blakemertz
commented
1 month ago Question for anyone out there -- does python -m inference assume that you are running in the DiffDock home directory where inference.py can be referenced to generate a module named 'inference'?
I tried the following:
python -m inference --config default_inference_args.yaml --protein_path examples/1a46_protein_processed.pdb --ligand examples/1a46_ligand.sdf --out_dir test
in the DiffDock home directory and am patiently waiting for results. However, it really looks like my terminal is just hanging (been waiting for 10+ minutes now....)
starwingc
After working on setting up the proper environment for diffdock, I thought I had a system that would allow me to roll with pytorch 2 and cuda 12. I went to run a pdb/ligand pair out of the example folder and got the following error:
I found one other closed issue #175 that had the exact same error message but no expalnation for how it was resolved, so I am opening up a new issue. To take this from the beginning.....
My cloud environment is using cuda 12:
and this was not downgradeable to cuda11. After looking through the many issues trying to install OpenFold to get DiffDock to work (and failing several times myself to set up a DiffDock environment first and then installing OpenFold), I decided to try to get a working version of OpenFold on my cluster. Following the meticulous work and advice from another contributor on the OpenFold git page, I was able to get a working version of OpenFold based on pytorch2 and cuda12 installed.
I then cloned my OpenFold environment into a DiffDock environment so I could work on getting DiffDock running successfully.
Next step was to pull in the remaining dependencies for DiffDock:
I specified the versions of ProDy, e3nn, rdkit, and gradio to stay consstent with the environment.yml in the DiffDock source code. I specified the versions for the torch_ dependencies based on my installed version of pytorch (2.1.2). I had to pull a forked version of esm-fold to stay compatible with a pytorch2/cuda12 installation.
The following is my list of conda packages for this environment:
And here are my modules:
You'll notice that inference is conspicuosly absent from my module list. Which brings up a couple of questions:
Thanks for the help -- been banging my head against this problem for nearly 3 weeks and thought I was THISCLOSE to a working install after getting OpenFold to work.