Public Preview of Azure OpenAI Service for Enhanced Data Management

by HubSite 365 about Microsoft

Software Development Redmond, Washington

Pro UserAll about AIAzure OpenAILearning Selection

Explore Azure OpenAI Service for private ChatGPT creation using personal data, addressing potential fine-tuning issues and utilizing knowledge base for accurate

Introducing Azure OpenAI Service On Your Data in Public Preview In this article, Mick Vleeshouwer explores the feasibility of creating a private ChatGPT model for Q&A using one's own data. He outlines the architecture and data requirements for this, and also discusses the limitations of fine-tuning pre-trained models. Some of the drawbacks of fine-tuning include problems with factual correctness, traceability, access control, and costs. To overcome these, Vleeshouwer suggests separating the knowledge base from the language model, thereby ensuring users receive accurate answers based on relevant data.

Key steps in the process involve:

User asking a question.
The application finds relevant text that likely contains the answer.
A prompt with the relevant text is sent to the large language model.
The user receives an answer or 'No answer found' response.

He also emphasizes the importance of context, recommending the use of a knowledge base for finding relevant documents via semantic search. This involves breaking down data into smaller, manageable pieces.

Azure OpenAI Service on your data supports connecting to multiple sources, including:
✅ Azure Cognitive Search index.
✅ Azure Blob storage container.
✅ Local files

Read the full article Introducing Azure OpenAI Service On Your Data in Public Preview

How to create a private ChatGPT with your own data

In this article, Mick Vleeshouwer explores the feasibility of creating a private ChatGPT model for Q&A using one's own data. He outlines the architecture and data requirements for this, and also discusses the limitations of fine-tuning pre-trained models.

Some of the drawbacks of fine-tuning include problems with factual correctness, traceability, access control, and costs. To overcome these, Vleeshouwer suggests separating the knowledge base from the language model, thereby ensuring users receive accurate answers based on relevant data.

Key steps in the process involve:

1. User asking a question.
2. The application finds relevant text that likely contains the answer.
3. A prompt with the relevant text is sent to the large language model.
4. The user receives an answer or 'No answer found' response.

Vleeshouwer then talks about 'prompt engineering' as a crucial part of the ChatGPT implementation to avoid hallucinations and suggests the inclusion of a footnote to the original document for factual accuracy checks.

He concludes by providing some resources for further exploration, including projects and notebooks, emphasizing that the possibilities for building a Q&A engine with large language models are virtually endless.

https://medium.com/@imicknl/how-to-create-a-private-chatgpt-with-your-own-data-15754e6378a1

Developing Customized Q&A Solutions with Azure OpenAI Service

With Azure OpenAI Service, it is now possible to create tailored Q&A systems that leverage your organization's data for accurate and relevant responses to user queries. By separating the knowledge base from the language model and implementing semantic search, you can overcome common limitations of fine-tuning pre-trained models, such as issues with factual correctness and traceability. This approach provides both a more secure and cost-effective solution for harnessing the power of large language models in a way that meets your organization's unique needs.

Learn about Introducing Azure OpenAI Service On Your Data in Public Preview

Azure OpenAI service is a public preview tool that can be used to create a private ChatGPT model for Question and Answer (Q&A) purposes. This model is created by fine-tuning pre-trained models, however there are drawbacks to this approach, such as problems with factual correctness, traceability, access control, and costs. To tackle these issues, Vleeshouwer suggests separating the knowledge base from the language model, thereby ensuring users receive accurate answers based on relevant data. The process for creating a Q&A model includes the user asking a question, the application finding relevant text, a prompt with the relevant text being sent to the large language model, and the user receiving an answer or 'No answer found' response. Context is also important, and Vleeshouwer suggests using a knowledge base for finding relevant documents via semantic search, which involves breaking down data into smaller, manageable chunks.

Keywords

Microsoft Azure, OpenAI Service, ChatGPT, Fine-Tuning, Knowledge Base, Semantic Search

Public Preview of Azure OpenAI Service for Enhanced Data Management

Explore Azure OpenAI Service for private ChatGPT creation using personal data, addressing potential fine-tuning issues and utilizing knowledge base for accurate

Learn about Introducing Azure OpenAI Service On Your Data in Public Preview

More links on about Introducing Azure OpenAI Service On Your Data in Public Preview

Keywords