Errors installing Tensorflow and RASA on Raspberry Pi 4

Im trying to install RASA on a Raspberry Pi using the script from " Nov '19

I had to remove dopamine from the tensor2tensor file as suggested in the github issue.

I’ve posted a modified install script as a gist and I now have Rasa running on a Raspberry Pi 4! "

I get as far as installing the tensorflow sudo pip3 install tensorflow-2.1.0-cp37-cp37m-linux_armv7l.whl

and get this error ERROR: tensorflow-2.1.0-cp37-cp37m-linux_armv7l.whl is not a supported wheel on this platform.

I have tried 2.2.0 and 2.4.0 with the same response

Im not sure as to where to go from here, and help would be aprechiated

Further to this attempt, I have an i5 gen 7 intel NUC with windows 10 and anaconda with python 3.7 which I have had several attempts at installing rasa on, all ending in lockups or version errors or the install I have running currently which is now been going 4 10+ hours scrolling through the components saying requirement already satisfied

Update I stopped the windows 10 version and restarted the install for the third time. This time it is working I can get into the console but the responses are very slow, at least a second or more wih the base data and story.

Would this run on the Raspberry pi 4 fasted than on the I5 NUC?

Hi @channe, Tensorflow unfortunately is causing a lot of issues when it comes to installing it on smaller devices like Raspberry Pi. @JulianGerhard I wonder if you have any input in this issue?


As I’m not a native Linux user I relied on other users attempts at getting RASA working on the Pi. The issue I found was one of incompatibility with tensorflow causing rasa to not install. I did try 3 versions of tensorflow 2.1, 2.2 and 2.4 all with anaconda with python 3.7.0 to overcome this before I stopped this and went onto the NUC.

I also found this with the Windows 10 NUC that I am currently using and it was only by trial and error that I got it going.

Ideally I would prefer to use the Pi to cut the power usage down but am prepared to stay with the NUC if this is all that I can use

I was wondering if there is a table of know good configurations over time for different hardware and software, even if it is a little bit older I could build it on the NUC and copy over to the Pi.

Look forward to hearing back from you