CODEWITHSUNDEEP
pythontensorflowapple-m1
Tanish Shah
Tanish Shah

Reputation: 39

Why after running "pip install tensorflow-metal" is causing error 'StatefulPartitionedCall_10' on apple m1?

The error is caused by the command “python -m pip install tensorflow-metal”

What this does is it boost the performance of compiling model while training them (basically compiles very fast using GPU)

So I tried running models in another environment where I didn’t run this command and they worked fine but took way more time so now I don’t know what to do cause performance is way to bad without using apple’s GPU processing

tf.print("Fit model on training data")
history = model.fit(x_train, y_train, batch_size=32, epochs=10,
    # We pass some validation for
    # monitoring validation loss and metrics
    # at the end of each epoch
    validation_data= [x_val, y_val],
    verbose= 2
)

Error:

NotFoundError Traceback (most recent call last)
    
tf.print("Fit model on training data")
history = model.fit(x_train, y_train, batch_size=32, epochs=10, validation_data= [x_val, y_val], verbose= 2)

NotFoundError: Graph execution error:

Detected at node 'StatefulPartitionedCall_10'

Need help with the metal command

Upvotes: 0

Views: 928

Answers (1)

Tanish Shah
Tanish Shah

Reputation: 39

Update:

So I found out what the problem was, the latest versions of both tensorflow-macos and tensorflow-metal are not compatible with each other, so I tried to install the earlier versions of each and it worked.

Here's a table of version compatibility

So just take care that you download the correct versions.

For example:

if installing:

python -m pip install tensorflow-macos==2.9

then run command

python -m pip install tensorflow-metal==0.5.0

Posting this because any of this is not well documented in either of tensorflow or apple developer's website, Hope this helps!

Upvotes: 1

Related Questions