https://github.com/krishnaura45/aimo-problem-solving

🧮Olympiad Math Reasoning 🤖LLMs for LaTeX Problem Solving ⚙️GPT-OSS-120B
https://github.com/krishnaura45/aimo-problem-solving

aimo artificial-intelligence custom-metrics deepseek-r1 gpt-oss-120b gpt-oss-20b kaggle-competition latex llm mathematical-olympiad qwen-3 qwen-3-5

Last synced: 2 months ago
JSON representation

🧮Olympiad Math Reasoning 🤖LLMs for LaTeX Problem Solving ⚙️GPT-OSS-120B

• Host: GitHub
• URL: https://github.com/krishnaura45/aimo-problem-solving
• Owner: krishnaura45
• Created: 2026-05-07T19:46:02.000Z (3 months ago)
• Default Branch: main
• Last Pushed: 2026-05-15T21:24:50.000Z (3 months ago)
• Last Synced: 2026-05-16T00:04:46.132Z (3 months ago)
• Topics: aimo, artificial-intelligence, custom-metrics, deepseek-r1, gpt-oss-120b, gpt-oss-20b, kaggle-competition, latex, llm, mathematical-olympiad, qwen-3, qwen-3-5
• Language: Jupyter Notebook
• Homepage:
• Size: 128 KB
• Stars: 1
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# AI Mathematical Olympiad - Progress Prize 3 ~ `91st Place Solution`

> Solving olympiad-level mathematical reasoning problems using open-source LLM inference pipeline and competitive reasoning workflow.

![Python](https://img.shields.io/badge/Python-3776AB?style=for-the-badge&logo=python&logoColor=white)

![Kaggle](https://img.shields.io/badge/Kaggle-20BEFF?style=for-the-badge&logo=kaggle&logoColor=white)

![LLM](https://img.shields.io/badge/Open--Source-LLMs-purple?style=for-the-badge)

![Math Reasoning](https://img.shields.io/badge/AI-Math%20Reasoning-blueviolet?style=for-the-badge)

![Best Score](https://img.shields.io/badge/Best%20Score-43.0-2ECC71?style=for-the-badge)

![Rank](https://img.shields.io/badge/Rank-92%20of%204138-brightgreen?style=for-the-badge)

![Solo](https://img.shields.io/badge/Submission-Solo-orange?style=for-the-badge)

### **Project Duration**: April 01, 2026 - April 15, 2026

---

## File Structure

```bash

├── INFO.md                      # Competition details, evaluation, rules and timeline

├── README.md

├── aimo-inference.ipynb         # Main competition inference pipeline

├── dependency-install.ipynb     # Dependency installation and environment setup

├── inference.py                 # Standalone inference script

├── utils.ipynb                  # Utility experiments and helper workflows

├── certificate.png              # Certificate of Achievement from Kaggle

├── assets

│   └── eval_info.png            # helper image

└── output

    └── submission.parquet       # Best submission file

```

### Installation

```bash

git clone https://github.com/krishnaura45/aimo-problem-solving.git

cd aimo-problem-solving

```

### Usage

```bash

# Open notebook workflow

jupyter notebook aimo-inference.ipynb

# or run standalone inference

python inference.py

```

---

## Problem Statement

The goal of the **[AI Mathematical Olympiad - Progress Prize 3](https://www.kaggle.com/competitions/ai-mathematical-olympiad-progress-prize-3)** Kaggle competition was to create open-source algorithms capable of solving olympiad-level mathematical reasoning problems written entirely in LaTeX notation.

The competition featured highly challenging problems spanning:

- Algebra

- Combinatorics

- Geometry

- Number Theory

with difficulty ranging from national olympiad level up to IMO-standard mathematics.

Hosted on Kaggle, the challenge evaluated submissions using a specialized penalized accuracy framework across public and hidden private reruns. Each problem required predicting a non-negative integer answer between `0` and `99999`.

The competition emphasized genuine mathematical reasoning capabilities and robust inference under constrained notebook environments.

---

## Approach

### Inference Pipeline

The primary workflow was implemented in:

- `aimo-inference.ipynb`

- `inference.py`

The overall pipeline focused on efficient large language model inference under strict Kaggle notebook constraints.

---

### Problem Processing

- Processed olympiad-style mathematical problems written in LaTeX

- Structured prompts for reasoning-oriented inference

---

### Reasoning Workflow

The inference workflow emphasized:

- Multi-step mathematical reasoning

- Symbolic interpretation of problem statements

- Integer answer extraction and normalization

---

## Competition Results

- Announced on: May 12, 2026.

- **Public/Private Leaderboard Scores**:

  - `37`

  - `38`

  - `39`

  - `40`

  - `42`

  - `43`

- **Performance**:

  - **Best Private Score**: `43.0`

  - Placed **91st out of 4066 participants** and **4138 teams** as a **solo participant**.

---

## References

- Dataset: [AIMO - PP3](https://www.kaggle.com/competitions/ai-mathematical-olympiad-progress-prize-3/data)

- Competition Info: `INFO.md`

- Main Inference Notebook: `aimo-inference.ipynb`

- Official Models Page: https://www.kaggle.com/competitions/ai-mathematical-olympiad-progress-prize-3/models

---

## Tech Stack

- **Language**: Python

- **Libraries / Frameworks**:

  - `transformers`

  - `torch`

  - `pandas`

  - `numpy`

- **Techniques**:

  - LLM Inference

  - Mathematical Reasoning

  - Sequential API-based Evaluation

  - Prompt-based Solving

- **Tools**:

  - Jupyter Notebook

  - Kaggle Notebooks

  - GPU-based inference environments

---

📌 *This project demonstrates the growing capability of open-source reasoning systems in solving olympiad-level mathematical problems under competitive inference constraints.*