[support] Add an ARM build for edge computing

fg-mindee commented 2 years ago

Low-power devices such as Raspberry PIs are widely used by developers to come up with exciting products. Adding customized builds for ARM architectures would greatly help their efforts!

charlesmindee commented 2 years ago

Any updates on that @fg-mindee ?

fg-mindee commented 2 years ago

My bad, I'll jump back into this this week

erdkse commented 1 year ago

what is the status of this? @fg-mindee

felixdittrich92 commented 1 year ago

@frgfm

frgfm commented 1 year ago

Hey everyone :wave:

Sorry, I didn't get the notification until Felix ping! Here are my thoughts on this matter:

Potential struggles to do an ARM build: We need one for PyTorch & one for TensorFlow + some IO-related lib will have an issue I suspect.
How this should be handled: with every release (or even every commit), a build will be needed. On other projects where this was required, we had a Raspberry PI running Github actions to build it for each release & commit. In this case, I cannot allocate my device to do this :sweat_smile: And it can be done on adhoc basis, but it's a long process (I used to be among the first to build PyTorch for ARM, and I can tell you that on a Pi 4 with 4Gb RAM, it was taking between 6 and 8 hours. So here it would be longer to do both the pytorch & tensorflow build :upside_down_face:

That being said @erdkse, if we put up a script to build it on your device when you see fit, would that help? Cheers :v:

mindee / doctr

[support] Add an ARM build for edge computing #430