How do you decide which model is best on v1.5.2 ?

Hello all,

I know it sounds like a stupid question, but which "recycling" corresponds to the PDB outfile for each model? Furthermore, does the last recycling (the third in my example) always result in a better quality prediction?

2023-06-07 18:01:32,537 Running colabfold 1.5.2 (05c0cb38d002180da3b58cdc53ea45a6b2a62d31)
2023-06-07 18:03:48,407 Running on GPU
2023-06-07 18:03:48,859 generated new fontManager
2023-06-07 18:03:49,454 Found 7 citations for tools or databases
2023-06-07 18:03:49,455 Query 1/1: module_2786_emile.fasta (length 992)
2023-06-07 18:03:50,625 Sleeping for 6s. Reason: PENDING
2023-06-07 18:03:57,785 Sleeping for 9s. Reason: RUNNING
2023-06-07 18:04:10,169 Sequence 0 found no templates
2023-06-07 18:04:10,169 Sequence 1 found no templates
2023-06-07 18:04:11,321 Sleeping for 5s. Reason: PENDING
2023-06-07 18:04:18,966 Setting max_seq=508, max_extra_seq=1630
2023-06-07 18:05:36,820 alphafold2_multimer_v3_model_1_seed_000 recycle=0 pLDDT=62.6 pTM=0.44 ipTM=0.178
2023-06-07 18:06:23,893 alphafold2_multimer_v3_model_1_seed_000 recycle=1 pLDDT=61.7 pTM=0.448 ipTM=0.172 tol=11.9
2023-06-07 18:07:10,961 alphafold2_multimer_v3_model_1_seed_000 recycle=2 pLDDT=60.9 pTM=0.454 ipTM=0.182 tol=3.81
2023-06-07 18:07:57,967 alphafold2_multimer_v3_model_1_seed_000 recycle=3 pLDDT=61 pTM=0.451 ipTM=0.179 tol=2.42
2023-06-07 18:07:57,983 alphafold2_multimer_v3_model_1_seed_000 took 214.8s (3 recycles)
2023-06-07 18:08:46,222 alphafold2_multimer_v3_model_2_seed_000 recycle=0 pLDDT=60 pTM=0.444 ipTM=0.182
2023-06-07 18:09:33,186 alphafold2_multimer_v3_model_2_seed_000 recycle=1 pLDDT=61 pTM=0.451 ipTM=0.169 tol=8.74
2023-06-07 18:10:20,171 alphafold2_multimer_v3_model_2_seed_000 recycle=2 pLDDT=60.7 pTM=0.443 ipTM=0.162 tol=8.36
2023-06-07 18:11:07,092 alphafold2_multimer_v3_model_2_seed_000 recycle=3 pLDDT=60.5 pTM=0.444 ipTM=0.163 tol=4.75
2023-06-07 18:11:07,093 alphafold2_multimer_v3_model_2_seed_000 took 188.0s (3 recycles)
2023-06-07 18:11:55,374 alphafold2_multimer_v3_model_3_seed_000 recycle=0 pLDDT=59.8 pTM=0.449 ipTM=0.18
2023-06-07 18:12:42,516 alphafold2_multimer_v3_model_3_seed_000 recycle=1 pLDDT=60.2 pTM=0.449 ipTM=0.17 tol=9.08
2023-06-07 18:13:29,547 alphafold2_multimer_v3_model_3_seed_000 recycle=2 pLDDT=59.9 pTM=0.443 ipTM=0.167 tol=5.16
2023-06-07 18:14:16,586 alphafold2_multimer_v3_model_3_seed_000 recycle=3 pLDDT=59.6 pTM=0.447 ipTM=0.172 tol=3.81
2023-06-07 18:14:16,601 alphafold2_multimer_v3_model_3_seed_000 took 188.4s (3 recycles)
2023-06-07 18:15:04,836 alphafold2_multimer_v3_model_4_seed_000 recycle=0 pLDDT=57.7 pTM=0.457 ipTM=0.241
2023-06-07 18:15:51,928 alphafold2_multimer_v3_model_4_seed_000 recycle=1 pLDDT=58.4 pTM=0.437 ipTM=0.17 tol=14.8
2023-06-07 18:16:39,011 alphafold2_multimer_v3_model_4_seed_000 recycle=2 pLDDT=58.2 pTM=0.442 ipTM=0.181 tol=6.9
2023-06-07 18:17:26,120 alphafold2_multimer_v3_model_4_seed_000 recycle=3 pLDDT=58.2 pTM=0.444 ipTM=0.176 tol=4.91
2023-06-07 18:17:26,135 alphafold2_multimer_v3_model_4_seed_000 took 188.4s (3 recycles)
2023-06-07 18:18:14,210 alphafold2_multimer_v3_model_5_seed_000 recycle=0 pLDDT=58.2 pTM=0.467 ipTM=0.224
2023-06-07 18:19:01,085 alphafold2_multimer_v3_model_5_seed_000 recycle=1 pLDDT=59.4 pTM=0.46 ipTM=0.19 tol=8.7
2023-06-07 18:19:47,975 alphafold2_multimer_v3_model_5_seed_000 recycle=2 pLDDT=59.3 pTM=0.454 ipTM=0.185 tol=4.25
2023-06-07 18:20:34,844 alphafold2_multimer_v3_model_5_seed_000 recycle=3 pLDDT=58.8 pTM=0.455 ipTM=0.185 tol=4.21
2023-06-07 18:20:34,845 alphafold2_multimer_v3_model_5_seed_000 took 187.6s (3 recycles)
2023-06-07 18:20:35,978 reranking models by 'multimer' metric

No files, tests, or entry points are named. Start by tracing the ColabFold 1.5.2 output terms for model numbers, recycling, PDB files, and the final multimer ranking. Done means providing a clear explanation of how outputs map to models and whether the final recycle is expected to be better.