How to output 10 predicted structures at one time and how to correctly input HEME as ligand

[KunShi-ECUST]

Hello! I would like to output more (say, 10) predicted results at one time in the locally deployed Chai-1 by me. However, I haven't found relevant content in the tutorial. Could you please provide me with the method?
Besides, I'm unable to input the Smiles file of HEME correctly, and it will report errors. So, currently, I split it into an Fe ion and a porphyrin ring. But the conformation of HEME in the output results is not very accurate. I'm looking forward to your help. Thank you!

[Ruibin-Liu]

My understanding is to change the value 5 in this line to 10 if you want 10.

chai-lab/chai_lab/chai1.py

Line 535 in 44375d5

num_diffn_samples = 5 # Fixed at export time

Not sure about the HEME problem.

[KunShi-ECUST]

I have already changed it to 10, but it still outputs 5 results.

[jackdent]

Updating this constant won't work: the model will always output 5 samples, since that is fixed in the exported model graph.

[jackdent]

The simplest way to do this is to run run_inference multiple times in a loop. Each iteration will produce 5 samples.