Previously the default GPU gpu0 is always used in the script.
This small change allows you to chose another GPU and also makes sure DAN-msa doesn't just use gpu0 as default, which then doesn't allow CUDA to distribute the workload itself, which results in OOM errors in case that GPU is busy, even when other GPUs might be idle.
Previously the default GPU gpu0 is always used in the script. This small change allows you to chose another GPU and also makes sure DAN-msa doesn't just use gpu0 as default, which then doesn't allow CUDA to distribute the workload itself, which results in OOM errors in case that GPU is busy, even when other GPUs might be idle.