tensorflow: Tensorflow profiling tool does not work on Mac M1
Click to expand!
Issue Type
Bug
Have you reproduced the bug with TF nightly?
Yes
Source
source
Tensorflow Version
2.11
Custom Code
Yes
OS Platform and Distribution
M1 Mac
Mobile device
No response
Python version
3.9
Bazel version
No response
GCC/Compiler version
No response
CUDA/cuDNN version
No response
GPU model and memory
No response
Current Behaviour?
I cannot do profiling on M1 macbook, but with the same code I can profile with linux machine (tried on colab).
Standalone code to reproduce the issue
import tensorflow as tf
from datetime import datetime
cifar = tf.keras.datasets.cifar100
(x_train, y_train), (x_test, y_test) = cifar.load_data()
model = tf.keras.applications.ResNet50(
include_top=True,
weights=None,
input_shape=(32, 32, 3),
classes=100,
)
logs = "logs/" + datetime.now().strftime("%Y%m%d-%H%M%S")
tboard_callback = tf.keras.callbacks.TensorBoard(
log_dir=logs, histogram_freq=1, profile_batch=(50, 150)
)
loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
optimizer = tf.keras.optimizers.Adam()
legacy_optimizer = tf.keras.optimizers.legacy.Adam()
# switch optimizer to test performance
model.compile(optimizer=optimizer, loss=loss_fn, metrics=["accuracy"])
model.fit(
x_train,
y_train,
epochs=2,
batch_size=16,
steps_per_epoch=200,
callbacks=[tboard_callback],
)
### Relevant log output
_No response_</details>
About this issue
- Original URL
- State: closed
- Created 4 years ago
- Comments: 16 (12 by maintainers)
Hi @mohantym and @chenmoneygithub, thanks for reaching out and your interest . Currently we haven’t implemented the pluggable profiler api for Metal plug-in . Is this something you are interested in?
@mohantym Thanks! I have the same profiling output, and it does generate a file. However, if you check the profile results on Tensorboard you will see nothing gets really traced, please see the screenshot in this comment