Math QA Batch Evaluation

A full-stack application that provides batch inference capabilities for math problem solving.

Most of the frontend is developed using Cursor

Features

• Batch Inference
  • Upload CSV files with math problems
  • Configurable parameters (model name, system prompt, max sequences, max length)
  • Progress tracking and results visualization
  • Export results to CSV

Prerequisites

• Python 3.8+
• Node.js 14+
• CUDA-capable GPU
• SGLang library
• Your model

Setup

1. Clone the repository:

git clone https://github.com/bogoconic1/aimo-inference-service.git
cd aimo-inference-service

2. Initial Setup:

bash setup.sh

Running the Application

1. Start the backend server:

cd backend
python main.py

2. Start the frontend development server:

cd frontend
npm start

The application will be available at:

• Frontend: http://localhost:3000
• Backend API: http://localhost:8000

Usage

Batch Inference

1. Upload a CSV file with 'id', 'problem', and 'answer' columns
2. Configure max_num_seqs (default: 3) and max_length (default: 1000) parameters
3. Click Process Batch
4. View results in the table
5. Download results as CSV

CSV Format for Batch Inference

The input CSV file should have the following columns:

• id: A unique identifier for each problem
• problem: The math problem text
• answer: The correct answer

Example:

id,problem,answer
1,"What is 2 + 2?",4
2,"Solve for x: 3x + 5 = 20",5

Click "Submit" and leave it all to the server after that! Intermediate progress will be shown as the evaluation goes!

Intermediate Progress

License

Apache 2.0