Skip to content

Commit 80bccd7

Browse files
StevenBucher98StevenBucher98
authored
Adding docs and files for deploying an Azure OpenAI instance via Bicep file (#324)
1 parent d93ef53 commit 80bccd7

File tree

2 files changed

+196
-0
lines changed

2 files changed

+196
-0
lines changed

‎docs/development/AzureOAIDeployment/DeployingAzureOAI.md

Lines changed: 138 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,138 @@
1+
# Deploying Azure OpenAI Service via Bicep
2+
3+
In order to use the `openai-gpt` agent you will either need a public OpenAI API key or an Azure
4+
OpenAI deployment. Due to its additional features and manageability, we recommend using the Azure
5+
OpenAI Service. This document provides easy step-by-step instructions on how to deploy the Azure
6+
OpenAI Service using Bicep files.
7+
8+
There are two things needed to begin a chat experience with Azure OpenAI Service, an Azure OpenAI
9+
Service account and an Azure OpenAI Deployment. The Azure OpenAI Service account is a resource that
10+
contains multiple different model deployments. The Azure OpenAI Deployment is a model deployment
11+
that can be called via an API to generate responses.
12+
13+
## Prerequisites
14+
15+
Before you begin, ensure you have the following:
16+
17+
- An active Azure subscription
18+
- Azure CLI or Azure PowerShell installed
19+
- Proper permissions to create resources in your Azure subscription
20+
21+
## Steps to Deploy
22+
23+
### 1. Getting and modifying the Bicep file
24+
25+
Clone the repository and navigate to the `./docs/development/AzureOAIDeployment` directory:
26+
27+
```sh
28+
git clone www.github.com/PowerShell/AIShell
29+
cd AIShell/docs/development/AzureOAIDeployment
30+
```
31+
32+
You will need to modify the `./main.bicep` file to include your own values. You will have to modify
33+
the parameters at the top of the file.
34+
35+
```bicep
36+
@description('This is the name of your AI Service Account')
37+
param aiserviceaccountname string = '<Insert own account name>'
38+
39+
@description('Custom domain name for the endpoint')
40+
param customDomainName string = '<Insert own unique domain name>'
41+
42+
@description('Name of the deployment')
43+
param modeldeploymentname string = '<Insert own deployment name>'
44+
45+
@description('The model being deployed')
46+
param model string = 'gpt-4'
47+
48+
@description('Version of the model being deployed')
49+
param modelversion string = 'turbo-2024-04-09'
50+
51+
@description('Capacity for specific model used')
52+
param capacity int = 80
53+
54+
@description('Location for all resources.')
55+
param location string = resourceGroup().location
56+
57+
@allowed([
58+
'S0'
59+
])
60+
param sku string = 'S0'
61+
```
62+
63+
The above is defaulted to use your resource groups location as the location for the account and
64+
`gpt-4` version `turbo-2024-04-09`. You can modify this based on the particular model you feel best
65+
fits your needs. You can find more information on available models at
66+
[Azure OpenAI Service models][03]. Additionally, you may need to modify the capacity of the
67+
deployment based on what model you use, you can find more information at
68+
[Azure OpenAI Service quotas and limits][04].
69+
70+
### 2. Deploy the Azure OpenAI Service
71+
72+
Now that you have modified the bicep files parameters, you are ready to deploy your own Azure OpenAI
73+
instance! Simply use either Azure CLI or Azure PowerShell to deploy the bicep files.
74+
75+
#### Using Azure CLI
76+
77+
```sh
78+
az deployment group create --resource-group <resource group name> --template-file ./main.bicep
79+
80+
// Get the endpoint and key of the deployment
81+
az cognitiveservices account show --name <account name> --resource-group <resource group name> | jq -r .properties.endpoint
82+
83+
az cognitiveservices account keys list --name <account name> --resource-group <resource group name> | jq -r .key1
84+
```
85+
86+
#### Using Azure PowerShell
87+
88+
```powershell
89+
New-AzResourceGroupDeployment -ResourceGroupName <resource group name> -TemplateFile ./main.bicep
90+
91+
// Get the endpoint and key of the deployment
92+
Get-AzCognitiveServicesAccount -ResourceGroupName <resource group name> -Name <account name> | Select-Object -Property Endpoint
93+
94+
Get-AzCognitiveServicesAccountKey -ResourceGroupName <resource group name> -Name <account name> | Select-Object -Property Key1
95+
```
96+
97+
### 3. Configuring the agent to use the deployment
98+
99+
Now that you have the endpoint and key of the deployment, you can open up the `openai-gpt` agent and
100+
run `/agent config` to edit the json configuration file with all the details of the deployment. The
101+
example below shows the default system prompt and the fields that need to be updated.
102+
103+
```jsonc
104+
{
105+
// Declare GPT instances.
106+
"GPTs": [
107+
{
108+
"Name": "ps-az-gpt4",
109+
"Description": "<insert description here>",
110+
"Endpoint": "<insert endpoint here>",
111+
"Deployment": "<insert deployment name here>",
112+
"ModelName": "gpt-4",
113+
"Key": "<insert key here>",
114+
"SystemPrompt": "1. You are a helpful and friendly assistant with expertise in PowerShell scripting and command line.\n2. Assume user is using the operating system `osx` unless otherwise specified.\n3. Use the `code block` syntax in markdown to encapsulate any part in responses that is code, YAML, JSON or XML, but not table.\n4. When encapsulating command line code, use '```powershell' if it's PowerShell command; use '```sh' if it's non-PowerShell CLI command.\n5. When generating CLI commands, never ever break a command into multiple lines. Instead, always list all parameters and arguments of the command on the same line.\n6. Please keep the response concise but to the point. Do not overexplain."
115+
}
116+
],
117+
// Specify the default GPT instance to use for user query.
118+
// For example: "ps-az-gpt4"
119+
"Active": "ps-az-gpt4"
120+
}
121+
```
122+
123+
## Conclusion
124+
125+
You have successfully deployed the Azure OpenAI Service and configured your `openai-gpt` agent to
126+
communicate with it! If you would like to go further in the model training, filters and settings you
127+
can find more information about Azure OpenAI deployments at
128+
[Azure OpenAI Service documentation][02].
129+
130+
A big thank you to Sebastian Jensen's medium article,
131+
[Deploy an Azure OpenAI service with LLM deployments via Bicep][01] for inspiring the Bicep code and
132+
guidance on how to deploy the Azure OpenAI Service using Bicep files. Please check out his blog for
133+
more great AI content!
134+
135+
[01]: https://medium.com/medialesson/deploy-an-azure-openai-service-with-llm-deployments-via-bicep-244411472d40
136+
[02]: https://docs.microsoft.com/azure/cognitive-services/openai/
137+
[03]: https://learn.microsoft.com/azure/ai-services/openai/concepts/models?tabs=global-standard%2Cstandard-chat-
138+
[04]: https://learn.microsoft.com/azure/ai-services/openai/quotas-limits

