How to make colabfold_batch use more CPU cores for large ensemble generation?

[nematiaram]

Hi,

I am running colabfold_batch inside a Docker container as part of a webserver pipeline.

We are trying to generate a large ensemble of models for one protein. In one test, we used n_models=1500, but the job ran for about 5 days and did not finish, so it had to be canceled.

During the run, we noticed that only about 6–9 CPUs were being used, even though 64 CPUs were available to the container.

My questions are:

1. Is there a recommended way to make colabfold_batch's use more CPU cores during the modeling/post-MSA steps?
2. Are there specific flags or environment variables for controlling CPU/thread usage?
3. For large ensemble generation, is it better to split the job into many smaller colabfold_batch runs instead of one run with a very large n_models value?

Any advice would be very helpful.

Thank you!