-
-
Notifications
You must be signed in to change notification settings - Fork 611
reinstate enzyme tests #2603
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
base: master
Are you sure you want to change the base?
reinstate enzyme tests #2603
Conversation
@@ -140,6 +136,7 @@ for (trainfn!, name) in ((Flux.train!, "Zygote"), (train_enzyme!, "Enzyme")) | |||
l2 = sum(pen2, Flux.trainables(m)) | |||
err + 0.33 * l2 | |||
end |
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.
@wsmoses without the change ReverseWithPrimal -> set_runtime_activity(ReverseWithPrimal) made above this gradient fails
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.
Hm when were these tests disabled (and do you know why).
my guess offhand is this is another consequence of Julia 1.11’s now triple pointer array causing performance and analysis to be worse (assuming it worked without on 1.10)
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 don't know, they were disabled already prior to #2602
Is it ok to keep set_runtime_activity in all our gradient infrastructure?
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #2603 +/- ##
==========================================
- Coverage 33.41% 33.38% -0.04%
==========================================
Files 33 33
Lines 2053 2055 +2
==========================================
Hits 686 686
- Misses 1367 1369 +2 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
if name ∈ ["Bilinear", "MultiHeadAttention"] | ||
# these are failing with error: | ||
# Error handling recursive stores for Symbol which has a fieldcount of 0 | ||
@test_broken test_gradients(model, x; loss, compare_finite_diff=false, test_enzyme=true) |
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.
@wsmoses these also used to work
No description provided.