I received an e-mail from the pi shop about a hardware AI accelerator board for the Raspberry Pi.
@cyclicalobsessive
This might be worth a look?
Maybe we can move some of the heavier AI lifting off of the Raspberry Pi itself?
I received an e-mail from the pi shop about a hardware AI accelerator board for the Raspberry Pi.
@cyclicalobsessive
This might be worth a look?
Maybe we can move some of the heavier AI lifting off of the Raspberry Pi itself?
I believe it only works on a RPi 5
Indeed, since it integrates 8GB of memory the HAT+ 2 allows use of a “lesser equipped Pi5”, BUT my tests of the “fit in 8gb LLM” (and those reported in the media) suggest we are far from useful AI on the Raspberry Pi, with or without the HAT+ or HAT+ 2.
Long ago, OpenCV packaged up the useful image processing algorithms into an API with interface standards that really made image processing available on an edge device. Large language models are a bucket of baby alligators quick to bite the hands of those that dip in.
For image processing, using the RPi5 plus a HAT+ 2 is a complicated niche.
Until they can manage what TFLite did to make large image models perform in real time on Raspberry Pi, I’m not touching Large Language Models again.
I suspect that I know less about this than I should, but can something like this help with what the Pi can do? Like maybe unload some of the Tensor Flow stuff from the Pi?
We don’t need to know more - too early, too much hyperbole.
Only if you want the Pi to heat your fish tank.