Hi @YoshitakaMo and localcolabfold team,

First of all, huge thanks for providing this tool.

I have a question: What are the set of parameters to use with colabfold_batch and colabfold_search when running everything in a local server with gpus and no data traversal to the MSA server on ColabFold on Google Colaboratory - #258 (comment)

In this regard, I happened to see this issue in the ColabFold repository and within that issue, I happened to see a detailed blog post from @mavericb as a comment with the blog being here: https://www.blopig.com/blog/2024/04/dockerized-colabfold-for-large-scale-batch-predictions/

I notice that @mavericb uses colabfold_batch and colabfold_search this way (with docker):

colabfold_search \
--mmseqs /usr/local/envs/colabfold/bin/mmseqs \
input.fasta database msas \
> /search.log 2>&1 \
&& colabfold_batch msas predictions \
> batch.log 2>&1

Questions:

1. Is the above command line from @mavericb the one to use (I would probably add --amber --use-gpu-relax or should I use the command line as shown here with the use of --use-env 1 --use-templates 1 --db-load-mode 2 for colabfold_search and use of -pdb-hit-file ... --local-pdb-path in colabfold_batch? What's the main difference between the 2 ways
2. The FAQS here suggest that ColabFold does not support multiple GPUs. Is it true for localcolabfold too where I want to run everything on my localserver via a slurm queue which has some single GPU machines and some multi-GPU machines? Related to this, what happens if a job lands on a multi-gpu machine (I am assuming the job will only run on 1 GPU - is my assumption correct?)

Thanks in advance.