Testing Tutorial notebook, very slow

[GattiMh]

Hello!

I'm working on M2 Mac and I had no issue with installation. Currently, testing your tutorial notebooks optimising the 4 smiles from the smiles.smi file provided.

Upon running these lines:

if __name__ == "__main__":
    path = os.path.join(root, "example/files/smiles.smi")
    args = options(path, k=1, use_gpu=False)   #specify the parameters for Auto3D 
    out = main(args)            #main acceps the parameters and run Auto3D
    print(out)

The notebook mentions taking it less than 1 minute. In my case is quite slow and seems to take multiple minutes. Am I the only one that has these issue?

Many thanks

[LiuCMU]

Hello, I think you were doing it correctly. I can think of 2 reasons for the speed differences, mainly:

• We updated the underlying default NNP model within Auto3D. AIMNet was updated. It's now more accurate, but is also slower compared with other models. ANI2xt and ANI2x are generally faster, you can sepcify isomer_engine=ANI2xt if you want to try one of them.

Model	Elements Covered	Charge	Training Dataset Size	Target DFT	Input	Output
ANI2x	H, C, N, O, F, S, Cl	only neutral molecules	~ 4.5M	ωB97X/6-31G*	elements, coordinates	energy, forces
ANI2xt	H, C, N, O, F, S, Cl	only neutral molecules, with enhancement for tautomers	~ 13M	B97-3c	elements, coordinates	energy, forces
AIMNET	H, B, C, N, O, F, Si, P, S, Cl, As, Se, Br, I	both neutral and charged molecules	~ 20 M	wB97M/Def2-TZVPP	elements, coordinates, molecular charge	energy, forces

• Hardware difference. Mine was run on an Intel Mac. In both Intel and M silicon Macs, only CPUs are used. Auto3D will be much faster on Nvidia GPUs.

Please let me know if you have additional questions :-)

[GattiMh]

That's really useful, thanks. I think it's a matter of using GPUs then, probably those timelines are related to having GPU performance accounted. Maybe you should update the comment on the notebook? As it stands, it implies the CPU version will be very fast.

[GattiMh]

When running the thermodynamic_calculator notebook, the first stage is fine generate low energy conformers but the second part # Calculate thermodynamic properties with the 3D structures I get this error:

0%|                                                                                                           | 0/4 [00:00<?, ?it/s]
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In[5], line 14
     10     t = 298
     11     return (id, t)
---> 14 out_thermo = calc_thermo(out, "AIMNET", get_mol_idx_t, opt_tol=0.003)
     15 print(out_thermo)

File ~/mambaforge/envs/auto3d/lib/python3.12/site-packages/Auto3D/ASE/thermo.py:264, in calc_thermo(path, model_name, get_mol_idx_t, gpu_idx, opt_tol, opt_steps)
    262     T = 298
    263 else:
--> 264     idx, T = get_mol_idx_t(mol)
    265 print(idx)
    267 try:

Cell In[5], line 9, in get_mol_idx_t(mol)
      8 def get_mol_idx_t(mol):
----> 9     id = mol.data["ID"]
     10     t = 298
     11     return (id, t)

AttributeError: 'Mol' object has no attribute 'data'

[LiuCMU]

Change out_thermo = calc_thermo(out, "AIMNET", get_mol_idx_t, opt_tol=0.003) to out_thermo = calc_thermo(out, "AIMNET", opt_tol=0.003) will solve the problem.

Sorry that the examples are outdated. I already updated them :-)