buaacyw/code-world-model

What it solves

Code World Model (CWM) addresses the challenge of creating a coding agent that acts as a "world brain," providing a framework for coding agents to understand and manipulate the same environment they are operating in.

How it works

CWM uses a specialized model (available as LoRA checkpoints) to perform inference and generate outputs based on specific configurations and prompts. The system allows users to download pre-defined examples and assets to validate, prepare, and generate results using a dedicated inference pipeline.

Who it’s for

Researchers and developers interested in advanced coding agents and world models for software development environments.

Highlights

• LoRA Checkpoints: Provides public LoRA checkpoints for efficient model adaptation.
• Deterministic Expansion: Includes scripts to expand compact Hugging Face examples into a local format for testing.
• Inference Pipeline: A structured process for validating, preparing, and generating outputs via configuration files.