[WIP] export weights as a constants in graph, so can do constant folding to them #19278
Conversation
This file contains 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
zhijxu-MS
changed the title
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
orttraining/orttraining/python/training/ortmodule/_graph_execution_manager.py
Fixed
Show fixed
Hide fixed
zhijxu-MS
force-pushed
the
zhijxu/improve-ortmodule-inference
branch
from
b70cd75 to
604695a
Compare
orttraining/orttraining/python/training/ortmodule/_graph_execution_manager.py
Fixed
Show fixed
Hide fixed
orttraining/orttraining/python/training/ortmodule/_graph_execution_manager.py
Fixed
Show fixed
Hide fixed
zhijxu-MS
force-pushed
the
zhijxu/improve-ortmodule-inference
branch
from
604695a to
8bcd1d8
Compare
zhijxu-MS
force-pushed
the
zhijxu/improve-ortmodule-inference
branch
from
8bcd1d8 to
955cb5b
Compare
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Outdated suggestions cannot be applied.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.
when supporting one 1p model inference, the weights of model are graph inputs and thus ORT's constant folding optimizer can't optimize the weight. after export weights as constants in graph, we see 10%+ gain in the model.
this is a draft
as we need to think of the scenario: model trained > model eval > model trained > model eval > ..., in such case, the weights are changed so we can't just export them as constant directly. one idea is we keep an model_version in trainingsession and also in inferencesession, if we found the model_version number mismatch, then we reexport it.