Skip to content

Latest commit

 

History

History
57 lines (46 loc) · 2.31 KB

INSTALL.md

File metadata and controls

57 lines (46 loc) · 2.31 KB

UV Installation

First, install uv:

curl -LsSf https://astral.sh/uv/install.sh | sh

Note: You may need to set CUDA_HOME and have it pointing to a valid CUDA 12.x installation. To use a different CUDA version, please change the sources in pyproject.toml (and the torch/torchvision versions). See this guide for more details.

Next, run:

uv sync
uv add --no-build-isolation \
'torch-scatter==2.1.2' \
'flash-attn~=2.7.4' \
'git+https://github.com/facebookresearch/detectron2.git@9604f5995cc628619f0e4fd913453b4d7d61db3f' \
'git+https://github.com/facebookresearch/pytorch3d.git@7a3c0cbc9d7b0e70ef39b7f3c35e9ce2b7376f32'
uv run bash docs/init.sh # Warning: You may need to set CUDA_HOME and TORCH_CUDA_ARCH_LIST (see below)

Conda Installation

Note: If you are using micromamba, set:

alias conda='micromamba'

Note: The instructions below install a standalone CUDA installation in your conda enviorment instead of using the system installation. You may need to use a different CUDA version based on your system drivers, GPU, etc. You can skip this step and instead just set CUDA_HOME, e.g., export CUDA_HOME='/usr/local/cuda-12.4'.

conda create -n univlg python=3.10
conda activate univlg
conda install cuda cuda-nvcc -c nvidia/label/cuda-12.4.1 # Optional
export LD_LIBRARY_PATH="$VIRTUAL_ENV/lib:$LD_LIBRARY_PATH" # Optional
export CUDA_HOME=$CONDA_PREFIX # Optional
conda install pytorch torchvision torchaudio pytorch-cuda=12.4 -c pytorch -c nvidia
pip install torch-scatter -f https://data.pyg.org/whl/torch-2.5.1+cu124.html
pip install 'git+https://github.com/facebookresearch/detectron2.git'
pip install flash-attn==2.6.3 --no-build-isolation
pip install -r docs/requirements.txt
bash docs/init.sh # Warning: You may need to set CUDA_HOME and TORCH_CUDA_ARCH_LIST (see below)

NTLK (Optional)

python -m spacy download en_core_web_sm
python -c 'import nltk; nltk.download("stopwords")'

Troubleshooting

To support multiple GPU architectures, you will need to set TORCH_CUDA_ARCH_LIST. For example:

export TORCH_CUDA_ARCH_LIST="8.0 8.6"

See this guide for more details.