georgkempf

This would be really helpful. I have no idea how to interpret the content of the tertiary file. Is there some documentation?

Hi, can you check the log output in baseDirectory/logs after running protling.py? I also had some problems in the beginning and it gave me the right hints. In my case...

I tried again with --inplace --chunk_size 1 and PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:15000 but still same error. The card has 80 GB memory. Is my understanding correct that using 2 GPUs would speed-up the...

This would be the traceback with CUDA_LAUNCH_BLOCKING=1. I saw in the installation instructions that cuda >= 11.4 is suggested for building triton but the environment.yml installs cudatoolkit 11.3 and there...

Great, thank's a lot! Now it was running for some time but another error occurred. Command line args were: `--gpus 4 --inplace --chunk_size 1` ``` Traceback (most recent call last):...

Thanks again for the super fast fix. It is now running for some hours but at some point it still crashes with OOM. I set the max_split_size already down to...

Looking forward to these optimizations. Overall great project!

I experienced that for some protein sequences (zinc fingers) the raw MSA size (before truncation/filtering) can be > 100 GB. The out-of-memory probably occurs when the sequences from a HHblits/Jackhmmer...