[SLM] Implement pattern replacement #16686
Open
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
|
This PR is currently marked as a draft, both because the unit tests depend on functionality introduced in #16697, and to allow discussion on the feature. |
Lunderberg
force-pushed
the
slm_inject_optimized_kernel
branch
2 times, most recently
from
c6fdea8 to
6493a71
Compare
Lunderberg
force-pushed
the
slm_inject_optimized_kernel
branch
from
559cae7 to
2401b40
Compare
Lunderberg
force-pushed
the
slm_inject_optimized_kernel
branch
2 times, most recently
from
32e071d to
f94dab6
Compare
Lunderberg
changed the title
|
Finally getting back around to this one, ready for review. |
Prior to this commit, if an optimized `nn.Module` implementation were
defined, either the model definition needed to be rewritten to include the
optimized implementation, or the user needed to write their own
`nn.Mutator` implementation.
This commit implements `nn.replace_implementation`, which provides a
`nn.Mutator` that replaces all instances of a `nn.Module` with an
optimized implementation. This allows a user to write optimized
implementations as subclasses, such as shown below.
```python
class BaseImplementation(nn.Module):
...
class OptimizedImplementation(BaseImplementation):
...
```
After defining the optimized implementation,
`nn.replace_implementation(OptimizedImplementation)` returns a
`nn.Mutator` that can inject the optimized module into an existing
end-to-end SLM model. In addition, the SLM-to-SLM transformation can
be converted into a Relax-to-Relax transformation, allowing an easy
path for migrating optimized kernels into a Relax optimation pipeline.
Lunderberg
force-pushed
the
slm_inject_optimized_kernel
branch
from
f94dab6 to
0198ba7
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.
Prior to this commit, if an optimized
nn.Moduleimplementation were defined, either the model definition needed to be rewritten to include the optimized implementation, or the user needed to write their ownnn.Mutatorimplementation.This commit implements
nn.replace_implementation, which provides ann.Mutatorthat replaces all instances of ann.Modulewith an optimized implementation. This allows a user to write optimized implementations as subclasses, such as shown below.After defining the optimized implementation,
nn.replace_implementation(OptimizedImplementation)returns ann.Mutatorthat can inject the optimized module into an existing end-to-end SLM model. In addition, the SLM-to-SLM transformation can be converted into a Relax-to-Relax transformation, allowing an easy path for migrating optimized kernels into a Relax optimation pipeline.