Train and run transformers directly on Apple's Neural Engine (github.com)

Can you elaborate on the project? Because I just see a vibe coded project with failing builds and no associated interesting development story or example applications.

[dead]

Premise seems interesting, I stopped reading at

  No CoreML. No per-token recompilation. Just ...

What RAM can be allocated to the NE? I am skeptical this is as easy as the GitHub repo suggests.