Questions about the tensorflow, tkan, tkat versions used

[zxy426-cyber]

Hello, could you please provide the version of the package you used? There was an error when I used tensorflow2.8.0 tkan=0.3.0 tkat=0.1.1, but I found that tensorflow would also be upgraded and unable to use gpu when using the latest version of tkat.

[remigenet]

Hi, I always use jax[cuda] latest version as backend of keras currently - that's the most performant setup I found (when using nvidia gpu, I do not think amd gpus are handled by jax, but for having tested trying to use tensorflow with a radeon card it was a pain too..)
Still, I would recommend you to update tkat and tkan version to the latest as some fix and improvement have been added to them !
If you have a nvidia card I would advise you to use jax with cuda enable as backend while using keras (pip install "jax[cuda]"), if on an amd gpu, well if I remember well there were previously tensorflow-gpu package (maybe why you are with tensorflow2.8, which is kinda old) but now it should be working with base tensorflow package if all rocm dependencies are installed (should..) !
Anyway without GPU last version of tkat and tkan are tested on CPU with latest tensorflow, jax and torch !