- Name: TensorFlow
- Homepage: https://www.tensorflow.org/
- Why should this be included in the repository? It is a very popular machine learning framework.
- Is it Open Source: yes
- How many users do you anticipate will use this software? Anyone doing machine learning will want to have this, so quite a few that I imagine use a pip installation for now.
- Link to source tarball/zip file: https://github.com/tensorflow/tensorflow/archive/v1.5.0.tar.gz
If/when cuda gets included in Solus, it can then be included in Tensorflow too. In the meantime I don't see any other way, it's CPU only or just wait for cuda. The only downside to include it with CPU only is that I'd expect uninformed users to be tempted to report issues about it.
Personally I'd still use it even without GPU support. =)
Everybody who has an AMD gpu has now a possibility for machine learning under Solus. I just got an gpu accelerated MNIST example to run :) With kernel 4.18 just type the command
echo 'SUBSYSTEM=="kfd", KERNEL=="kfd", TAG+="uaccess", GROUP="video"' | sudo tee /etc/udev/rules.d/70-kfd.rules
Update: Works without command line. Docker is enough.
@DataDrake The build of tensorflow takes massive memory and storage space and that causes problem for me when building it on my desktop. I had to limit the ram usage by adding "--ram_utilization_factor=10" to limit ram usage. Could this be a problem for our build server? If not, could we try building it again on the server and see whether it is reproducible?