From Scratch#

This guide is intended for standard network environments and walks you through the complete installation and basic evaluation environment setup for FluxVLA.

Installation Steps#

1) Create a Conda Environment#

conda create -n fluxvla python=3.10 -y
conda activate fluxvla

2) Install PyTorch#

pip install torch==2.6.0 torchvision==0.21.0 torchaudio==2.6.0 --index-url https://download.pytorch.org/whl/cu124

3) Install flash-attention#

Option 1: Install via pip

pip install flash-attn==2.5.5 --no-build-isolation --find-links https://github.com/Dao-AILab/flash-attention/releases

Option 2: Build from source

git clone https://github.com/Dao-AILab/flash-attention.git
cd flash-attention
git checkout v2.5.5
MAX_JOBS=4 python setup.py install

4) Install av#

conda install -c conda-forge av=14.4.0

5) Install FluxVLA#

pip install -r requirements.txt
pip install -e . --no-build-isolation

6) Fix numpy Version (Required)#

Some dependencies may install a newer version of numpy, but this project requires numpy 1.26.4. Run the following commands to fix it:

pip uninstall numpy
pip install numpy==1.26.4

Configure the Online Evaluation Environment#

To evaluate LIBERO on devices without ray-tracing capability (e.g., A100), refer to the EGL Device GPU Rendering Configuration.

Install Dependencies#

export MUJOCO_GL=egl
sudo apt install libegl-dev libgl1-mesa-dev libx11-dev libglew-dev libosmesa6-dev

Environment Check#

Verify that /proc/1/environ contains the following environment variables:

  • NVIDIA_DRIVER_CAPABILITIES=all

  • NVARCH=x86_64

  • NVIDIA_REQUIRE_CUDA=cuda>=12.4

  • brand=tesla and driver>=470

Create the EGL Configuration File#

Create the file /usr/share/glvnd/egl_vendor.d/10_nvidia.json with the following content:

{
    "file_format_version": "1.0.0",
    "ICD": {
        "library_path": "libEGL_nvidia.so.0"
    }
}