Remove type restrictions on normalization layers' scale and bias#2099
Open
darsnack wants to merge 2 commits intoFluxML:masterfrom
Open
Remove type restrictions on normalization layers' scale and bias#2099darsnack wants to merge 2 commits intoFluxML:masterfrom
darsnack wants to merge 2 commits intoFluxML:masterfrom
Conversation
ToucheSir
reviewed
Nov 3, 2022
src/layers/normalise.jl
Outdated
Comment on lines
327
to
328
| β::V # bias | ||
| γ::V # scale | ||
| γ::U # scale |
Member
There was a problem hiding this comment.
I don't know where the original type param names came from, but would it make sense to harmonize them with the actual parameter names? e.g. B,S instead of U,V.
Member
|
Should we consider wrapping these two in a Scale sub-layer? Replaced with |
Member
|
It's complicated because backends like cuDNN fuse in the affine transform as well. One idea would be to extract the |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Right now, the normalization scale and bias are required to be the same type. This is an unnecessary restriction that isn't there for the weight and biases of other layers.
PR Checklist
Tests are addedEntry in NEWS.mdDocumentation, if applicable