TensorFlow >=2.3

HPC Technical
1

Hello,

I have a code which requires a version of TensorFlow >=2.3. I was wondering if I could install a version with pip in a new virtual python environment or if I should wait for the admin to install it as part of a new module so that it is fully compatible with the hardware capabilities ?

Best,

Hugues

2

Hi there,

I have a code which requires a version of TensorFlow >=2.3.

We can compile the following versions (cf. Redirecting... ):

version versionsuffix toolchain
2.3.1 -Python-3.7.4 foss/2019b, fosscuda/2019b
2.3.1 -Python-3.8.2 foss/2020a, fosscuda/2020a
2.4.1 foss/2020b, fosscuda/2020b
2.4.1 -Python-3.7.4 fosscuda/2019b

I was wondering if I could install a version with pip in a new virtual python environment or if I should wait for the admin to install it as part of a new module so that it is fully compatible with the hardware capabilities ?

Both options are fine, if your code already requires a specific Python virtualenv than I would try also to install TensorFlow there.

Another option would be to use an upstream Docker/Singularity image, we can find a test .sbatch on the HPC software repository at https://gitlab.unige.ch/hpc/softs/-/blob/master/t/tensorflow/cuDNN_7.4.24-CUDA-10.0.130_-test_is_gpu_available-singularity-_TensorFlow_1.7.0-gpu.sbatch .

Thx, bye,
Luca

3

Thank you very much for your quick reply. I will try with a virtualenv where I install a more recent version of TensorFlow.

Best regards,
Hugues

4

4 posts were split to a new topic: Issue with TensorFlow 2.3