Chat stream async

A chat flow defined using async class entry that return output in stream mode. It demonstrates how to create a chatbot that can remember previous interactions and use the conversation history to generate next message.

Prerequisites

Install promptflow sdk and other dependencies in this folder:

pip install -r requirements.txt

What you will learn

In this flow, you will learn:

• how to compose a chat flow that return output in stream mode.
• prompt template format of LLM tool chat api. Message delimiter is a separate line containing role name and colon: "system:", "user:", "assistant:". See OpenAI Chat for more about message role.

  system:
  You are a chatbot having a conversation with a human.
  
  user:
  {{question}}
  

• how to consume chat history in prompt.

  {% for item in chat_history %}
  {{item.role}}:
  {{item.content}}
  {% endfor %}
  

Run flow

• Prepare your Azure Open AI resource follow this instruction and get your api_key if you don't have one.

• Setup connection

Go to "Prompt flow" "Connections" tab. Click on "Create" button, select one of LLM tool supported connection types and fill in the configurations.

Or use CLI to create connection:

# Override keys with --set to avoid yaml file changes
pf connection create --file ../../connections/azure_openai.yml --set api_key=<your_api_key> api_base=<your_api_base> --name open_ai_connection

Note in flow.flex.yaml we are using connection named open_ai_connection.

# show registered connection
pf connection show --name open_ai_connection

• Run as normal Python file

python flow.py

• Test flow

pf flow test --flow flow:ChatFlow --init init.json

• Test flow with yaml You'll need to write flow entry flow.flex.yaml to test with prompt flow.

# run chat flow with default question in flow.flex.yaml
pf flow test --flow . --init init.json

# run chat flow with new question
pf flow test --flow . --init init.json --inputs question="What's Azure Machine Learning?"

pf flow test --flow . --init init.json --inputs question="What is ChatGPT? Please explain with consise statement."

• Create run with multiple lines data

pf run create --flow . --init init.json --data ./data.jsonl --column-mapping question='${data.question}' --stream

You can also skip providing column-mapping if provided data has same column name as the flow. Reference here for default behavior when column-mapping not provided in CLI.

• List and show run meta

# list created run
pf run list

# get a sample run name

name=$(pf run list -r 10 | jq '.[] | select(.name | contains("chat_async_stream_")) | .name'| head -n 1 | tr -d '"')
# show specific run detail
pf run show --name $name

# show output
pf run show-details --name $name

# visualize run in browser
pf run visualize --name $name

Run flow in cloud

• Assume we already have a connection named open_ai_connection in workspace.

# set default workspace
az account set -s <your_subscription_id>
az configure --defaults group=<your_resource_group_name> workspace=<your_workspace_name>

• Create run

# run with environment variable reference connection in azureml workspace
pfazure run create --flow . --init ./init.json --data ./data.jsonl --column-mapping question='${data.question}' --stream
# run using yaml file
pfazure run create --file run.yml --stream