google-deepmind
AlphaFold Colab runs keep getting stuck at AMBER relaxation step
AlphaFold Colab runs keep getting stuck at AMBER relaxation step. Runtime type: Python 3; hardware accelerator: T4 GPU.
AMBER relaxation: 83% 5/6 [elapsed: 30:18 remaining: 06:00] WARNING:root:Warning: importing 'simtk.openmm' is deprecated. Import 'openmm' instead.
Sometimes completes the run; other times it crashes.
Same problem here, but it still works sometimes if you just ignore this warning and let it keep running.
how long you you have to let it run (i know this most likely relates to the complexity of what one is doing)? I am experiencing the same thing on a multimer run right now. The actual run have not crashed it just appears to be stuck with a similar message: 2023-12-04 13:02:00,435 reranking models by 'multimer' metric 2023-12-04 13:02:02,095 Warning: importing 'simtk.openmm' is deprecated. Import 'openmm' instead.
But as I have not seen any progression for a while, I was afraid it was just hanging. It seems ppl get the oenmm warning at different parts of the run?
.@rukibuki in my experience the run throws up this warning and continues; most of the time the run is completed, but sometimes it does not. There is no way to tell. But I have noticed a strong anti-correlation between the size of the system and completion of the run.
This appears to be an issue with the latest version of openmm. Where before it would exit once converged (or reached some maximum number of iterations), now it can keep going forever if clashes remain in structure...
Thank you @sokrypton for checking. I think the issue is not too bad but can be frustrating when no results are generated repeatedly. On one occasion, I tried to skip the minimization step but it still gave me several warnings including the fact that I should be minimizing the structure and then failed to yield a structure. Unfortunately, I did not save the warnings it gave.
Is it possible you have a stale copy of the Colab notebook? We removed all imports via simtk in https://github.com/google-deepmind/alphafold/commit/955737f12cc4b27398fe2369417fc2cc79815eb6.
I am using the AlphaFold Colab that makes use of the version from DeepMind: https://colab.research.google.com/github/deepmind/alphafold/blob/main/notebooks/AlphaFold.ipynb
Are you running relaxation with GPU? Sometimes it will not work. Try to run the prediction with GPU, then relax with CPU. It may take longer, but always work for me (set --use_gpu_relax=False).
