Replies: 1 comment 2 replies
-
|
It is needed for fine-tuning. |
Beta Was this translation helpful? Give feedback.
2 replies
-
|
ok thanks. Are those heads similar in concept to the heads in attention layers? |
Beta Was this translation helpful? Give feedback.
-
|
No, it is just multiple readout to fit different inconsistent electronic structure methods. See here: https://mace-docs.readthedocs.io/en/latest/guide/multihead_training.html |
Beta Was this translation helpful? Give feedback.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
why did the developers add heads to the layers in the MACE model? did they improve performance in MP foundation models? if so, how much did the performance or accuracy?
more details on why the heads were added to the layers will be appreciated
Beta Was this translation helpful? Give feedback.
All reactions