federated
federated copied to clipboard
anyway to install tff with tensorflow-metal
I'm using my gpu version on macos(tensorflow-metal). https://github.com/apple/tensorflow_macos Tensorflow version:tensorflow2.4.0-rc0 However I can not installing a tff version on tensorflow2.4.0-rc0 When I am trying to install tensorflow_federated using pip install tensorflow_federated==0.18.0, it would automatically install another cpu version of tensorflow macos and replace tensorflow2.4.0-rc0 with tensorflow2.4.0cpu. Anyway to have a gpu version of tff in macos?
In a pyhton3.8 virtual env, for the tf base version 2.5 and tensorflow_macos==2.5.0 and tensorflow-metal seem to work:
SYSTEM_VERSION_COMPAT=0 pip install tensorflow-macos==2.5.0 tensorflow-federated==0.19.0 tensorflow-metal
I tested the tff example from https://www.tensorflow.org/federated, it is kind of weird to see the cpu is faster. But why?
| type | time in secs |
|-----------|--------------|
| cpu | 3.46 |
| gpu/metal | 8.57 |
An update on this with an intel based MBP works with [email protected]
SYSTEM_VERSION_COMPAT=0 pip install tensorflow-macos==2.8.0 tensorflow-metal==0.4.0;
SYSTEM_VERSION_COMPAT=0 pip install tensorflow-federated==0.20.0
Closing as this does not appear to be an issue anymore.