Costa Rica
Last updated: 2024-11-19
List of References (Click to expand)
Table of Contents (Click to expand)
Deploymentsin Azure OpenAI Studio involve making AI models, workflows, and applications available for use in production environments. This includes hosting models on servers or in the cloud and creating APIs or other interfaces for users to interact with the models.
Types of deployments available in Azure OpenAI Studio:
| Deployment Method | Description | Billing Model | Deployment Process |
|---|---|---|---|
| Serverless API | Deploy models as serverless APIs, accessible via endpoints without managing infrastructure. | Pay-as-you-go (per token/request) | 1. Create a project in Azure OpenAI Studio. 2. Navigate to the project dashboard. 3. Click on Deployments.4. Click + Deploy model.5. Select a model from the catalog. 6. Configure deployment settings. 7. Click Deploy.8. Use the generated API endpoint to integrate the model. |
| Models as a Service (MaaS) | Managed service where models are hosted and maintained by Azure. | Pay-as-you-go (per token) | 1. Create a project in Azure OpenAI Studio. 2. Navigate to the project dashboard. 3. Click on Deployments.4. Click + Deploy model.5. Select a model from the catalog. 6. Configure deployment settings. 7. Click Deploy.8. Use the generated API endpoint to integrate the model. |
| Models as a Platform (MaaP) | Deploy models on dedicated virtual machines (VMs) for more control over resources. | Billed as VMs per-hour | 1. Create a project in Azure OpenAI Studio. 2. Navigate to the project dashboard. 3. Click on Deployments.4. Click + Deploy model.5. Select a model from the catalog. 6. Choose VM specifications. 7. Configure deployment settings. 8. Click Deploy.9. Use the generated API endpoint to integrate the model. |
| Flows and Web Applications | Deploy models as part of larger workflows or web applications. | Varies based on usage | 1. Create a project in Azure OpenAI Studio. 2. Navigate to the project dashboard. 3. Click on Deployments.4. Click + Deploy model.5. Select a model from the catalog. 6. Configure deployment settings. 7. Integrate the model into your workflow or web application. 8. Click Deploy.9. Use the generated API endpoint to integrate the model. |
| Managed Compute | Use Azure's managed compute resources to deploy and run models. | Combination of compute and storage costs | 1. Create a project in Azure OpenAI Studio. 2. Navigate to the project dashboard. 3. Click on Deployments.4. Click + Deploy model.5. Select a model from the catalog. 6. Choose managed compute resources. 7. Configure deployment settings. 8. Click Deploy.9. Use the generated API endpoint to integrate the model. |
| Provisioned-Managed | Ensures consistent performance by allocating specific throughput capacity using PTUs. | Billed based on the number of PTUs provisioned. | 1. Create a project in Azure OpenAI Studio. 2. Navigate to the project dashboard. 3. Click on Deployments.4. Click + Deploy model.5. Select a model from the catalog. 6. Configure deployment settings, including the number of PTUs. 7. Click Deploy.8. Use the generated API endpoint to integrate the model. |
- Create a Project: Go to Azure OpenAI Studio, select
New Project, enter a name, and choose a hub. - Deploy a Model: Navigate to the project, select
Deployments, and click+ Deploy model. - Select a Model: Choose a model from the model catalog, such as an Azure OpenAI model.
- Configure Deployment: Specify the deployment name and modify settings as needed.
- Deploy: Click
Deployto make the model available as an API. - Integrate and Use: Use the generated API endpoint to integrate the deployed model with your applications.
- Create an Azure OpenAI Resource
-
Sign in to the Azure portal.
-
Navigate to Create a resource and search for Azure OpenAI.
-
Follow the prompts to create your resource, ensuring it's in a region that supports the models you need.
-
Note
When creating your Azure OpenAI resource, ensure you select a resource group and region that supports the specific model versions you want to use. This can affect latency and availability.
- Deploy the Model
-
Once your resource is created, go to the Azure OpenAI Studio.
-
Select Deployments and click on Create.
-
Choose the model version you want to deploy (e.g.,
gpt-4o-mini-2024-07-18).
-
Note
After deploying the model, you'll receive an API key. Keep this key secure and manage it through Azure's Key Vault for added security.
Be aware of the usage quotas and pricing associated with each model. You can monitor your usage and set up alerts in the Azure portal to avoid unexpected costs.
- Configure the Model
- Fine-tuning (Optional): If you need to fine-tune the model, you can do so by preparing your training data and following the fine-tuning workflow in the Azure OpenAI Studio.
- Structured Outputs: For the GPT-4o-2024-08-06 model, you can specify JSON Schemas within your API calls to ensure the AI output adheres to your defined structure.
Note
Temperature and Max Tokens: Adjust the temperature parameter to control the randomness of the output and max_tokens to limit the length of the response.
Top-p and Frequency Penalty: Use top_p to control the diversity of the output and frequency_penalty to reduce repetitive responses.
- Integration and Testing
- Integrate the deployed model into your application using the provided API endpoints.
- Test thoroughly to ensure the model meets your requirements and adheres to any specified output formats.
Note
Implement robust error handling in your application to manage API rate limits, timeouts, and other potential issues. Use logging to track API calls and responses for debugging and performance monitoring.
Here's a basic example of how you might call the model using Python:
import openai
openai.api_key = "YOUR_API_KEY"
response = openai.Completion.create(
model="gpt-4o-2024-08-06",
prompt="Your prompt here",
max_tokens=100
)
print(response.choices[0].text)