Skip to content

Latest commit

 

History

History
67 lines (47 loc) · 2.1 KB

File metadata and controls

67 lines (47 loc) · 2.1 KB

Install

We recommend using conda to set up the environment.

1. Create and activate conda environment

conda create -n cjepa python=3.10 -y
conda activate cjepa
git clone https://github.com/galilai-group/cjepa.git
cd cjepa

2. Install system dependencies

We use ffmpeg for video processing:

conda install anaconda::ffmpeg

3. Install basic Python dependencies

We recommend using uv to install dependencies. You can also use pip if you prefer.

pip install uv
uv pip install seaborn webdataset swig einops torchcodec av accelerate tensorboard tensorboardX hickle pycocotools wget

4. Install third-party libraries

  • Every third party library should be installed under src/third_party. Please follow the instruction to install the library.
  • Below will install stable-pretraining, stable-worldmodel and nerv.
  • All third-party repositories are installed in editable mode to ensure smooth development.
  • Specific commits/tags are pinned for reproducibility.
  • The environment is tested with Python 3.10.
cd src/third_party
git clone https://github.com/galilai-group/stable-pretraining.git
cd stable-pretraining
git checkout 92b5841
uv pip install -e .

cd ..
git clone https://github.com/galilai-group/stable-worldmodel.git
cd stable-worldmodel
git checkout 221ac82  ## If it doesn't work, please see below
uv pip install -e .


cd ../
git clone https://github.com/Wuziyi616/nerv.git
cd nerv
git checkout v0.1.0   # tested with v0.1.0 release
uv pip install -e .

5. Stable-Worldmodel Manual Download Instruction

Sorry for the inconvenience.