nanogpt-training.txt

Source: https://docs.azure.cn/en-us/guides/developer/azure-developer-guide
Training note: this is a condensed Azure developer guide text for the ukazure.llm nanoGPT demo.

Azure is a cloud platform for hosting existing applications and building new ones.
Developers can use Azure to develop, test, deploy, monitor, and manage applications.
Applications can start small and scale as customer demand grows.
Azure services can support high availability, regional failover, cloud storage, authentication, monitoring, and DevOps.

Choosing where to start means choosing how much platform management you want.
Virtual Machines give developers infrastructure as a service and full control over the operating system and server configuration.
App Service gives developers a managed platform for web apps, API apps, and mobile back ends.
Azure Functions gives developers serverless execution for code triggered by HTTP requests, events, webhooks, or schedules.
Service Fabric helps developers build, deploy, and manage scalable microservices.
Azure Spring Apps supports Java Spring applications with managed service discovery, configuration, scaling, and monitoring.

App Service is useful when a team wants a fast path to publish web projects.
App Service supports authentication, autoscaling, production testing, continuous deployment, container deployment, and multiple languages.
Developers can host websites, mobile services, and REST APIs from the same platform.
DevOps workflows can publish App Service applications through tools such as GitHub, Jenkins, Azure DevOps, TeamCity, and other automation systems.

Virtual Machines are useful when a team needs full control over the application infrastructure.
With virtual machines, the team manages server software, configuration, maintenance, and operating system patches.
This model works for workloads that do not fit neatly into a platform as a service model.
Virtual machines can host Windows or Linux workloads and can be connected through Azure Virtual Network.

Azure Functions is useful when a team wants to run small pieces of code without managing a complete hosted application.
Function code can be written in languages such as C#, F#, Java, JavaScript, Python, and PowerShell.
Consumption based hosting means the platform can scale execution and the team pays for the time the code runs.
Functions are a good fit for event driven tasks, scheduled jobs, lightweight APIs, and automation glue.

Most applications need data.
Azure Cosmos DB supports globally distributed document, table, graph, and MongoDB compatible workloads.
Azure Storage supports blobs, queues, files, tables, and other nonrelational storage patterns.
Azure SQL Database supports relational data, transactions, referential integrity, performance, scalability, and data protection.
Azure Data Factory can help move data from on premises systems into Azure.
Hybrid Connections can connect an App Service application to on premises resources.

Containers help developers package applications so they run consistently between development, testing, and production.
Azure Kubernetes Service runs and manages clusters for production container workloads.
Azure Container Apps runs microservices and containerized applications on a managed serverless platform.
App Service for Linux can run custom container images for web applications.
Container based deployment lets teams use familiar Docker tools and existing open source workflows.

Authentication matters because applications need to know who users are and protect access to resources.
Microsoft Entra ID provides cloud identity and access management.
Developers can add single sign on, OAuth, OpenID Connect, and Microsoft Graph integration through Microsoft Entra ID.
App Service Authentication can add built in authentication support to App Service applications.
Security design should keep credentials out of source code and prefer managed identity where possible.

Monitoring is part of running software in production rather than merely hoping very professionally.
Application Insights monitors live web applications and helps teams understand performance and usage.
Azure Monitor collects metrics and logs from infrastructure and application resources.
Good monitoring helps developers find issues, understand customer behaviour, and improve reliability over time.

Azure integrates with common DevOps tools.
Teams can use GitHub, Jenkins, Puppet, Chef, TeamCity, Ansible, Azure DevOps, command line tools, and automation pipelines.
Continuous integration and continuous deployment help teams move from local code to cloud-hosted applications repeatably.
Infrastructure and application deployment should be boring, reviewable, and automated whenever possible.

Azure regions are physical locations where applications and data run.
Choosing a region can affect latency, compliance, availability, and disaster recovery.
Teams often keep application data near the application hosting region.
High availability applications may use more than one region so that failover is possible if a datacenter has a serious outage.
Azure operated by 21Vianet has Azure regions in China, including China East, China East 2, China East 3, China North, China North 2, and China North 3.

Developers can manage Azure resources through the Azure portal, command line interfaces, PowerShell, REST APIs, SDKs, and infrastructure as code.
Azure CLI lets developers script tasks against Azure resources from a terminal.
Azure PowerShell provides cmdlets for managing Azure resources from PowerShell.
REST APIs and Azure SDKs let applications and automation tools manage resources programmatically.
SDKs are available for platforms including .NET, JavaScript, Java, Python, PHP, Ruby, and Go.

Azure Resource Manager treats related cloud services as a logical group.
A web application might include App Service, SQL Database, Storage, Azure Cache for Redis, and a content delivery network.
Resource Manager can deploy, update, or delete related resources in a coordinated operation.
Bicep is an infrastructure as code language for defining Azure deployments.
ARM templates are JSON documents for declarative deployment.
Bicep and templates help teams make environments consistent across development, testing, staging, and production.

Developers should also understand Azure accounts, subscriptions, directories, and billing.
An Azure account is an identity that can work with Azure subscriptions.
Subscriptions trust a Microsoft Entra directory for users, services, and devices.
Understanding identity and billing is not glamorous, but it stops cloud projects from becoming expensive archaeology.

For this nanoGPT demo, the training document gives the model Azure vocabulary and developer-shaped phrases.
The generated text will still be character level and imperfect.
The question loop uses this same document as grounded evidence so you can ask about App Service, Functions, storage, authentication, monitoring, regions, Azure CLI, SDKs, Resource Manager, and Bicep.