openai chat gpt microsoft azure的进一步展开说明

# Table of Contents

– Introduction
– What is Azure OpenAI on your data
– Getting Started
– Data Source Options
– Azure Cognitive Search Index
– Ingesting your data into Azure Cognitive Search
– Data formats and file types
– Virtual Network Support & Private Endpoint Support
– Azure OpenAI Resources
– Azure OpenAI Resources in Virtual Networks
– Azure Cognitive Search Resources
– Storage Accounts
– Azure Role-based Access Controls (Azure RBAC)
– Document-level Access Control
– Azure OpenAI Studio
– Web App Customization
– Using the API
– Conclusion

# Introduction

Azure OpenAI on your data is a powerful solution that allows you to run supported chat models like GPT⑶5-Turbo and GPT⑷ on your data without the need for model training or fine-tuning. By running models on your data, you can enhance your chat experience, analyze your data more accurately and quickly, and unlock valuable insights for better decision-making, trend identification, and operational optimization.

# What is Azure OpenAI on your data

Azure OpenAI on your data leverages OpenAI’s GPT⑶5-Turbo and GPT⑷ language models to provide responses based on your data. You can access Azure OpenAI on your data using a REST API or the web-based interface in Azure OpenAI Studio. This solution connects to your data source and enables an enhanced chat experience.

One of the key features of Azure OpenAI on your data is its ability to retrieve and utilize data in a way that enhances the model’s output. It works in conjunction with Azure Cognitive Search to determine what data to retrieve from the designated data source based on user input and conversation history. The retrieved data is then augmented and resubmitted as a prompt to the OpenAI model, ensuring that the model provides responses based on the most up-to-date information in the designated data source. This eliminates the generation of responses based on outdated or incorrect information.

# Getting Started

To get started with Azure OpenAI on your data, you need to have already been approved for Azure OpenAI access and have an Azure OpenAI Service resource with either the GPT⑶5-Turbo or GPT⑷ models deployed. Once you have these prerequisites in place, you can proceed with the following steps:

## Data Source Options

Azure OpenAI on your data uses an Azure Cognitive Search index to determine what data to retrieve based on user inputs and conversation history. You can create your index from a blob storage or local files using Azure OpenAI Studio, which provides a web-based interface for easy data source configuration. The quickstart article provides detailed instructions on how to create your index.

### Ingesting your data into Azure Cognitive Search

For documents and datasets with long text, it is recommended to use the available data preparation script to ingest the data into Azure Cognitive Search. This script chunks the data, ensuring more accurate responses from the service. It also supports scanned PDF files and images, ingesting the data using Document Intelligence.

### Data formats and file types

Azure OpenAI on your data supports various filetypes including .txt, .md, .html, Microsoft Word files, Microsoft PowerPoint files, and PDFs. However, there are some caveats when it comes to document structure and its impact on response quality. For example, the model provides the best citation titles from markdown files and extracts text contents from PDFs as a preprocessing step. If a document includes images or other visual content, the model’s response quality depends on the quality of the extracted text.

## Virtual Network Support & Private Endpoint Support

Azure OpenAI resources, including Azure OpenAI on your data, can be protected in virtual networks and accessed via private endpoints just like any other Azure AI service. Azure Cognitive Search resources can also be protected by a private network. To enable Azure OpenAI on your data to access a protected Azure Cognitive Search resource, an application form needs to be completed. The request will be reviewed, and if approved, a private endpoint request will be sent to the search service for approval.

Please note that virtual networks and private endpoints are only supported for the API and not currently supported for Azure OpenAI Studio.

# Azure OpenAI Resources

Azure OpenAI on your data leverages various Azure resources, including Azure OpenAI resources and Azure Cognitive Search resources. This section provides an overview of the resources involved and their configuration requirements.

### Azure OpenAI Resources in Virtual Networks

Azure OpenAI resources can be protected in virtual networks, allowing you to secure your chat models and related services. The same configuration options available for other Azure AI services can be applied to Azure OpenAI resources. This ensures that your chat models are accessible only within the specified virtual network, providing an additional layer of security.

### Azure Cognitive Search Resources

If you have an Azure Cognitive Search resource that is protected by a private network, and you want to allow Azure OpenAI on your data to access the search service, you need to complete an application form. The application will be reviewed, and if approved, a private endpoint request will be sent to the search service for your approval. Once the request is approved, you can start using the chat completions extensions API.

