I’d like to know which tensorflow-datasets versions are compatible with which versions of tensorflow? I’m using tensorflow together with autokeras on GPU, it took some time to find matching versions that would recognize the GPU. Now I installed tensorflow-datasets, autokeras doesn’t work anymore. It would be helpful if someone could point me to a list of which TFDF versions go with which TF versions so I can install specific versions in my conda env.
Hi @Julia_Wasala Welcome to the TF Forum
Can you please share your versions of CUDA, cuDNN, GPU driver, etc (GPU support | TensorFlow)
Will try to chase the right team to look into this. Which version of Autokeras, TFDS and TensorFlow do you have right now?
As far as I understand the min version of TF for TFDS is 2.1: datasets/tf_compat.py at 93c1d96dee42b451038e9303166e4c7e83a26b82 · tensorflow/datasets · GitHub
Also I think that we still don’t officially support conda.
@Julia_Wasala Can you try installing TFDS, Autokeras and TF in a
virtual environment /
venv with PiPy (
I’m using cuDNN 8.2.1, cuda 11.3, not sure which GPU driver.
In the end I did find a working configuration: it seems like the culprit was python 3.9
so what I have working now is:
I figured the problem out by looking at all the dependencies on libraries.io, one of the packages required an older version of scikit-learn that was not available in 3.9.
Once I get around to it I will contact the autokeras team about it, because the confusion seems to be stemming from the autokeras website (it says tensorflow above 2.3, python above 3 or something like that but actually there are more constraints)