LLMGSI

The LLMGSI Project is a research project aiming to use large language models (LLM) to simulate realistic office hour (OH) experiences with real GSIs. Except that instead of waiting for long OH queues, students can get almost instant guidance.

Repository Structure

LLMGSI
├───README.md
├───requirements.txt
├───.gitignore  
└───data (Course specific data)
│   ├───data_config.json
│   ├───cs61a_fa20
│   │   ├───hw1_data.json
│   │   ├───hw2_data.json
│   │   └───...   
│   └───...   
└───src (main source code)
│   ├───baselines
│   ├───nanoserver (backend server code)
│   ├───webapp (frontend web interface)
│   │   └───...   
│   └───...   
│   ├───main.py (main entry point for the program)
│   ├───query_engines (llamaindex query engines)
│   └───setup_utils.py (setup utilities)
└───storage (Prebuilt retrieval index)
│   └───...

Usage (Not Up-to-Date)

Clone the repository with

git clone https://github.com/TCSo/LLMGSI.git

Install all required dependencies with

pip install -r requirements.txt

Replace "YOUR_OPENAI_API_KEY" in src/setup_utils.py with a valid openAI API key and run make to start the application

make run

Start Server

Dependency

pip install django==3.2.10
pip install djangorestframework
pip install django-cors-headers
pip install django-filter

Set up database, and run

cd src/nanoserver
python manage.py makemigrations
python manage.py migrate
python manage.py runserver 0.0.0.0:8080

Local test with this url:

127.0.0.1:8080/api/option:<gpt><llama>/infer

Request body can be empty for default response, or

{"input": "What's gradient descent?"}

then you may get response like:

{
    "data": "Gradient descent is an optimization algorithm commonly used in machine learning and optimization problems. It is used to minimize a function by iteratively adjusting the parameters in the direction of steepest descent. In other words, it helps find the minimum of a function by taking steps proportional to the negative of the gradient at the current point. The gradient represents the direction of the steepest increase in the function, so moving in the opposite direction allows us to find the minimum. Gradient descent is often used in training machine learning models to find the optimal values for the model's parameters.",
    "succeed": 1
}