fix default uniform init range for kvzch #3336
Open
+53
−3
Conversation
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
meta-cla
bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D81296621 |
emlin
force-pushed
the
export-D81296621
branch
from
48609e0 to
2503b81
Compare
emlin
added a commit
to emlin/torchrec
that referenced
this pull request
Sep 4, 2025
Summary: since zch v.Next is using a very large virtual table size, 2^50, the default uniform init value becomes very small, and when the weight dtype is half, those value essentially becomes 0. We have observed the weight init value is all 0 from the debug log: https://fburl.com/mlhub/aea9mbzf {F1981621246} Differential Revision: D81296621
|
This pull request was exported from Phabricator. Differential Revision: D81296621 |
Summary: since zch v.Next is using a very large virtual table size, 2^50, the default uniform init value becomes very small, and when the weight dtype is half, those value essentially becomes 0. We have observed the weight init value is all 0 from the debug log: https://fburl.com/mlhub/aea9mbzf {F1981621246} Differential Revision: D81296621
emlin
force-pushed
the
export-D81296621
branch
from
2503b81 to
148ce39
Compare
|
This pull request was exported from Phabricator. Differential Revision: D81296621 |
emlin
added a commit
to emlin/torchrec
that referenced
this pull request
Sep 4, 2025
Summary: since zch v.Next is using a very large virtual table size, 2^50, the default uniform init value becomes very small, and when the weight dtype is half, those value essentially becomes 0. We have observed the weight init value is all 0 from the debug log: https://fburl.com/mlhub/aea9mbzf {F1981621246} Reviewed By: kathyxuyy Differential Revision: D81296621
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Summary:
since zch v.Next is using a very large virtual table size, 2^50, the default uniform init value becomes very small, and when the weight dtype is half, those value essentially becomes 0.
We have observed the weight init value is all 0 from the debug log: https://fburl.com/mlhub/aea9mbzf
{F1981621246}
Differential Revision: D81296621