Rasa Stack on Google Cloud Platform


(Audun Gulbrandsen) #1

Hi! I am new to Rasa and Google Cloud Platform, and was wondering if anybody know were I can find a good tutorial on how to install Rasa Stack on Google Cloud Platform (GCP) ?

Is it recommended to install via docker? Should one use AppEngine, Compute Engine or Kubernetes?

(Pierre Paul) #2

We use docker + Compute Engine. We have build trigger on the master branch that creates a new docker build then pushes it to the Compute instance (via gcloud beta compute instances update-container).

(Palash Vishwas Karmore) #3

What if I have an in-house server instead of Compute Engine. Do you have any suggestions for that?

(Pierre Paul) #4

As a matter of fact, I do. Make sure your server CPU supports the AVX instruction set (cat /proc/cpuinfo | grep -i avx). Rasa ships (by default anyway) with tensor flow >= 1.7 which is precompiled with the AVx instruction set. You can compile tensor flow manually, but if you use docker it becomes even more complicated. My servers from 2010 don’t supports AVX (Dell r710).

(Pierre Paul) #5

If your servers don’t support it, you will get “Illegal instructions” error and crash. You can test with python -c "import tensorflow".

If your servers supports it, I highly recommend that you look into dokku. It also works great with or without docker.