Rasa NLU docker installation fails?

I tried to install Rasa NLU following the instructions in the docs but after downloading the files I get this error standard_init_linux.go:211: exec user process caused "exec format error"

I also tried to install Rasa follwing the instructions on their website but that also fails when trying to install tensorflow.

You’re probably trying this on a Raspberry Pi? The message "exec format error" means that Docker is trying to run an image for another architecture. If you look at the OS/architecture of the rasa image on Docker Hub, it only lists linux/amd64.

Yes that’s right, I am trying to install on a PI 4. I haven’t been able to install Rasa on a PI yet, I have tried several instructions pages on gitgub which claim were sucessful but everything failed so far.

Do you remember which ones you have tried? I’m giving it a go now, I’ll let you know if I succeed.

Sorry for teh slow reply, apart from the docker install in teh rhasspy docs I tried this link and specifically these instrauctions. The install fails when trying to install scipy, but I got a workaround doing this:

# standard update

sudo apt update && sudo apt upgrade -y

# install latest pip

curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py

sudo python3 get-pip.py

# install dependencies

sudo apt install python3-dev libatlas-base-dev libhdf5-dev libblas-dev liblapack-dev gfortran -y

sudo pip3 install setuptools --upgrade --ignore-installed

sudo pip3 install wrapt --upgrade --ignore-installed

sudo pip3 install botocore

sudo pip3 install scipy==1.3.3

Then run the install command for Rasa again.

I gave up when the installation of the spacy’s models failed…

I also tried this link in the past which failed

This is really a nightmare. I got really far, even managed to install TensorFlow from a wheel built for ARMv7, but I can’t find a wheel for tensorflow-addons, so I have to build it manually, but that requires Bazel, which doesn’t have a package for Raspbian, and off we go… Currently building Bazel from scratch after having to install a whopping 1 GB JDK…

I already had to switch to a bigger microSD card and to the 2 GB Raspberry Pi 4 because 8 GB storage and 1 GB RAM wasn’t enough.

If this all works out, I’ll publish a Rasa Docker image for ARMv7.

OK, I managed to build a Rasa Docker image for ARMv7, and it successfully trained the demo app. Training is really slow on my Raspberry Pi 4, inference is fast:

I’m not sure this will be usable. Rhasspy probably times out before Rasa has finished training.

In making this work I had to resort to some ugly hacks and magic spells and I have seen errors that I never want to see again, but I’ll clean this up and publish a Docker image and a build script to reproduce this.