search

koenvervloesem / rasa-docker-arm

Rasa Docker image for ARMv7. Runs on a Raspberry Pi.
https://hub.docker.com/r/koenvervloesem/rasa
MIT License
19 stars 2 forks source link
arm armhf armv7 armv7l bot bot-framework bots chatbot chatbots chatbots-framework docker docker-image machine-learning natural-language-processing nlp nlu rasa raspberry-pi raspberry-pi-4 raspberry-pi-4b

readme

Rasa Docker image for ARM

Continous Integration GitHub license

Rasa is an open source machine learning framework to automate text- and voice-based conversations. It doesn't have an official Docker image for ARM, so this project builds such an image (for the linux/arm/v7 architecture) that you can run on a Raspberry Pi.

The resulting image stays as close as possible to Rasa's official Docker container. The only changes are added pre-built wheels for TensorFlow and TensorFlow Addons, some extra dependencies installed and the user ID changed to 1000 to align with the pi user on Raspberry Pi OS (Raspbian).

System requirements

If you manage to build the Docker image on another version of Raspbian, another Linux distribution, another model of the Raspberry Pi or even another ARM computer or cross-build it on another architecture, please let me know so I can generalize the script.

Usage

First make sure to have Docker installed:

curl -sSL https://get.docker.com | sh 
sudo usermod -aG docker pi

After this, log out and log in again.

Then, if you want to build a Docker image of the latest Rasa version supported by this project (which is stored in the file RASA_VERSION), just run:

make docker

If you want to build an image for a specific Rasa version, run the build script with the version number as an argument:

./scripts/build_docker.sh 1.10.5

The build script downloads the version's archive, patches it and builds the Docker image.

If all goes well, at the end you should be able to run the Rasa Docker container on your Raspberry Pi, for instance with a Docker Compose file:

docker-compose up -d

After this, you can create a Rasa project with:

docker-compose run rasa init

Patching other Rasa versions

If you want to build a Docker image for another Rasa version than the ones supported by this project, just copy the unpacked Rasa archive, make your changes, create the patch and put it in the directory patches with the version number in the file name. Then the build script will pick this up. For instance:

wget -O rasa-1.10.5.tar.gz https://github.com/RasaHQ/rasa/archive/1.10.5.tar.gz
tar xzf rasa-1.10.5.tar.gz
cp -r rasa-1.10.5 rasa-1.10.5-2
# Make your changes in the `rasa-1.10.5-2` directory
diff -ruN rasa-1.10.5 rasa-1.10.5-2 > patches/rasa-1.10.5-arm.patch
./scripts/build_docker.sh 1.10.5

If you have successfully built the release, please contribute your patch to this project by a pull request so others can benefit from it too.

Motivation

Someone on the Rhasspy forum failed to install Rasa on his Raspberry Pi 4, and I became curious. How difficult could it be? Well, quite difficult, as it turned out to be.

I had to build a wheel for TensorFlow Addons on ARM, which was quite a challenge. To build this wheel, I needed to build Bazel on ARM. After these slight detours, I finally managed to build a working Docker image for Rasa on ARM.

Docker image on Docker Hub

You don't have to build this Docker image yourself, you can find it on Docker Hub as koenvervloesem/rasa. Download the latest version with:

docker pull koenvervloesem/rasa

If you want to download a specific version, use:

docker pull koenvervloesem/rasa:1.10.5

References

When I was searching for a way to run Rasa on a Raspberry Pi, I encountered the following projects, which I didn't use for various reasons but they gave some helpful background information:

License

This project is provided by Koen Vervloesem as open source software with the MIT license. See the LICENSE file for more information.