ChatGPT-Repeater

The aim of this project is to automate the process of repeatedly requesting ChatGPT to handle data.

This program concatenates prompt_prefix and prompt_suffix with prompt(content in partx.txt) and saves the response of ChatGPT.

How To Execute

1. Generate and fill out config.json (Referencing config.example.json)

2. Put the tasks into the ./data folder. The format for each task is as follows:

   └── task folder
       ├── meta.json
       ├── part1.txt
       ├── part2.txt
       ├── part3.txt
       └── part4.txt
   

   • Every part(1-based) will be used as prompt.

   • Each task will be outputted to its corresponding folder. The original information in meta.json will be preserved and additional information will be appended, such as "token usage" and "cost".

   • example

     •   ./data
         ├── Task A
         │   ├── meta.json
         │   ├── part1.txt
         │   ├── part2.txt
         │   ├── part3.txt
         │   ├── part4.txt
         │   └── part5.txt
         └── Task B
             ├── meta.json
             ├── part1.txt
             ├── part2.txt
             ├── part3.txt
             └── part4.txt
       

3. Install python dependencies

   pip install -r requirements.txt

4. This script will save all ChatGPT completions to the output folder

   python get_result.py

Output format

• The output of get_result.py will be in the following format:

  20230524_220036
  └── Task A
      ├── meta.json
      ├── output_1.json
      ├── output_2.json
      ├── output_3.json
      ├── output_4.json
      └── output_5.json
  

• The file meta.json will preserve its original key-value pairs and add information about token usage and cost. For example:

  {
      "title": "Paper title",
      "filename": "Paper filename",
      "token_usage": <total token usage>,
      "cost": <total cost>
  }
  

• output_x.json is the response of OpenAI API with part_x.txt

Supported task (post-processing)

• Generating Q&A related to agriculture from paper

  • Use this setting for generating output

    • prompt_prefix / suffix: see ./task_record/task_prompt.md
    • The format for meta.json is as follows:

      {
          "title": "Paper title",
          "filename": "Paper filename"
      }
      

  • Generate a CSV file with the following columns:

    • "question", "answer", "title", "filename"-

    python post_process.py --task AgricultureQA --result_folder_name <generated folder name under output folder>