update python to 3.10 and Keras minimum version to 3.8

[sachinprasadhs]

Description of the change

• Updated python version from 3.9 to 3.10
• Updated Keras version from 3.5 to 3.6, since couple of models use dot_product_attention, which was introduced in keras from 3.6 version.
• Updated stale script to run only on main repo and skip forked repo and added operations-per-run to cover all the issues and PRs.

Reference

Colab Notebook

Checklist

• I have added all the necessary unit tests for my change.
• I have verified that my change does not break existing code and works with all backends (TensorFlow, JAX, and PyTorch).
• My PR is based on the latest changes of the main branch (if unsure, rebase the code).
• I have followed the Keras Hub Model contribution guidelines in making these changes.
• I have followed the Keras Hub API design guidelines in making these changes.
• I have signed the Contributor License Agreement.

[mattdangerw]

Thanks!

[mattdangerw]

this is the oldest version of keras we support. i think it's to bump to up to 3.8 if we want, that version is preinstalled in colab, but whatever we do here we should also reflect in our minimum pip package requirement https://github.com/keras-team/keras-hub/blob/master/pyproject.toml#L30

also, we should do a search over the repo and remove any workarounds we had for keras 3.5 or earlier, like

keras-hub/keras_hub/src/models/roformer_v2/roformer_v2_attention.py

Line 182 in 625f510

if keras.__version__ < "3.6":

keras-hub/keras_hub/src/models/roformer_v2/roformer_v2_text_classifier_test.py

Line 49 in 625f510

if keras.__version__ < "3.6":

in short, we should always test against the oldest version of keras we want to support, whatever that is

[mattdangerw]

@sachinprasadhs did you see this comment? Are you still working on this?

Any change to testing should go with bumping the actual minimal version in pyproject.toml and removing old workarounds. Maybe let's try bumping to 3.8 and changing that stuff.

[sachinprasadhs]

I have updated the Keras version to 3.8, but JAX backend is constantly failing with same error, do we have to pin JAX version as well?

[sachinprasadhs]

I guess jax.spmd_mode was deprecated in recent JAX versions and we have it removed in Keras as well from 3.10 version keras-team/keras#21164

[sachinprasadhs]

@james77777778 , Could you please check and validate the failing cases with JAX and it's failing for now stable diffusion 3 cases.
Is it something related to Python 3.10 with JAX or SD3?

[james77777778]

@james77777778 , Could you please check and validate the failing cases with JAX and it's failing for now stable diffusion 3 cases. Is it something related to Python 3.10 with JAX or SD3?

This PR should fix the issue.
#2338