Prompt flow tutorial: local development and Azure runs

Uploaded on September 12, 2023, this tutorial demonstrates Prompt flow development across a local machine and Azure Machine Learning. The source gives no package or extension version, so the workflow reflects the upload period rather than a confirmed current setup.

The demo starts with the flow folder stored in the workspace. Its flow.dag.yaml file defines inputs, outputs, tool nodes and variants, while separate source files supply custom Python logic and prompt content. The presenter exports this folder through the UI, opens it in Visual Studio Code and checks it into a code repository.

For local testing, the presenter describes installing the Prompt flow SDK, selecting an environment and supplying flow and data paths to obtain a single-run result. Cloud batch execution uses the Prompt flow Azure package and requires a workspace connection plus a runtime and connection configured in Azure. Results can be inspected in the portal or through a localhost visualization page. A subsequent run uses earlier outputs for evaluation and metrics.

The Visual Studio Code extension demonstration covers prompt variants, debugging, bulk tests and JSONL output inspection. The final section imports an edited local folder back into the Azure UI. This is a local development workflow with cloud execution; the transcript does not establish offline model inference or a self-hosted deployment.