官方技术资料：Microsoft（第 5 页）
官方技术资料 · Microsoft：共 312 条，本页为第 61–75 条。
Microsoft Agent Framework Workflows Orchestrations - Handoff
In-depth look at Handoff Orchestrations in Microsoft Agent Framework Workflows.
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：workflow-orchestration · multi-agent
链接复核时间：2026-09-24
Microsoft Agent Framework Workflows Orchestrations - Magentic
In-depth look at Magentic Orchestrations in Microsoft Agent Framework Workflows.
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：workflow-orchestration · multi-agent
链接复核时间：2026-09-24
Microsoft Agent Framework Workflows Orchestrations - Sequential
In-depth look at Sequential Orchestrations in Microsoft Agent Framework Workflows.
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：workflow-orchestration · multi-agent
链接复核时间：2026-09-24
Designing and implementing a GenAI gateway solution
This document serves as an essential guide for engineering teams tasked with designing and implementing a GenAI gateway solution. This document applies to both Azure OpenAI and LLMs in general
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：deployment · observability
链接复核时间：2026-09-24
Key considerations for designing a GenAI gateway solution
Key considerations for designing a GenAI gateway solution
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：deployment · observability
链接复核时间：2026-09-24
GenAI gateway reference architecture using APIM
The selection below describes the reference design for key GenAI gateway.
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：deployment · observability
链接复核时间：2026-09-24
Generative AI
This section serves as a comprehensive guide for launching, operating, and enhancing Generative AI applications in a production environment. It encapsulates essential capabilities, from delivering innovative experiences to ensuring robust security and privacy, and managing the lifecycle of AI solutions.
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：agent · agent-frameworks
链接复核时间：2026-09-24
LLMOps - Operational management of LLMs
LLMOps is the collection of tools and processes that manages the end-to-end process of developing, deploying, and maintaining LLM-based applications.
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：deployment · observability
链接复核时间：2026-09-24
Monitoring Generative AI applications
As the adoption of generative AI applications continues to grow, so does the necessity for robust monitoring.
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：observability · deployment
链接复核时间：2026-09-24
Operationalize LLM Security & Safety Evaluations using Azure AI Evaluations SDK
This article explains how to implement local safety and security evaluations using the Azure AI Evaluations SDK. It describes how to securely integrate these evaluations into CI/CD pipelines for continuous security validation in Azure DevOps.
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：model-evaluation · llm
链接复核时间：2026-09-24
Security planning for LLM-based applications
This article discusses the Security planning for the sample Retail-mart application. It shows the architecture and data flow diagram of the example application.
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：governance · human-in-the-loop · agent
链接复核时间：2026-09-24
Evaluating the performance of LLM summarization prompts with G-Eval
Evaluating the performance of summarization prompts is a challenging concept. The normal comparison done by LLMs between the output and a ground truth is not feasible, as the answer is subjective and difficult to compare. One solution, known as G-Eval, uses GPT-4 to evaluate the quality of summaries without a ground truth. It performs evaluation across four dimensions: fluency, coherence, consistency, relevance; it then outputs numeric results, allowing for objective comparison between prompt iterations, and a cohesive and scientific experimentation process.
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：model-evaluation · llm
链接复核时间：2026-09-24
A list of metrics for evaluating LLM-generated content
Evaluating the performance of machine learning models is crucial for determining their effectiveness and reliability. To do that, quantitative measurement with reference to ground truth output (also known as evaluation metrics) are needed. However, LLM applications are a recent and fast evolving ML field, where model evaluation is not straightforward and there is no unified approach to measure LLM performance. Several metrics have been proposed in the literature for evaluating the performance of LLMs. It is essential to use the right metrics that are suitable for the problem we are attempting to solve.
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：model-evaluation · llm
链接复核时间：2026-09-24
Fostering appropriate reliance on GenAI
This article describes a framework that helps product teams identify, assess, and mitigate overreliance risk in AI products.
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：governance · human-in-the-loop · agent
链接复核时间：2026-09-24
Create an MCP Server knowledge source (preview)
Learn how to create an MCP Server knowledge source for agentic retrieval in Azure AI Search, which connects to any external Model Context Protocol server.
证据用途：用于理解、实现、选型、可靠性、安全或迁移判断。
使用边界：只直接证明 Microsoft 的实现与声明；预览功能和产品生命周期使用前应复核
关联概念：rag · embedding · vector-db
链接复核时间：2026-09-24