Switch to new model parallelism api #1267
Closed
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
mattdangerw
force-pushed
the
model-parallel-distribution
branch
from
18bff4f to
0770022
Compare
Member
Author
|
@qlzh727 any thoughts on how to best test this (once we have a nightly environment to test)? Previously we were just making a 1x1 mesh and passing it in to exercise the code path, which is not great but at least it is something. Is there more we can do here? |
Member
|
For single worker testing, I think you can take approaches like https://github.com/keras-team/keras/blob/c8a5a8969a8712a9a1939937ce34158e04cfc09d/keras/distribution/distribution_lib_test.py#L464. This will run with 8 virtual CPUs locally, and you can do some e2e verification. The config of 8 local virtual CPUs can be found in https://github.com/keras-team/keras/blob/c8a5a8969a8712a9a1939937ce34158e04cfc09d/keras/distribution/distribution_lib_test.py#L17. |
mattdangerw
added a commit
to mattdangerw/keras-hub
that referenced
this pull request
Oct 9, 2023
We will replace this with the work on keras-team#1267 But we have no coverage for that PR till we run tests against Keras 3, which will probably still be about a week. For now, let's just remove this usage, which is no longer needed and will break a Keras 3 install.
Merged
mattdangerw
added a commit
that referenced
this pull request
Oct 9, 2023
We will replace this with the work on #1267 But we have no coverage for that PR till we run tests against Keras 3, which will probably still be about a week. For now, let's just remove this usage, which is no longer needed and will break a Keras 3 install.
mattdangerw
added a commit
that referenced
this pull request
Nov 7, 2023
We will replace this with the work on #1267 But we have no coverage for that PR till we run tests against Keras 3, which will probably still be about a week. For now, let's just remove this usage, which is no longer needed and will break a Keras 3 install.
I813AWt29W
added a commit
to I813AWt29W/keras-nlp
that referenced
this pull request
Aug 26, 2024
We will replace this with the work on keras-team/keras-hub#1267 But we have no coverage for that PR till we run tests against Keras 3, which will probably still be about a week. For now, let's just remove this usage, which is no longer needed and will break a Keras 3 install.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
Move from
DTensorto thekeras.distributionAPI in Keras 3.The overall plan is that we will have a default layout map that will work with most of backbones (because most of our backbones share the same weights structure). A backbone that needs a separate layout map should override
create_layout_map.Note that this will not pass on CI for a while, because we need to tests agains the latest keras-nightlies for this to work. That is blocked on the
keras-nightlybeing available andtensorflow-text-nightlyno longer being a month+ old. Both in route I think.