Replace native module implementation with managed code. #1398
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
|
And thank you, @shaltielshmid for all the help on getting this one ready. |
This was
linked to
issues
Oct 29, 2024
…s and/or buffers were declared on a base class. 2. Addressed issue with ParameterDict and ParameterList not doing _to() properly.
… This is a minor performance improvement.
2. Removed redundant '_to()' overrides in parameter-less modules. 3. Added ParameterLessModule<...> for all the type signatures that Module<...> supports.
…List modules are held on.
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Outdated suggestions cannot be applied.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.
This PR, which has been in the works for a while, replaces the implementation of many (but not all) built-in modules with an equivalent managed code implementation.
The rationale is primarily to follow what is done in Pytorch, where most modules are now implemented in Python. It also makes it easier to support access to a number of module 'attributes' in Pytorch, i.e. properties, which was harder to do with the native implementations, which did not support access to all such attributes.
This is a massive PR, but the patterns for replacements are just a few:
Change the constructors to not call the native code module factories. Importantly, the constructors must not call
RegisterComponents()like most custom modules should.Change the
forwardmethods to invoke functional APIs instead of the native forward method.Implement each 'attribute' using a .NET property with a setter and a getter. The backing fields should be nullable, even if the property itself is not, because the constructor is expected to use the property setter for initialization.