LLM4Crop - Documentation

Overview

We developed LLM4Crop, a workflow as a Python script that automates the transformation of crop model source code from any languages/modelling platforms to an other language or platform. LLM4Crop uses the Crop2ML standardized format as an intermediate step. It leverages LLMs to refactor, analyze, and document source code through an agent-based workflow. LLM4Crop uses CyMLTh, a process that verifies if a Crop2ML package is in the correct format and if all CyML code are correctly constructed following Crop2ML rules. If not, CyMLTh leverages LLMs to autonomously correct them.

Purpose

The script facilitates the transformation of crop model implementations, written in various programming languages and software architectures into Crop2ML, therefore facilitates crop model component exchange. This process includes:

• Code Analysis & Documentation: Extract metadata and create comprehensive descriptions.
• Code Refactoring: Convert source code to standardized Python modules, in a functional structure.
• Algorithmic Metadata Generation: Extract algorithm inputs, outputs and parameters.
• XML Generation: Create Crop2ML-compliant XML model descriptions.
• Code Transpilation: Convert Python modules to CyML (Crop2ML Language).
• Project Generation: Build complete Crop2ML project structures using cookiecutter.
• Hybrid veryfier: Correct automatically CyML code if not following Crop2ML rules.
• Transpiler: Transpile Crop2ML component into a broad range of languages/platforms

Usage

Command Line Interface

From platform to Crop2ML

python main.py -u <unit_file> <helper_file> ... [-u <unit_file2> ...] [-c <composite_file>] -o <output_folder>

• -u, --unit (required, multiple): Model unit source file(s) to process
• -c, --composite (optional): Composite model file (defines how units connect)
• -o, --output (required): Output folder where results will be saved

Examples

Single Model Unit Processing

python main.py -u soil_temperature.java -o ./output

Processes a single soil temperature model file and generates a Crop2ML component.

Multiple Model Units (Composite Model)

python main.py -u growth.py -u stress.py -u weather.py -c composite.json -o ./output

Processes three model units and combines them into a composite model using the composite.json configuration.

Multiple Model Units separated in different files

python main.py -u surface_temperature.cs surface_temperature_info.txt -u soil_layers_temeprature.cs soil_layers_temperature_structure.json -o ./output

Generates a soil temperature model combining surface and soil layers temperature modules.

From Crop2ML to platform

python main.py -p <Crop2ML package>

• -p, --package (required): The Crop2ML package to transform in all languages/platforms supported

Examples

Single Model Unit Processing

python main.py -p SoilTemperature

Transform the Crop2MLpackage "SoilTemperature" into component compatible with a broad range of languages and platforms.

Configuration Files Required

• API_KEY_PATH: The path of a OpenAi or Claude API's key.