FIX DOCUMENTATION OF COMPATIBILITY

[coldhearti]

A challenge to those worthy, try finding this information from keras.io or the github page:

• Which JAX version is supported for the latest keras version ?
• Which Tensorflow version is supported the latest keras version ?
• Which Pytorch version is supported the latest keras version ?

You will probably find an answer at least from here: https://keras.io/getting_started
But this leaves a lot to be desired. A compatibility table would be much better.

In general the documentation at keras.io is pretty poor and lacking.

[dhantule]

Hi @coldhearti, thanks for reporting.

For details about latest Keras version compatibility (including supported versions of JAX, TensorFlow, and PyTorch), please refer the requirements.txt file.

[innat]

@dhantule Is there a requirements.txt file that supports all backends on GPU, ensuring the correct CUDA compatibility?

[dhantule]

Hi @innat, There are separate requirements files for each backend, these install the CUDA-enabled dependencies, you can find them here

requirements‑tensorflow‑cuda.txt
requirements‑jax‑cuda.txt
requirements‑torch‑cuda.txt

[innat]

@dhantule
Actually i was referring something like that would work in same environment, like in collab.

[dhantule]

@dhantule Actually i was referring something like that would work in same environment, like in collab.

Hi @innat, as per the documentation, it is recommended to use a clean Python environment for each
backend to avoid CUDA version mismatches, there are separate requirements-{backend}-cuda.txt for TensorFlow, JAX, and PyTorch. You can find this information here

[dhantule]

Hi @coldhearti, PR #21733 was merged and a backend compatibility table has been added to README.md you can find it here, feel free to close this issue.