽 👉 🔗: https://api.whatsapp.com/send/?phone=12242760576&text=Strategy
➡️Copy search input Reply "777" to WhatsApp: + 12242760576

@zephyr_z9 I share my real-time TRADE alert (entry & exit points) on WhatsApp, free to join ✅!!! 🔽 👉 🔗: https://api.whatsapp.com/send/?phone=12242760576&text=Strategy
➡️Copy search input Reply "777" to WhatsApp: + 12242760576

@zephyr_z9 I will share my detailed trading plan (including entry and exit points, investment analysis, etc.) on WA. This might be helpful to you. Get it for free!
👉Copy and reply with "TRADING PLAN" to my WA to get it for free👉+17869786054
My WA link:http://wa.me/17869786054/?text=TRADING

@zephyr_z9 I will share my detailed trading plan (including entry and exit points, investment analysis, etc.) on WA. This might be helpful to you. Get it for free!
My WA link:http://wa.me/17869786054/?text=777

@zephyr_z9 I will share my detailed trading plan (including entry and exit points, investment analysis, etc.) on WA. This might be helpful to you. Get it for free!
👉Copy and reply with "TRADING PLAN" to my WA to get it for free👉+17869786054
My WA link:http://wa.me/17869786054/?text=TRADING
AI Judge changed title after evaluation, original title: "Apple's new AFM Cloud Pro server model is built on Google's Gemini foundation and training data"
The models dynamically activate 1-4B parameters per inference