-
Notifications
You must be signed in to change notification settings - Fork 633
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[bugfix][Favor] Same feature map for q and k #183
Conversation
f3b613c
to
5ff7137
Compare
5ff7137
to
fff320d
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks!
I have one minor comment, otherwise LGTM
@@ -98,8 +98,7 @@ def __init__( | |||
"normalize_inputs": self.normalize_inputs, | |||
} | |||
|
|||
self.feature_map_query: FeatureMap = feature_map_constructor(**feature_settings) # type: ignore | |||
self.feature_map_key: FeatureMap = feature_map_constructor(**feature_settings) # type: ignore | |||
self.feature_map: FeatureMap = feature_map_constructor(**feature_settings) # type: ignore |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We should probably enforce now that the iter_before_redraw
is a multiple of 2, or multiply it by 2 in the implementation of the FeatureMap
, otherwise we can still sample different maps for k
/ q
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
oww, that would have been an annoying bug, great catch and I went too fast. I would do a x2 on the setting, because I think that from the outside "iter" means "number of steps we took"
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I updated the docstring for the attention to make it clear that this counted the number of "forward" calls
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks @blefaudeux !
What does this PR do?
Thanks @fmassa for the heads up !
Before submitting
PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.