Mozilla-Ocho / llamafile

Distribute and run LLMs with a single file.
https://llamafile.ai
Other
19.49k stars 985 forks source link

Coral Support #40

Open dipseth opened 10 months ago

dipseth commented 10 months ago

Is there anyway to run this using a CoraTel USB Accelerator?

stlhood commented 10 months ago

This is an interesting device, but my understanding is that it has very little onboard memory, which means that it's unlikely to be able to run even the smaller LLMs. Apparently it also lacks several key operations needed to run these models. Some further reading:

jart commented 10 months ago

Can it at least run Rocket 3b? Since Rocket runs great for me on a similarly priced $60 raspberry pi 4.

TomatoCo commented 9 months ago

The coral TPUs have only 8 megabytes of memory. Their main advantage is that they do about 4 trillion operations per second while drawing one watt.