tensorflow-metal plugin errors

Question

Created Jan ’23

Replies 2

Boosts 1

Views 2.2k

Participants 3

I followed the documentation the install tensorflow-metal but when I ran the verification example on that page I am getting a bunch of errors

11490434/11490434 [==============================] - 0s 0us/step
2023-01-07 17:01:16.644549: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:306] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support.
2023-01-07 17:01:16.644938: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:272] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 0 MB memory) -> physical PluggableDevice (device: 0, name: METAL, pci bus id: <undefined>)
Metal device set to: Apple M2

2023-01-07 17:01:16.907085: W tensorflow/tsl/platform/profile_utils/cpu_utils.cc:128] Failed to get CPU frequency: 0 Hz
Epoch 1/5
2023-01-07 17:01:17.101099: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:114] Plugin optimizer for device_type GPU is enabled.
2023-01-07 17:01:17.538659: W tensorflow/core/framework/op_kernel.cc:1830] OP_REQUIRES failed at xla_ops.cc:418 : NOT_FOUND: could not find registered platform with id: 0x11aeca870

Boost

Answer 1

Kearel OP

Jan ’23

Hi!

I've been facing this issue for a few days and it worked when I changed the optimizer from

model.compile(optimizer="adam", loss=loss_fn, metrics=["accuracy"])

to:

model.compile(optimizer=tf.keras.optimizers.legacy.Adam(), loss=loss_fn, metrics=["accuracy"])

Got the fix from here: https://developer.apple.com/forums/thread/722873.

Quoting the other post:

Is this on the latest wheels with tensorflow-macos==2.11 and tensorflow-metal==0.7.0? In that case has to do with recent changes on tensorflow side for version 2.11 where a new optimizer API has been implemented

Hope that helps!

0

Answer 2

tseliot OP

Jan ’23

Yes. I've just tested with the latest wheels for Tensorflow-Macos (2.11) and Tensorflow-Metal (.7). The only catch is that you need to use Python 3.9 in your environment. For some reason, Python 3.10 and 3.11 don't seem to be working.

Steps I used (after mini forge installation).

-- conda create -n tf

-- conda activate tf

-- conda install python==3.9

-- pip install tensorflow-macos

-- pip install tensorflow-metal

As mentioned above, you do need to change the test script provided to account for the deprecation of the Adam optimizer. After doing that, everything worked great for me.

0