search

google-deepmind / ferminet

An implementation of the Fermionic Neural Network for ab-initio electronic structure calculations
Apache License 2.0
740 stars 129 forks source link

Model parameters being marked as orphan when using KFAC optimizer #80

Closed Uernd closed 1 month ago

Uernd commented 1 month ago

I was trying to use KFAC optimizer to train the FermiNet, a well-designed PINN for solving atoms and molecules. But when I was trying to add the parity constrain to it, I noticed that some of the parameters was labeled as 'orphan', though the calculation was proceeding normally.

I wonder what 'orphan' means, and is this a normal circumstance? Could these parameters labelled as 'orphan' being updated while trainning normally? I mean even though it finally yielded a correct results, there still was a lot of normally-labelled parameters which ensured the model is trainable.

The following picture contains all my changes to FermiNet. I've only modified the networks.py file. all my changes to ferminet

Thank you for your assistance !

jsspencer commented 1 month ago

It's harmless; it just means KFAC is using a default for handling those layers rather than specialised algorithm (which are used if supported layers are detected -- see KFAC documentation and papers for more details).

Uernd commented 1 month ago

That would be great! Thanks for your assistance! 

Best, Zixiao Zhang     ------------------ Original ------------------ From: "James Spencer"; Date: 2024年9月24日(星期二) 晚上6:38 To: "google-deepmind/ferminet"; Cc: "Zack @.***>; "Author"; Subject: Re: [google-deepmind/ferminet] Model parameters being marked as orphan when using KFAC optimizer (Issue #80)

 

It's harmless; it just means KFAC is using a default for handling those layers rather than specialised algorithm (which are used if supported layers are detected -- see KFAC documentation and papers for more details).

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>