‎docs/development/AzureOAIDeployment/main.bicep

Lines changed: 58 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,58 @@
1+
@description('This is the name of your AI Service Account')
2+
param aiserviceaccountname string = '<Insert own account name>'
3+
4+
@description('Custom domain name for the endpoint')
5+
param customDomainName string = '<Insert own unique domain name>'
6+
7+
@description('Name of the deployment ')
8+
param modeldeploymentname string = '<Insert own deployment name>'
9+
10+
@description('The model being deployed')
11+
param model string = 'gpt-4'
12+
13+
@description('Version of the model being deployed')
14+
param modelversion string = 'turbo-2024-04-09'
15+
16+
@description('Capacity for specific model used')
17+
param capacity int = 80
18+
19+
@description('Location for all resources.')
20+
param location string = resourceGroup().location
21+
22+
@allowed([
23+
'S0'
24+
])
25+
param sku string = 'S0'
26+
27+
resource openAIService 'Microsoft.CognitiveServices/accounts@2024-10-01' = {
28+
name: aiserviceaccountname
29+
location: location
30+
identity: {
31+
type: 'SystemAssigned'
32+
}
33+
sku: {
34+
name: sku
35+
}
36+
kind: 'AIServices'
37+
properties: {
38+
customSubDomainName: customDomainName
39+
}
40+
}
41+
42+
resource azopenaideployment 'Microsoft.CognitiveServices/accounts/deployments@2024-10-01' = {
43+
parent: openAIService
44+
name: modeldeploymentname
45+
properties: {
46+
model: {
47+
format: 'OpenAI'
48+
name: model
49+
version: modelversion
50+
}
51+
}
52+
sku: {
53+
name: 'Standard'
54+
capacity: capacity
55+
}
56+
}
57+
58+
output openAIServiceEndpoint string = openAIService.properties.endpoint

0 commit comments

Comments
 (0)