[nnx] add compat #3921
Merged
[nnx] add compat #3921
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
cgarciae
force-pushed
the
nnx-compact
branch
2 times, most recently
from
17e7fc2
to
3dc8838
Compare
cgarciae
force-pushed
the
nnx-object-refactor
branch
3 times, most recently
from
c137583
to
24002bc
Compare
cgarciae
force-pushed
the
nnx-compact
branch
5 times, most recently
from
e4af67e
to
0da1116
Compare
chiamp
reviewed
May 18, 2024
| return x @ self.w + self.b[None] | ||
|
|
||
| @dataclasses.dataclass | ||
| class Foo(nnx.compat.Module): |
There was a problem hiding this comment.
Suggested change
| class Foo(nnx.compat.Module): | |
| class Foo(compat.Module): |
Change all nnx.compat references to compat for consistency?
chiamp
reviewed
May 18, 2024
| >>> import jax.numpy as jnp | ||
| ... | ||
| >>> class Linear(nnc.Module): | ||
| ... def __init__(self, dout, rngs: nnx.Rngs): |
There was a problem hiding this comment.
For clarification:
- if we define an
__init__method, we will be able to instantiate module parameters using the.initmethod? - if we define a
setupmethod or wrap__call__withcompact, we will be able to instantiate the module parameters by calling the module on a sample input and invoking shape inference? - the module parameters that are instantiated are bound to the module so they can be dot-accessed, which is different from Linen where they are returned separately as a variable dict?
- Instead of defining an
__init__method, can we define asetupmethod or wrap__call__withcompactto use the.initmethod as well?
There was a problem hiding this comment.
Sorry for the confusion here, this init method is the current init we have for nnx.Module but we are just moving it out to compat.Module, however its still need to create refactor the method so it follows the Linen API as closely as possible in a subsequent PR.
chiamp
approved these changes
May 18, 2024
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.
What does this PR do?
Adds
nnx.compatmodule with the goal of making it possible to port Linen codebases to an NNX system with as few changes as possible. It would contain the following functionality:Module: inherits fromnnx.Moduleand adds methods fromlinen.Module.compact: allows defining submodules inlinewrappers: some types that simplify NNX <-> Linen interop.