api_gen now excludes backend specific code.
#103
Merged
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
hertschuh
added a commit
to hertschuh/namex
that referenced
this pull request
May 23, 2025
This allows not exploring some directories for the purpose of finding symbols. Reason: - keras-team/keras#21321 - keras-team/keras-rs#103
hertschuh
added a commit
to hertschuh/namex
that referenced
this pull request
May 23, 2025
This allows not exploring some directories for the purpose of finding symbols. Reason: - keras-team/keras#21321 - keras-team/keras-rs#103
|
In draft for now, requires fchollet/namex#7 |
This: - Allows development (`api_gen` / git presubmit hooks) without all backends and backend specific dependencies installed and working. For instance, jax_tpu_embedding currently doesn't import on MacOS Sequoia, this allows running `api_gen` regardless. - Makes sure we don't accidentally create and honor exports that are backend specific.
hertschuh
force-pushed
the
api_gen_backends
branch
from
669d203 to
1287f4e
Compare
cantonios
approved these changes
May 27, 2025
| "keras_rs", | ||
| code_directory="src", | ||
| exclude_directories=[ | ||
| os.path.join("src", "layers", "embedding", "jax"), |
There was a problem hiding this comment.
Should we create a backend/ subfolder somewhere and move these files there? Might make it more explicit, and easier to exclude future backend-specific code.
I'm okay with it either way - just thought I'd make the suggestion.
There was a problem hiding this comment.
Yeah, I was thinking about this option too. On the one hand, it will be cleaner and more consistent with Keras, especially if we end up having more backend dependent code (but I'm hoping we won't). On the other hand, it moves the DistributedEmbedding subclasses far away from the location of the super class, which is not intuitive and harder to navigate.
In the case of keras, it doesn't actually make it easier to exclude backends: https://github.com/keras-team/keras/pull/21321/files because we need to keep the backend folder.
hertschuh
added a commit
that referenced
this pull request
Jun 3, 2025
* Ignore shard_map attr error in mypy. (#97) * Added TF specific documentation to `DistributedEmbedding`. (#94) * Fix symbolic calls for `EmbedReduce`. (#98) `EmbedReduce` was inheriting the behavior from `Embedding` and not correctly applying the reduction. * Move `DistributedEmbedding` declaration to its own file. (#99) Having it in `__init__.py` doesn't play nice with pytype. * Remove dependency on `tree` and use `keras.tree`. (#100) Keras can already depend on either `dmtree` or `optree` and use whichever is best or available on the current platform. * Only enable JAX on linux_x86_64. (#101) * Add out_sharding argument WrappedKerasInitializer. (#102) This is for forward-compatibility. Latest versions of JAX introduce the `out_sharding` argument. * Use Python 3.10 style type annotations. (#104) Now that we require Python 3.10, we can use the shorter annotation style, which should improve the readability of the documentation. * Do not bundle test utils in wheel. (#105) * Update version number to 0.2.1 (#106) As 0.2.0 was just released. * Fix invalid escape sequence in unit test. (#108) * Replace leftover `unflatten_as` to `pack_sequence_as`. (#109) This instance was missed as it is only run on TPU. * Make the declaration of `Nested` compatible with pytype. (#110) Which doesn't support `|` between forward declarations using a string. * Add ragged support for default_device placement on JAX. (#107) Requires calling `preprocess`. Internally, we currently convert ragged inputs to dense before passing to the embedding call(...) function. * Add documentation for using DistributedEmbedding with JAX. (#111) * `api_gen` now excludes backend specific code. (#103) This: - Allows development (`api_gen` / git presubmit hooks) without all backends and backend specific dependencies installed and working. For instance, jax_tpu_embedding currently doesn't import on MacOS Sequoia, this allows running `api_gen` regardless. - Makes sure we don't accidentally create and honor exports that are backend specific. * Enable preprocess calls with symbolic input tensors. (#113) This allows us to more-easily create functional models via: ```python preprocessed_inputs = distributed_embedding.preprocess(symbolic_inputs, symbolic_weights) outputs = distributed_embedding(preprocessed_inputs) model = keras.Model(inputs=preprocessed_inputs, outputs=outputs) ``` * Check for jax_tpu_embedding on JAX backend. (#114) This is to allow users to potentially run Keras RS _without_ the dependency. If a user doesn't have `jax-tpu-embedding` installed, but are on `linux_x86_64` and has a sparsecore-capable TPU available, and if they try to use `auto` or `sparsecore` placement with distributed embedding, will raise an error informing them to install the dependency. --------- Co-authored-by: C. Antonio Sánchez <[email protected]> Co-authored-by: hertschuh <[email protected]>
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:
api_gen/ git presubmit hooks) without all backends and backend specific dependencies installed and working. For instance, jax_tpu_embedding currently doesn't import on MacOS Sequoia, this allows runningapi_genregardless.