Links

Installation in GCP

Easily install Giskard in GCP
Installing Giskard in GCP enables you to inspect & test models that you created in the GCP environment (Workbench, Vertex AI, etc.). Here are the 3 steps to install Giskard in a new VM instance in GCP:

1. Create a Giskard VM Instance in GCP

  1. 1.
    Go to VM instances in Compute Engine and create a VM instance
  2. 2.
    In the configuration of your VM :
    1. 1.
      We recommend you choose at least an e2-standard-2 (2vCPU, 8GB memory)
    2. 2.
      Choose Allow full access to all Cloud APIs
    3. 3.
      In the firewall section, allow HTTP and HTTPS traffic
  3. 3.
    Connect to your VM in SSH by opening a browser window
  4. 4.
    Create a firewall rule to open ports 19000 and 40051 port of the Giskard instance. Here is the command line that you can execute in the terminal opened by your SSH connection:
gcloud compute firewall-rules create giskard-main --allow tcp:19000
gcloud compute firewall-rules create giskard-worker --allow tcp:40051
Make sure you have the proper rights to open a port. If not contact your GCP administrator.
Creating the firewall rules can also be done through UI in the VPC Networksection:
  • Go to the firewall in VPC Network section of GCP
  • Click on create a firewall rule
  • In Targets, select All instances in the network
  • In Source filter, choose IPv4 ranges
  • In source IPv4 ranges, select 0.0.0.0/0
  • In Protocols and ports, select Specified protocols and ports
  • Then select TCP, and type 19000
  • Repeat the same steps to open port 40051

2. Install Giskard in the GCP VM

  • Installation of the Giskard requirements (git and docker)
sudo apt install git
curl -fsSL https://get.docker.com -o get-docker.sh
sudo sh get-docker.sh
  • Installation of Giskard
git clone https://github.com/Giskard-AI/giskard.git
cd giskard
sudo docker compose up -d --force-recreate --no-build

3. Connect to your instance and start uploading ML model

  • Get the external IP address of your Giskard VM in the VM instances section of the Compute Engine
  • Go to http://<your IP address>:19000 in your web browser
You can stop the instance and restart it when you need to save your GCP compute costs. However, note that
  • the IP address will not necessarily be the same. So make sure you copy it again when it's launched.
  • you will need to re-run docker, by executing in the Giskard folder:
sudo docker compose up -d --force-recreate
  • The user id is admin and the password is admin
That's it, you are now ready to use Giskard in GCP! Now you can start uploading a model! To do that in GCP, you can use a workbench notebook, for example!