Outlook: Will local colabfold_batch running at multi-thread GPU mode be online soon?

Hi,

As MSA server was optimized for the broader scope of the beneficiary, large-scale protein-protein interaction predictions was recommended to be deployed at a private server, especially on GPU array. While colabfold_batch was updated to version 1.6.1, it seems that one cannot set multi-thread GPU mode or deploy the prediction period utilizing the full performance of a GPU array. I have deployed localcolabfold on my server but the only way I can do to make best use of the GPUs was to split my list of candidate peptides and sent those lists to individual GPUs seperately.

Is it on schedule that those local colabfold_batch is possible to run at multi-thread mode, so that time of prediction can be cut down by months.
I appreciate for your opensource of local colabfold and thank you for your consideration for this update on multithread GPU mode (for NVIDIA A100 etc.)

Thank you!

The issue names localcolabfold and colabfold_batch but no file or test. Start by tracing how colabfold_batch currently assigns prediction work to GPUs, then clarify the intended multi-thread or multi-GPU behavior for NVIDIA A100 arrays. Done means the supported behavior and implementation scope are defined, with validation for large prediction batches.