LOCA: Logical Chain Augmentation for Olympiad-Level Physics Reasoning

Overview

LOCA (Logical Chain Augmentation) is a novel multi-agent framework for complex physics reasoning, capable of solving Olympiad-level physics problems (IPhO/CPhO). LOCA leverages iterative review and augmentation to enhance LLM reasoning quality on challenging scientific problems.

Installation

Prerequisites

• Python 3.12+
• LLM API key
• uv (recommended)

Setup

1. Clone the repository

git clone xxx
cd LOCA

2. Install dependencies

Using uv (recommended):

uv sync

Project Structure

LOCA/
├── src/ # Source code
│ ├── loca/ # LOCA implementation (Solver, Reviewers, Augmentation)
│ ├── vanilla_review/ # Vanilla review implementation (Baseline)
│ ├── single_llm.py # Single LLM solver (Baseline)
│ ├── api/ # LLM API interfaces (OpenAI, Anthropic, LiteLLM)
│ ├── utils.py # Common utilities
│ └── ...
├── configs/ # Configuration files
├── problem_set/ # Test datasets
├── test_results/ # Evaluation results
└── scripts/ # Analysis scripts

Usage

Running LOCA

uv run main.py --config PATH_TO_YAML --config-name CONFIG_NAME_IN_YAML