AI Performance Testing

AI Performance Testing Skill

This skill provides a structured workflow for performance testing of AI systems, focusing on factual accuracy, completeness, and consistency. It leverages the Deepeval framework to measure the performance of LLM-based systems.

Workflow for AI Performance Testing

Follow these steps to conduct comprehensive AI performance testing:

Remind: you should create new files instead of modify template Remind: you should use local existing venv instead of directly create one Remind: you should only modify the tested model, do not change any config for the testing process/models

1. Generate Test Data:

   • Objective: Create a dataset for testing the AI system.
   • Action: Use the template/data_generator.py script to generate test data. You may need to ask the user for a data source to integrate. The output data file should be appropriately named for the test run.
   • Verification: A test data file is created and available for the performance test.

2. Set up the Target LLM-based System:

   • Objective: Integrate the LLM-based system to be tested with the Deepeval framework.
   • Action: Configure the template/performance_test.py script to connect to the target LLM system. This may involve setting API keys, model names, and other parameters.
   • Verification: The setup is complete and the system is ready for testing.

3. Execute Performance Tests:

   • Objective: Run the performance tests to measure the system's factual accuracy, completeness, and consistency.
   • Action: Execute the template/performance_test.py script. The script will use the generated test data to query the LLM system and evaluate its responses against the defined metrics.
   • Verification: The test script runs successfully and outputs the performance metrics.

4. Generate Performance Report:

   • Objective: Create a structured report summarizing the test findings.
   • Action: Process the test results to generate a markdown-formatted report that includes detailed outcomes for accuracy, completeness, and consistency.
   • Verification: Ensure the report is comprehensive, accurate, and saved as a .md file in an accessible location.

Converted and distributed by TomeVault — claim your Tome and manage your conversions.