Iterx Code Examples

This repository contains example tasks for IterX. Each task demonstrates how to integrate custom rewarding logic with the Iterx API for reinforcement learning from code feedback.

Installation

pip install -r requirements.txt

Tasks

Task	Category	Difficulty	Description
pairwise_ranking	Demo	⭐	illustration only
guess_lyric	Demo	⭐	Illustration only
online_packing	Algorithms	⭐⭐	Online bin packing optimization
sql_query_optim	Database	⭐⭐⭐	Optimize SQL queries for performance
smart_contract	Blockchain	⭐⭐⭐	Smart contract gas optimization
cuda_optimization	Systems / GPU	⭐⭐⭐	CUDA kernel optimization
exploit_contract	Blockchain / Security	⭐⭐⭐⭐	Smart contract exploit generation
optimizer	Machine Learning	⭐⭐⭐⭐	Design an optimizer better than Adam
mev_arbitrage	Blockchain / DeFi	⭐⭐⭐⭐⭐	MEV arbitrage strategy design

Note: Tasks with ⭐ (1 star) difficulty are for illustration purposes only, demonstrating how to set up an Iterx task.

Task Structure

Each task folder contains:

• run_iterx.py - Main script to create task and run evaluation loop
• eval_*.py - Custom evaluation logic with get_reward() function
• initial_code.* - Initial code/template provided to the model
• README.md - Task-specific documentation

Usage

cd <task_folder>
python run_iterx.py

The script will:

1. Create a new task (or load existing task_id from task_id.txt)
2. Poll for unevaluated code submissions
3. Evaluate each submission using the custom get_reward() function
4. Submit scores back to the Iterx API
5. Repeat until the task is finished