Prompt flow

The pf CLI is used to manage all Prompt Flow resources, including flows, connections, runs, tools, configurations, and services. It provides a unified interface for the entire development lifecycle from initialization to deployment.

# Summary of available command groups
pf flow      # Manage flows
pf connection # Manage connections
pf run        # Manage runs
pf tool       # Init or list tools
pf config     # Manage user config
pf service    # Manage prompt flow service
pf upgrade     # Upgrade CLI
pf trace      # Manage traces

Prompt flow is a development suite for the end-to-end lifecycle of LLM-based AI applications. It is designed to streamline ideation, prototyping, testing, evaluation, deployment, and monitoring.

Key Capabilities:

• Flow Creation: Link LLMs, prompts, Python code, and other tools into executable workflows.
• Debugging: Iterate on flows and trace interactions with LLMs.
• Evaluation: Calculate quality and performance metrics using large datasets.
• CI/CD Integration: Incorporate testing and evaluation into your continuous integration/deployment pipelines.
• Deployment: Deploy flows to serving platforms or integrate them directly into application codebases.
• Collaboration: Use the cloud version via Azure AI for team collaboration.

The Analyze Documents flow is a multi-step pipeline that utilizes the promptflow-azure-ai-language package to perform several NLP tasks.

Workflow Logic:

1. Reads a text file.
2. Translates text to the target language (via Azure AI Translator).
3. Redacts Personally Identifiable Information (PII).
4. Generates extractive and abstractive summaries.
5. Extracts named entities (NER).
6. Analyzes the sentiment of the abstractive summary.

Tools used:

• python tool
• translator
• pii_entity_recognition
• abstractive_summarization
• extractive_summarization
• entity_recognition
• sentiment_analysis

The Analyze Conversations flow uses Azure AI Language models to process conversation transcripts.

Capabilities:

• Language Detection: Identifies the language of the transcript.
• Key Phrase Extraction: Extracts important terms and concepts.
• Conversational PII: Redacts personally identifiable information.
• Conversation Summarization: Generates narrative summaries, recap summaries, follow-up tasks, and chapter titles.

Tools used:

• python tool
• language_detection (from promptflow-azure-ai-language)
• key_phrase_extraction (from promptflow-azure-ai-language)
• conversational_pii (from promptflow-azure-ai-language)
• conversation_summarization (from promptflow-azure-ai-language)

The promptflow-azure package allows users to leverage the cloud-based version of Prompt Flow in Azure AI. It is designed to streamline the development cycle of LLM-powered AI applications by providing tools for prototyping, experimenting, iterating, and deploying.

Key capabilities include:

• Creating executable flows that connect LLMs, prompts, and Python tools using visualized graphs or flex flow code.
• Debugging, sharing, and iterating on flows within a collaborative team environment.
• Evaluating flow performance through large-scale batch runs.

The Prompty CLI supports various use cases ranging from simple question-answering to complex evaluation logic. Common patterns include:

• Basic Chat API: Answering questions using the chat API, with connections configured via environment variables (basic/README.md).
• Chat with History: Leveraging Prompt Flow connections to answer questions while maintaining chat history (chat-basic/README.md).
• Conversation Evaluation: Using prompts to determine specific conversational traits, such as detecting if an assistant has offered an apology (eval-apology/README.md).
• Correctness Evaluation: Using prompts to determine if a generated answer is correct (eval-basic/README.md).

Prompt flow supports deployment to various environments depending on your development and production needs. You can deploy a flow to:

• Local Development Server: For testing and rapid iteration during development.
• Docker Containers: For containerized deployment and portability.
• Kubernetes Clusters: For scalable orchestration in production environments.
• Azure App Service: For managed cloud hosting on Azure.
• Executable Applications: Distributing the flow as a standalone executable.

Specific guides are available for each of these deployment targets to help you configure the appropriate runtime and infrastructure.

Using variants in Prompt Flow provides several advantages for LLM application development:

• Enhance generation quality: Identify the optimal combination of prompts and configurations through experimentation.
• Manage prompt history: Easily track and manage historical versions of LLM nodes, allowing you to revert to or update based on previous iterations without losing progress.
• Boost productivity: Streamline the optimization process by managing multiple variations within a single node structure.
• Data-driven comparison: Effortlessly compare results from different variants side-by-side to decide which performs best.