Error while building TensorFlow about tensorflow-on-raspberry-pi HOT 9 CLOSED

Thanks for opening this up- usually, exit status 1 or 4 indicates some sort of out of memory error.

Just to get started, could you confirm that you've installed some sort of USB (or other) memory drive as swap space? If so, what size is it? Additionally, could you use the top command to make sure that the swap is actually being recognized by the device?

from tensorflow-on-raspberry-pi.

Yes, I installed a memory drive of 4GB. I will only be able to access my device tomorrow, I'll check if the swap is recognized and tell you then. Thanks for your help.
EDIT : I checked, the swap is recognized.

from tensorflow-on-raspberry-pi.

I could fix this by editing some source code, it was caused by Eigen imports, but the main problem is TensorFlow 0.9, because you can't really disable GPU support, it requires CUDA for half floats even if you disabled GPU support in config. I'll downgrade to 0.8 for now, and wait for a fix to use 0.9.
I'll open an issue for that on tensorflow's git

from tensorflow-on-raspberry-pi.

Thanks for the info @MrSlayer02 - I'm going to leave this open until there is a documented way to work through this.

from tensorflow-on-raspberry-pi.

I opened an issue on TensorFlow's git , maybe this will be fixed soon ...
But until, you should edit your GUIDE.md to tell to checkout branch r0.8 (by adding -b r0.8 to the clone command or by git checkout r0.8).

from tensorflow-on-raspberry-pi.

I'll add a note in GUIDE.md indicating that there is currently an issue with more recent commits and that the most recent confirmed compilation was at tag r0.8.

from tensorflow-on-raspberry-pi.

It looks like the issue has to do with the Neon instruction flag- the Eigen 16-bit floating point doesn't play nice with Neon, it would seem.

If you still want to build it yourself, try again without the --copt="-mfpu=neon" flag. If you just want 0.9rc0, I'm uploading the binary now!

from tensorflow-on-raspberry-pi.

If you still plan on building from scratch, let me know if you can confirm that removing the -mfpu=neon flag works for you, and I'll close this issue.

from tensorflow-on-raspberry-pi.

Thanks for the pre-release :)
I won't have time to try this out, actually I did this for a school project and finally I'm happy with TensorFlow 0.8, which I could build. I think you can close this issue.

from tensorflow-on-raspberry-pi.