Tfci.py looking for CUDA libs even if tensorflow-gpu is not installed about compression

[jonycgn]

Hi,

I actually suspect some problem in TF 1.15, since this project doesn't actually have any GPU-specific code. Have you tried running any other TensorFlow code? Like simply python3 -c "import tensorflow"?

Please also post a stack trace, I'd like to know in which line of tfci.py the code fails.

Thanks!

[medgives]

HI,
python3 -c "import tensorflow" works without any problems. The python -m tensorflow_compression.python.all_test command suggested in the readme also returns no errors.

There's no stack trace related to tfci.py, which in facts makes me think too it's some kind of problem with TF.

I think it's also worth mentioning I get the exact same problem on my laptop (also running Ubuntu 18.04, TF 1.15, Python 3.6 and tf-compression 1.3).

Thanks for the help

[jonycgn]

I can't reproduce this on my Mac (running Python 3.7, otherwise same versions). Just compressed and decompressed an image with the model you mentioned successfully.

I will need more time to look into this on Linux. In the mean time, if you have any updates, please post them here.

[medgives]

Ok, I think I discovered what's the problem: it's just a console output problem. The .tfci file is correctly created and it appears to be fine (I was able to decompress it also). My fault, I should have checked before opening the issue.