First, clone the Conch repo.
git clone https://github.com/stackav-oss/conch.git
cd conchthen install:
pip install -e .To install torch/triton, specify either the cuda or rocm extra (depending on your platform).
# For Nvidia/CUDA
pip install -e ".[cuda]"# For AMD/ROCm
pip install -e ".[rocm]" --extra-index-url https://download.pytorch.org/whl/rocm6.2.4You can install conch from PyPi via:
pip install conch-triton-kernelsNote: by default, without any extras specified, this will not install torch or triton.
This allows usage of Conch as long as Torch and Triton are already installed in your environment.
For Nvidia/CUDA platforms, you can specify the [cuda] extra to install torch and triton for Nvidia/CUDA platforms.
pip install "conch-triton-kernels[cuda]"For AMD/ROCm platforms, you can specify the [rocm] extra to install torch and triton for AMD/ROCm platforms.
You must also specify the appropriate --extra-index-url.
pip install "conch-triton-kernels[rocm]" --extra-index-url https://download.pytorch.org/whl/rocm6.2.4