Unable to see Memory usage on MMSeq Server

Current Behavior

I have deployed local MMseqs2 Server for ColabFold on a cloud VM with 196G of Memory and 20 Cores CPU. I also indexes in my storage. However I could not see RAM usage beyond 3Gigs for any AA sequence length of 200 to 300. I did not use vmtouch as I did not have sufficient RAM as ~1TB.

Expected Behavior

I expect to see some RAM usage so that processing time can be reduced. Is there any option I am missing. Please guide me in this

Steps to Reproduce (for bugs)

This is the command running on client side:
colabfold_batch --host-url <host_url> --model-type alphafold2_multimer_v3 --templates --amber 230aa_new.fasta output/230_aa_amber

230aa_new.fasta is a fasta file with one sequence of 23AA
I need information to change and configure the options of --prefilter-mode and --db-load-mode from the client side i.e, with colabfold_batch command or start the mmseqs2 server with different values for the options other than the default.

Environment

• MMSEQS_COMMIT=${1:-4589151554eb83a70ff0c4d04d21b83cabc203e4}
  *BACKEND_COMMIT=${2:-14e087560f309f989a5e1feb54fd1f9c988076d5}

• VM Specification:
  Standard E20s v5 (20 vcpus, 160 GiB memory) with Storage Type: Premium SSD LRS, Size: 3TB, Max IOPS: 7500, Max Throughput: 2500MBps, Encryption: SSE with PMK

• Operating system and version:
  Linux (ubuntu 24.04)

No source file or test is named. Start with the colabfold_batch invocation and the MMseqs2 server settings for --prefilter-mode and --db-load-mode; determine where these options are configured and whether the client exposes them. Done means the supported configuration path and expected memory behavior are documented and reproducible.