`colabfold_search` fails GPU search

TL;DR -> colabfold_search fails search on "GPU" DB setup

Error

INFO:colabfold.mmseqs.search:Running mmseqs align /home/ec2-user/result/prof_res /home/ec2-user/msa-work/db/uniref30_2302_db_seq /home/ec2-user/result/res_exp /home/ec2-user/result/res_exp_realign --db-load-mode 0 -e 10 --max-accept 100000 --threads 64 --alt-ali 10 -a
align /home/ec2-user/result/prof_res /home/ec2-user/msa-work/db/uniref30_2302_db_seq /home/ec2-user/result/res_exp /home/ec2-user/result/res_exp_realign --db-load-mode 0 -e 10 --max-accept 100000 --threads 64 --alt-ali 10 -a 

Input /home/ec2-user/result/prof_res does not exist
Traceback (most recent call last):
  File "/home/ec2-user/msa-work/localcolabfold/colabfold-conda/bin/colabfold_search", line 8, in <module>
    sys.exit(main())
  File "/home/ec2-user/msa-work/localcolabfold/colabfold-conda/lib/python3.10/site-packages/colabfold/mmseqs/search.py", line 461, in main
    mmseqs_search_monomer(
  File "/home/ec2-user/msa-work/localcolabfold/colabfold-conda/lib/python3.10/site-packages/colabfold/mmseqs/search.py", line 131, in mmseqs_search_monomer
    run_mmseqs(mmseqs, ["align", base.joinpath("prof_res"), dbbase.joinpath(f"{uniref_db}{dbSuffix1}"), base.joinpath("res_exp"), base.joinpath("res_exp_realign"), "--db-load-mode", str(db_load_mode), "-e", str(align_eval), "--max-accept", str(max_accept), "--threads", str(threads), "--alt-ali", "10", "-a"])
  File "/home/ec2-user/msa-work/localcolabfold/colabfold-conda/lib/python3.10/site-packages/colabfold/mmseqs/search.py", line 46, in run_mmseqs
    subprocess.check_call([mmseqs] + params)
  File "/home/ec2-user/msa-work/localcolabfold/colabfold-conda/lib/python3.10/subprocess.py", line 369, in check_call
    raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command '[PosixPath('mmseqs'), 'align', PosixPath('/home/ec2-user/result/prof_res'), PosixPath('/home/ec2-user/msa-work/db/uniref30_2302_db_seq'), PosixPath('/home/ec2-user/result/res_exp'), PosixPath('/home/ec2-user/result/res_exp_realign'), '--db-load-mode', '0', '-e', '10', '--max-accept', '100000', '--threads', '64', '--alt-ali', '10', '-a']' returned non-zero exit status 1.

Steps to reproduce

1. DB setup

git clone https://github.com/sokrypton/ColabFold.git
GPU=1 ./setup_databases.sh /path/to/db_folder

2. Install mmseqs

wget https://mmseqs.com/latest/mmseqs-linux-gpu.tar.gz;
tar xvfz mmseqs-linux-gpu.tar.gz;
	# In ~/.bashrc
	export PATH=$(pwd)/mmseqs/bin/:$PATH
source ~/.bashrc

3. Install localcolabfold

### INSTALL colabfold ###
wget https://raw.githubusercontent.com/YoshitakaMo/localcolabfold/main/install_colabbatch_linux.sh
bash install_colabbatch_linux.sh
	# In ~/.bashrc
	export PATH="/path/to/your/localcolabfold/colabfold-conda/bin:$PATH"	
source ~/.bashrc

4. Running coldfold_search

colabfold_search --mmseqs mmseqs  --gpu 1 input_sequences.fasta /home/ec2-user/msa-work/db/ ~/result/

Infrastructure setup

• Nvidia 1xL40S
• 16 CPU cores (AMD EPYC 7R13 Processor)
• 128GB Ram
• 3TB SSD

Start in colabfold/mmseqs/search.py, especially mmseqs_search_monomer and the run_mmseqs call that invokes align on prof_res. Reproduce the GPU database setup and colabfold_search command, then check why prof_res is absent before alignment. Done means GPU-backed search completes without the missing prof_res error.