code-world-model README

The project's README from github.com/buaacyw/code-world-model. As one markdown file: README.md.

Code World Model:
Coding Agent as World Brain

Yiwen Chen1,2   ·   Guosheng Lin2   ·   Chi Zhang1

1 AGI Lab, Westlake University      2 Nanyang Technological University

Project Page   |   arXiv

https://github.com/user-attachments/assets/4c13c5fb-ab45-4a0e-9e59-b85499ef7390

Quick start

  1. Follow the environment and installation guide.
  2. From the repository root, download the public CWM LoRA and deterministically expand the 40 compact Hugging Face examples:
./scripts/install_release_assets.sh "$PWD/release"
  1. Enter the prepared release directory and run a released configuration without changing its inputs, prompts, or seed:
cd release
CONFIG=examples/example_01/config.json
python -m cwm_h3_inference validate --config "$CONFIG"
CUDA_VISIBLE_DEVICES=0 python -m cwm_h3_inference prepare --config "$CONFIG"
CUDA_VISIBLE_DEVICES=0 python -m cwm_h3_inference generate --config "$CONFIG"

Use a new cache directory and output path for every changed experiment. The Hugging Face examples contain compact NPZ conditions, not caches or output videos; the installation script performs the required local format expansion.

Documentation

code-world-model (buaacyw/code-world-model)

Searching Database…

512 ·GitHub
Code-world-model
What is this?Code World Model (CWM) is an academic research project on using a coding agent as a 'world brain', released with a public LoRA checkpoint and example data for reproducing experiments.
Is this a ready-to-use app or a research codebase?It's a research codebase from Westlake University and Nanyang Technological University's AGI Lab, distributed as installation scripts, a LoRA checkpoint, and release configurations rather than a polished app.
How do I get it running?Follow the environment/installation guide, download the public CWM LoRA from the repository root and expand the 40 compact Hugging Face examples, then enter the prepared release directory and run a released configuration.
Do I need the Hugging Face examples before running anything?Yes - the Hugging Face examples only ship compact NPZ condition files, not the caches or output videos a run needs; the installation script expands them into usable local format first.
Can I tweak the inputs or seed of a released experiment?The quick start says to run a released configuration without changing its inputs, prompts, or seed, and to use a fresh cache directory and output path for any experiment you do modify.
What does running a released config actually produce?Running a released configuration produces output videos and cache files, generated locally from the expanded NPZ condition data.
What's in the 40 Hugging Face examples?They are compact NPZ-format condition files for 40 example experiments, which the installation script expands locally into full caches and output videos.
Where's the documentation for input format and configuration?The docs cover environment/installation/model/LoRA downloads, input format and configuration for inference, and license/notice information, as separate linked guides.
What are similar projects?H3-World, jeeves, SolarWM, Hy-MT2, LTX-Video, EasyWAM, Marionette, airllm
Which topics is it in?model, world-models, configuration, lora, dataset, fine-tuning, video-generation, academic-research, artificial-intelligence, inference, generative-ai, huggingface
Creator
buaacyw
Size
87.3 MB
Stars
512
Updated
CommitAug 27
Age
1mo
Commits · 90 Days