-
|
Is there a closed form expression to calculate how many learnable parameters a MACE model has? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
One might be able to write one, but for example the number of parameters in the symmetric contraction is not easy to compute. We always print that number in the log file when you start a training, that is the easiest. |
Beta Was this translation helpful? Give feedback.
-
|
Note that most of the parameters in MACE are in the linear layers, which are not the most expensive parts and neither the most expressive, so relating parameters count to either efficiency or accuracy is not straightforward. |
Beta Was this translation helpful? Give feedback.
Note that most of the parameters in MACE are in the linear layers, which are not the most expensive parts and neither the most expressive, so relating parameters count to either efficiency or accuracy is not straightforward.