Skip to content

Latest commit

 

History

History
100 lines (76 loc) · 2.8 KB

README.md

File metadata and controls

100 lines (76 loc) · 2.8 KB
Raw

Customer Intent Extraction

This sample is using OpenAI chat model(ChatGPT/GPT4) to identify customer intent from customer's question.

By going through this sample you will learn how to create a flow from existing working code (written in LangChain in this case).

This is the existing code.

Prerequisites

Install promptflow sdk and other dependencies:

pip install -r requirements.txt

Ensure you have put your azure OpenAI endpoint key in .env file.

cat .env

Run flow

  1. init flow directory - create promptflow folder from existing python file
pf flow init --flow . --entry intent.py --function extract_intent --prompt-template chat_prompt=user_intent_zero_shot.jinja2

The generated files:

  • extract_intent_tool.py: Wrap the func extract_intent in the intent.py script into a Python Tool.
  • flow.dag.yaml: Describes the DAG(Directed Acyclic Graph) of this flow.
  • .gitignore: File/folder in the flow to be ignored.
  1. create needed custom connection
pf connection create -f .env --name custom_connection
  1. test flow with single line input
pf flow test --flow . --inputs ./data/sample.json
  1. run with multiple lines input
pf run create --flow . --data ./data --column-mapping history='${data.history}' customer_info='${data.customer_info}'

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.

  1. list/show
# list created run
pf run list
# get a sample completed run name
name=$(pf run list | jq '.[] | select(.name | contains("customer_intent_extraction")) | .name'| head -n 1 | tr -d '"')
# show run
pf run show --name $name
# show specific run detail, top 3 lines
pf run show-details --name $name -r 3
  1. evaluation
# create evaluation run
pf run create --flow ../../evaluation/eval-classification-accuracy --data ./data --column-mapping groundtruth='${data.intent}' prediction='${run.outputs.output}' --run $name
# get the evaluation run in previous step
eval_run_name=$(pf run list | jq '.[] | select(.name | contains("eval_classification_accuracy")) | .name'| head -n 1 | tr -d '"')
# show run
pf run show --name $eval_run_name
# show run output
pf run show-details --name $eval_run_name -r 3
  1. visualize
# visualize in browser
pf run visualize --name $eval_run_name # your evaluation run name

Deploy

Serve as a local test app

pf flow serve --source . --port 5123 --host localhost

Visit http://localhost:5213 to access the test app.

Export

Export as docker

# pf flow export --source . --format docker --output ./package