Please note that public network access can be disabled for the search service once the private endpoint is approved.

### Storage Accounts

Currently, storage accounts in virtual networks and private endpoints are not supported by Azure OpenAI on your data. This means that your data source should not be a storage account protected by a private network. It is recommended to use other supported data sources like Azure Cognitive Search indexes or local files.

### Azure Role-based Access Controls (Azure RBAC)

To add a new data source to your Azure OpenAI resource, you need to have certain Azure RBAC roles assigned to your account. These roles include:
– Cognitive Services Contributor: Allows you to use Azure OpenAI on your data.
– Search Index Data Contributor: Allows you to use an existing Azure Cognitive Search index instead of creating a new one.
– Storage Blob Data Contributor: Allows you to use an existing Blob storage container instead of creating a new one.

Make sure you have these roles assigned to your account before adding a new data source.

### Document-level Access Control

Azure OpenAI on your data provides document-level access control, allowing you to restrict the documents that can be used in responses for different users. This feature leverages Azure Cognitive Search security filters and can be enabled on existing Azure Cognitive Search indexes. To enable document-level access control:
1. Register your application and create users and groups by following the steps in the Azure Cognitive Search documentation.
2. Index your documents with their permitted groups, making sure to include the necessary security fields in the index schema.
3. Add your data source in Azure OpenAI Studio and specify the field that contains the permitted groups information.
4. Ensure that each sensitive document in the index has the correct value set on the security field to indicate the permitted groups of the document.
5. In Azure OpenAI Studio, enable document-level access control in the data source configuration.

Once enabled, Azure OpenAI Studio will provide document access based on the Azure Active Directory (AD) group membership of the logged-in user.

# Azure OpenAI Studio

Azure OpenAI Studio provides a web-based interface for configuring and managing your Azure OpenAI on your data solution. With Azure OpenAI Studio, you can easily connect your data source, define index field mappings, customize chat behavior, and deploy your model as a web app or to Power Virtual Agents.

To connect your data source, simply follow the instructions provided in Azure OpenAI Studio. You can select an existing Azure Cognitive Search index or ingest your data into Azure Cognitive Search directly from the studio.

In addition to data source configuration, Azure OpenAI Studio allows you to customize the behavior of your chat model. You can define a system message to provide instructions or context to the model, set a maximum response length, and limit responses to your data only.

Azure OpenAI Studio also supports web app customization, allowing you to modify the frontend and backend logic of the app. You can customize the app’s appearance, behavior, and even integrate it with other Azure services.

# Web App Customization

The web app provided by Azure OpenAI Studio can be customized to suit your specific needs. You can modify the frontend and backend logic of the app to enhance its appearance and functionality.

For example, you can change the app’s icon by updating the respective file in the source code. You can also customize the app’s behavior, such as resetting the chat session when the user changes settings, and clearly communicating the impact of each setting on the user experience.

When customizing the app, it is recommended to:
– Reset the chat session if the user changes settings to avoid confusion.
– Clearly communicate the impact of each setting on the user experience.
– Keep the app up-to-date by regularly pulling changes from the main branch of the source code.

Please note that when you rotate API keys for your Azure OpenAI or Azure Cognitive Search resource, you need to update the app settings of your deployed apps to use the new keys.

# Using the API

If you prefer to interact with Azure OpenAI on your data programmatically, you can make API calls to the service. The API allows you to send user queries and receive responses from the chat models.

When making API calls, it is recommended to include the following parameters for optimal results:
– fieldsMapping: Explicitly set the title and content fields of your Azure Cognitive Search index. This improves the quality of search retrieval and citation generation.
– roleInformation: Corresponds to the “System Message” in Azure OpenAI Studio. It provides instructions or context to the chat model. Refer to the System Message section for best practices and recommendations.

You can also enable streaming data by setting the “stream” parameter to true in your API request. This allows data to be sent and received incrementally, improving performance and user experience for large or dynamic data.

# Conclusion

Azure OpenAI on your data is a powerful solution that enables you to run supported chat models on your data without the need for training or fine-tuning. By leveraging Azure Cognitive Search and other Azure resources, you can enhance your chat experience, analyze your data more accurately and efficiently, and gain valuable insights for better decision-making. Whether you use Azure OpenAI Studio, the web app, or the API, Azure OpenAI on your data provides the flexibility and tools you need to unlock the full potential of your data. Get started today and start chatting on top of your data!

openai chat gpt microsoft azure的常见问答Q&A