AF2 Using MMseqs ColabFold 1.5.4 crash with ValueError: unknown model_type af2_multimer_v3

When I ran an open ColabFold1.5.3 on my browser this morning for a multimer_v3 job, it crashed at the "Run Prediction" stage right after collecting the MSAs and drawing the sequence coverage plot, with an error message signaling ValueError: Unknown model_type alphafold2_multimer_v3. This error recurred when choosing 'multimer_v3' or 'auto' settings. When I restarted a clean AF2, it upgraded to ColabFold1.5.4––a happy surprise!––& I guessed that that ValueError occurred because I was trying to run an outdated ColabFold. So I tried again, first sampling the new "DeepFold v1" choice (how does this compare with Multimer_v3?), and this ran well....but when I tried to purposefully run model_type = auto (or multimer_v3), the program crashed with that same ValueError message, see below. Thanks very much in advance for tracking down the bug, & looking forward to using the new ColabFold version!

2023-12-28 17:52:48,081 Setting max_seq=508, max_extra_seq=2048

ValueError Traceback (most recent call last)
in <cell line: 65>()
63
64 download_alphafold_params(model_type, Path("."))
---> 65 results = run(
66 queries=queries,
67 result_dir=result_dir,

2 frames
/content/colabfold/alphafold/models.py in get_model_haiku_params(data_dir, model_type, model_number, use_fuse, to_jnp)
35 is_deepfold = True
36 else:
---> 37 raise ValueError(f"Unknown model_type {model_type}")
38
39 path = os.path.join(data_dir, "params", file)

ValueError: Unknown model_type alphafold2_multimer_v3

Start by reproducing the multimer_v3 or auto run in the ColabFold notebook, then inspect alphafold/models.py, especially get_model_haiku_params and the model_type handling shown in the traceback. Compare the accepted model names with the values passed during Run Prediction. Done means those settings no longer raise the reported ValueError and the prediction proceeds past MSA collection.