Setup p0 on your on-prem server on Google Cloud Provider (GCP)
Welcome to the on-premise installation instructions for p0 on Google Cloud Provider (GCP). Follow this guide to get set up with your own version of p0.
Create service account with required permission
Ensure that you have gcloud CLI (Google Cloud command line interface) installed, authenticated, and configured with the correct project ID.
Refer to the CLI Management Documentation for installation and setup instructions.
Create a service account using either:
gCloud CLI (see command below)
Important notes
Ensure the service account has access to
Vertex API
to use Claude 3 Haiku model. Please refer to this guide for configuration details.New Google Cloud accounts receive free credits, but a billing account is required to use the Claude 3 Haiku model.
To create a service account using gCloud CLI: (Replace
p0-vertex-ai-sa
with your preferred service account name, if needed.)
Assign required roles to the newly created service account by replacing
<PROJECT_ID>
in the following command:
This grants roles/aiplatform.user
, which is required for accessing your on-premise LLM from p0 instance.
Import p0 image
On the Google Cloud Console, select the project where you want to set up the p0 application.
In the left sidebar, select
Migrate to Virtual Machines
Navigate to the
Targets
tab and add your project for this API.Go to the
Image Imports
tab and click theCreate Image
button.
Fill in the following details:
Image Name:
imported-p0-dashboard
Source Cloud Storage File:
public-p0/p0-dashboard.vmdk
Region: Select your preferred region
Target Project: Choose the project created in Step 1
Click
Create
Wait approximately 20 minutes for the image import to complete.
Create p0 instance
Once the p0 image import is complete, go to
Compute Engine
>Storage
>Images
: https://console.cloud.google.com/compute/images?tab=images
You should see the imported image. If not, search for the image name from the previous step:
imported-p0-dashboard
.Once you locate the image, click the three-dot Action menu next to it, then select
Create Instance
Instance Configuration Page
will open. Put in the name of your choice for the machine.
Under
Machine Configuration
, select your preferred region and zone.In the
Machine Configuration
section, choose:E2
>Standard
e2-standard-4
(4 vCPUs, 2 cores, 16 GB memory)
(Optional) Under
OS and Storage
, if you have a large number of repositories where 50GB won’t be enough, add a new disk or proceed with the default 50GB.In the
Networking
section, underFirewall
, checkAllow HTTP and HTTPS traffic
.
Under
Security
>Identity and API Access
, select the service account created in the previous steps.
(Optional) Under
Security
>Manage Access
, enableBlock project-wide SSH keys
for a more controlled experience.In the
Advanced
section, underProvisioning Model
, selectStandard Instances
(avoid Spot instances, as they can be terminated during runs).
Click
Create
at the bottom of the page.In about 10 minutes, visit the external IP of the newly created instance to begin onboarding.