Skip to content

Generate multiple draft responses and then use a second model to judge the answers and pick a winner, which is then returned to the user.

Notifications You must be signed in to change notification settings

FormulaMonks/gptgladiator

Repository files navigation

Problem:

You can't always rely on LLMs to give you great answers every time. They hallucinate, make stuff up, etc.

Solution:

Generate multiple draft responses and then use a second model to judge the answers and pick a winner, which is then returned to the user. Google's Bard uses this same approach. With this library you can apply the pattern to gpt-3.5 and gpt-4.

Streamlit UI:

We provide a streamlit UI demo to test it out yourself.

How it works

The number of drafts generated is set at 3 but can be configured by the user.

The generation of multiple drafts serves the purpose of providing you with the most optimal response. By considering all potential answers and the requisite level of detail, this approach ensures that the delivered response is accurate, comprehensive, and informative.

Content is then evaluated with GPT4 which takes into account:

It is accurate and factual.
It is comprehensive and informative.
It is clear and easy to understand.
It is relevant to the request.
It is creative and engaging.
The quality of the language used.
The accuracy of the information.
The relevance of the information to the request.
The clarity and conciseness of the writing.
The creativity and engagement of the writing.

Token usage warning

This will use a lot of tokens. For example to generate 3 drafts, you are at 3x + you need to feed those drafts into another prompt + get that response, so >7x.

How to use it:

First run:

pip install gptgladiator

Then use in your code as follows:

from gptgladiator import Gladiator

gladiator = Gladiator(num_drafts=3, api_key="your key here")
winner = gladiator.run("what is a hotdog?")
print("winner = ", winner)

print("drafts generated = ", gladiator.drafts)

print("grades generated = ", gladiator.grades)

Prerequisites

How to run it locally?

Clone this repo:

$ git clone https://github.com/TheoremOne/gptgladiator.git

Navigate to the directory where the sources where cloned to:

$ cd gptgladiator

Create and activate a python virtual environment:

Windows:
$ python -m venv venv
$ .\venv\Scripts\activate

Linux:
$ python3 -m venv venv
$. venv/bin/activate

Upgrade pip if needed and install requirements:

$ pip install --upgrade pip
$ pip install -r requirements.txt

About

Generate multiple draft responses and then use a second model to judge the answers and pick a winner, which is then returned to the user.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •  

Languages