Empty hit_pdb files because of Error: Could not find CIFs

Expected Behavior

colabfold_batch predicts structures for all complexes

Current Behavior

For some of the complexes, colabfold_batch produces an error (see example in screenshot). I suspect this to be caused by the generated hit_pdb file being empty as for the failing complexes the respective m8 files has a size of 0 bytes (see red rectangle in screenshot).
The remaining complexes with a filled m8 file get predicted correctly.

ColabFold Output (for bugs)

Context

I am following the suggested pipeline with colabfold_search and colabfold_batch. The first step runs, but results in many empty hit_pdb files for my PP complexes, which fail during the second step colabfold_batch

Your Environment

localcolabfold setup script on Debian 12 with L40S GPU.

Question

What am I doing wrong? How can this be fixed?

Thank you in advance!

Start by reproducing the suggested colabfold_search then colabfold_batch pipeline on the PP complexes, using the ColabFold output and empty m8 and hit_pdb files as checkpoints. Compare failing and successful complexes and verify that colabfold_batch completes predictions without the “Could not find CIFs” error.