Skip to content

Latest commit

 

History

History
69 lines (40 loc) · 1.76 KB

readme.md

File metadata and controls

69 lines (40 loc) · 1.76 KB

Table of Contents

  1. Installation
  2. Usage
  3. Functionality
  4. Contributing

Literal AI Generations Downloader & Analyzer

This project provides a Python script for downloading AI generations using the Literal AI API and analyzing the results by calculating various relevance metrics.

Installation

  • Make sure you have Python 3 installed, then install the dependencies with.

    pip install -r requirements.txt

  • Set up your environment variables

    export CD_AI_TUTOR_LITERAL_AI_API_KEY='your-literal-ai-api-key' export CD_AI_TUTOR_OPEN_AI_API_KEY='your-openai-api-key'

Usage

  • This script can download AI generations and calculate scores for groundedness and relevance.

  • Adjust --batch_size to control the number of generations fetched per batch.

    python src/fetch_generations.py --batch_size 50

  • After downloading generations, run the following script to generate and push scores to literalai

    python src/generate_scores.py --log_file logs/generations_10.json

  • Make sure you specify the correct path to your generation log file within the script.

Functionality

  • Download Generations with Pagination:

    • Fetches AI generations from the Literal AI API.
    • Saves the fetched data as JSON files in a specified logs directory.
  • Score Calculation:

    • Calculates groundedness and relevance scores.
    • Uses OpenAI's feedback functions to determine the quality of AI responses.
    • Pushes the scores to literalai

Contributing

Contributions to enhance this project are welcome! Feel free to fork the repository and submit pull requests.