doctr
doctr copied to clipboard
[support] Add an ARM build for edge computing
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!
Any updates on that @fg-mindee ?
My bad, I'll jump back into this this week
what is the status of this? @fg-mindee
@frgfm
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: