Solver Pipeline

This repository contains an asynchronous solution generation pipeline using OpenAI's API. The pipeline is designed to take a problem statement and iteratively generate, refine, and produce a final code solution.

Overview

The pipeline follows these steps:

1. Problem Input: The user provides a problem statement.
2. Solution Sketch Generation: Three initial solution sketches are generated using different temperature settings.
3. Sketch Refinement: The sketches are refined into a final solution sketch.
4. Pseudocode Generation: Pseudocode is generated from the refined sketch.
5. Initial Code Generation: Initial code is generated from the pseudocode.
6. Code Refinement: The initial code is refined multiple times to produce the final code.

Pipeline Steps

1. Problem Input

The user inputs a problem statement that serves as the basis for generating solutions.

2. Solution Sketch Generation

• Temperatures: Three sketches are generated using temperatures of 0.6, 0.8, and 1.0.
• Asynchronous Tasks: Each sketch is generated asynchronously to improve efficiency.

3. Sketch Refinement

• The three sketches are refined into a single, coherent solution sketch.

4. Pseudocode Generation

• Pseudocode is generated from the refined solution sketch to provide a structured plan for coding.

5. Initial Code Generation

• Python code is generated from the pseudocode, forming the initial version of the solution.

6. Code Refinement

• The initial code undergoes three rounds of refinement to ensure quality and correctness.

Usage

To use the pipeline, instantiate the Solver class and call the solve method with a problem statement. The method returns the final refined code.

solver = Solver() final_code = await solver.solve(problem="Your problem statement here")

Configuration

Ensure you have your OpenAI API key set in the config.py file.