Installing and Running Boltz2 with CUDA 13.2 and Nvidia RTX A6000 GPU

[rjrich]

In case this might help others with a similar issue, I am posting a solution to a problem that I encountered when I first attempted to install Boltz2 in a new conda environment on a Linux workstation with an Nvidia RTX A6000 GPU and running CUDA 13.2. With this setup, there were errors related to the CUDA version and associated libraries. I got a working installation with the following strategy:

1. Remove the boltz2 conda environment.

2. Create a new boltz conda environment with python=3.12

3. Install PyTorch for CUDA 13.2 by using the nightly builds:

   pip3 install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu132

4. Install boltz2 without installing the default torch and cuda packages:

   pip3 install boltz --no-deps

5. Install other required dependencies:

   pip3 install boltz[cuda]

This procedure yielded a working boltz2 installation. It ran inferences that completed without frank errors. However, there were some warnings. I will post a sample output in a separate issue to ask for suggestions about how to remove the warnings or if they can be safely irgnored.

[volgin]

Take a look at https://github.com/Novel-Therapeutics/boltz-community

It has 40+ bug fixes compared to this repo, as well as upgraded dependencies.

[rjrich]

@volgin Thank you! I will check it out.