MSA search is taking a very long time

So the question is whether there is a way of speeding the MSA search compared to how it is now. For details please see below.

Expected Behavior

I would like to be able to run the script search.py (possibly ad-hoc) to build MSAs within reasonable time.

Current Behavior

Running search.py on a .fasta file with only two sequences took 40 min. This holds both for an ad-hoc version (when I just used the script and pasted in all dependencies without cloning the repo) and for running from the cloned repo (actually from the cloned repo it took even longer than 40 min)

Steps to Reproduce (for bugs)

The .fasta file used

>seq1
VCIHTENQNQVSFYPFVLHEISVLIELTLGHLRYRLTDVPPQPNSQPDSATNYVWML
>seq2
MRFSDNLAKILDKYENLGNKLSSGIMGDEFVKASKEYAELEDVVAKIKEYNKAKSELEEANNFKLEVGLDNATLEMIEDEIHTLENSLPKLERAVKIALLPKDDADSKSAIIEVRAGSGGEEAALFAAVLFNMYQRYAELKGWRFEILAISDTGIGGYKEASASIKGKDVFSKLKFESGVHRVQRVPETESQGRIHTSAATVAVLPEAEEVDIQIEDKDLRIDTYRASGAGGQHVNTTDSAVRITHIPTGITVALQDEKSQHKNKAKALKILRARIYEEERRKKEQERADSRRGQVGSGDRSERIRTYNFPQGRVSDHRINLTLYKIDEVVKNGQLDEFVEALIADDEAKKLLGIYSKNTA

Database dbbase created by running setup_databases.sh and contains uniref30_2302 and colabfold_envdb_202108 and default parameters

ColabFold Output

two .a3m files with 1893 and 20000 sequences. If I lower the parameter --max-seqs to 50 and --max-seq-id to 0.9 it gives two files with 150 and 3000 sequences respectively, but the running time is still 40 min.

Your Environment

Using a ColabFold v1.5.2 within a linux environment

Start with search.py and reproduce the reported runtime using the two-sequence FASTA and the dbbase created by setup_databases.sh. Compare the default run with --max-seqs 50 and --max-seq-id 0.9, using the uniref30_2302 and colabfold_envdb_202108 databases. Done means identifying and addressing the MSA-search slowdown, then verifying the two-sequence case completes in a reasonable time.