Use local model parameters

[tfisher013]

Hi developers,

I originally posted this issue under ani-model-zoo, but perhaps it belongs here.

I'm looking for a way to load TorchANI model parameters locally (I downloaded the 2x model parameters from github.com/aiqm/ani-model-zoo/archive/ani-2x.zip). Specifically, is there a way that I can change the line below to load the model parameters from a local path instead of from the aiqm repo?

model = torchani.models.ANI2x(periodic_table_index=True).to(device).double()

Thanks for your help!

[yueyericardo]

Hi,

ani-model-zoo belongs to torchani, and it could only be used within torchani. The reason we move it to ani-model-zoo repo (100MB) is to reduce the size of TorchANI for deployment. Line39 parse_resources.py

Why you need to use it outside of torchani?

[tfisher013]

Thanks for your reply.

I'm using TorchANI in a web app that's behind a firewall and that was causing some issues accessing the parameters on Github. While that issue has been resolved for now, I don't maintain the firewall settings and being able to ensure that the parameters are able to load should something change would be great.

I understand that the parameters aren't loaded each time the model is run, so this would be insurance against using a different model sometime in the future (or perhaps having to replace the current one) with firewall settings that would be causing the issue I had earlier.

[yueyericardo]

Hi,

Good to hear you are using torchani, what's the web app for?

Depends on how you install torchani, you can alway download the zip file from ani-model-zoo, and unzip it into torchani/torchani/resources folder, then it should work. A way to find the resources folder might be

python -c 'import os; import torchani; ani_root = torchani.__file__; ani_resource = os.path.dirname(ani_root) + "/resources"; print(ani_resource)'

/home/richard/dev/torchani/torchani/resources

[tfisher013]

Thanks for your reply.

The web app is used to help determine favorable chemical reaction pathways. It's been really helpful due to its speed.

TorchANI is installed via pip on the web server. I'll give this solution a try.

[IgnacioJPickering]

@tfisher013 has this been solved?

[tfisher013]

Sorry for the delay. Yes, I've been able to use the solution @yueyericardo shared to fix this issue. Thanks for your help!