This repository keeps only the Cortex System-2 evaluation, optional WebSocket serving, and subtask visualization paths.
conda create -n cortex python=3.10 gcc_linux-64 gxx_linux-64 -c conda-forge -y
conda activate cortexconda install -c conda-forge av==15.0.0 sentencepiece==0.2.1 tiktoken
pip install -r requirements.txt
pip install -e .If your datasets are stored in an object store such as S3, Ceph, or Petrel,
install the storage backend required by your environment and make sure
mmengine.fileio can read the configured data_path and annotation_path
URIs. For local files, no extra storage backend is required.
Install FlashAttention 2 for local Qwen-VL inference:
pip install flash-attn --no-build-isolationIf building from source fails, install a wheel matching the local CUDA/PyTorch/Python versions. Example for CUDA 12, PyTorch 2.6, Python 3.10:
wget https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.3/flash_attn-2.7.3+cu12torch2.6cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
pip install ./flash_attn-2.7.3+cu12torch2.6cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
python -c "import flash_attn; print(f'version: {flash_attn.__version__}')"python -m py_compile cortex/inference/step_level_eval.py
python -m py_compile cortex/inference/episode_level_eval.py
python -m py_compile visualize_subtask_sampling.py
bash -n scripts/run_scripts/step_level.sh
bash -n scripts/run_scripts/episode_level.sh
bash -n scripts/run_scripts/run_subtask_visualization.sh