tensorflow-macos and tensorflow-metal #89
Description
Any comments on using 'tensorflow-macos' and 'tensorflow-metal' in the 'clrs' ecosystem?
I was able to install tensoflow-macos and tensorflow-metal in the clrs virtual enviornment.
My AMD GPU is being recognized ... but 'clrs' is looking for: 'tpu_driver' , 'cuda', 'tpu'.
Any ideas?
% python3 -m clrs.examples.run
I0605 17:03:39.042836 4560762368 run.py:196] Using CLRS30 spec: {'train': {'num_samples': 1000, 'length': 16, 'seed': 1}, 'val': {'num_samples': 32, 'length': 16, 'seed': 2}, 'test': {'num_samples': 32, 'length': 64, 'seed': 3}}
I0605 17:03:39.044355 4560762368 run.py:180] Dataset found at /tmp/CLRS30/CLRS30_v1.0.0. Skipping download.
Metal device set to: AMD Radeon Pro 5700 XT
systemMemory: 128.00 GB
maxCacheSize: 7.99 GB
...
devices = jax.devices()
logging.info(devices)
I0605 17:03:40.532486 4560762368 xla_bridge.py:330] Unable to initialize backend 'tpu_driver': NOT_FOUND: Unable to find driver in registry given worker:
I0605 17:03:40.534332 4560762368 xla_bridge.py:330] Unable to initialize backend 'gpu': NOT_FOUND: Could not find registered platform with name: "cuda". Available platform names are: Interpreter Host
I0605 17:03:40.536365 4560762368 xla_bridge.py:330] Unable to initialize backend 'tpu': INVALID_ARGUMENT: TpuPlatform is not available.
W0605 17:03:40.536445 4560762368 xla_bridge.py:335] No GPU/TPU found, falling back to CPU. (Set TF_CPP_MIN_LOG_LEVEL=0 and rerun for more info.)