Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Update Tensorflow Stub DType #6899

Open
j3soon opened this issue Aug 15, 2024 · 3 comments
Open

Update Tensorflow Stub DType #6899

j3soon opened this issue Aug 15, 2024 · 3 comments

Comments

@j3soon
Copy link

j3soon commented Aug 15, 2024

Environment information (required)

Not related.

Issue description

The tensorflow stub tensorboard/compat/tensorflow_stub/dtypes.py does not include certain types in the latest tensorflow tensorflow/python/framework/dtypes.py. Such as:

  • types_pb2.DT_FLOAT8_E5M2_REF
  • types_pb2.DT_FLOAT8_E4M3FN_REF
  • types_pb2.DT_INT4_REF
  • types_pb2.DT_UINT4_REF

These types exist in tensorboard/compat/proto/types.proto and in tensorflow/python/framework/dtypes.py.

I can open a PR to update the dtypes.py file according to the latest tensorflow code. However, I think it will require adding ml_dtypes as an extra dependency, and I'm not sure if this is the preferred approach.

j3soon added a commit to j3soon/tbparse that referenced this issue Aug 15, 2024
@arcra
Copy link
Member

arcra commented Aug 21, 2024

Hmmm... I don't know if we want to include this dependency at this point. Honestly, I'm not well aware at this time of how this impacts users.

Those types seem to be noted as "experimental" in some way.

Have you experienced any issues with the tool because of this?

@j3soon
Copy link
Author

j3soon commented Aug 22, 2024

Hi @arcra, I haven't encountered any issues with tensorboard, but I noticed that the stubs have significantly diverged from the main TensorFlow repository. Do you think we should update them now, or would it be better to resolve this when something breaks in the future?

@arcra
Copy link
Member

arcra commented Aug 22, 2024

Yes, honestly, we don't have bandwidth at the moment, and it's unclear if users are using these, and to what extent other parts of the code need to be updated. It looks like our tensor_util.py is the piece of code that would care about this, but we'd need to spend some time understanding if we want to and how to add support for these types there as well.

So I'd punt it until someone asks for support for this, but we can keep this open for now. Thanks for letting us know.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants