[Qwix] Conditionally store unquantized inputs in residuals to avoid memory bloat. - #361
Merged
Merged
Conversation
copybara-service
Bot
force-pushed
the
test_977925928
branch
2 times, most recently
from
September 9, 2026 17:02
adfeed1 to
17b16f9
Compare
…emory bloat. Avoid unconditionally saving `lhs_in` and `rhs_in` into the `residuals` tuple in `dot_general_qt_fwd`. Instead, only retain unquantized operands when required by backward contraction (e.g., `use_original_residuals=True` or tiled/microscaling fallback) or by active gradient clipping. PiperOrigin-RevId: 978634942
copybara-service
Bot
force-pushed
the
test_977925928
branch
from
September 9, 2026 17:53
17b16f9 to
228fdd6
Compare
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
[Qwix] Conditionally store unquantized inputs in residuals to avoid memory bloat.
Avoid unconditionally saving
lhs_inandrhs_ininto theresidualstuple indot_general_qt_fwd. Instead, only retain unquantized operands when required bybackward contraction (e.g.,
use_original_residuals=Trueor tiled/microscalingfallback) or by active gradient clipping.