Apple silicon m1 맥에 tensorflow설치하고 GPU가속 사용하기

어느날 애플의 텐서플로 공식 깃헙에 들어가보니 archived 되어있었고
이곳 으로 안내된다.

이제 apple silicon용 텐서플로가 알파 버전으로 지원되는것이 아니라 텐서플로 v2.5부터 공식 지원된다.

https://developer.apple.com/metal/tensorflow-plugin/
공식 페이지에 들어가보면 초장부터 조금 황당한 소리를 하고 있다.
macOS 12 라니 지금 현재는 베타버전인 OS를 설치하라고 한다.

그래도 혹시 모르니 한번 해보도록 한다.
프로시져는 공식 페이지의 것을 그대로 따라하도록 한다.

Download and install Conda env:

chmod +x ~/Downloads/Miniforge3-MacOSX-arm64.sh
sh ~/Downloads/Miniforge3-MacOSX-arm64.sh
source ~/miniforge3/bin/activate

Install the Tensorflow dependencies:

conda install -c apple tensorflow-deps

Install base tensorflow:

python -m pip install tensorflow-macos

Install metal plugin:

python -m pip install tensorflow-metal

이제 따로 환경을 만들어서 테스트 해본다:

conda create --clone base -n tensorflow-test

만든 환경을 활성화 하고:

conda activate tensorflow-test

벤치마크용 코드를 실행하기 위해 이것도 설치한다:

pip install tensorflow_datasets

vscode던 아무 코드에디터로 test.py를 만들고 다음 코드를 넣는다:
이곳 에서 벤치마크용으로 쓰이던 코드를 가져왔다.

import tensorflow.compat.v2 as tf
import tensorflow_datasets as tfds

tf.enable_v2_behavior()

from tensorflow.python.framework.ops import disable_eager_execution
disable_eager_execution()


(ds_train, ds_test), ds_info = tfds.load(
    'mnist',
    split=['train', 'test'],
    shuffle_files=True,
    as_supervised=True,
    with_info=True,
)

def normalize_img(image, label):
  """Normalizes images: `uint8` -> `float32`."""
  return tf.cast(image, tf.float32) / 255., label

batch_size = 128

ds_train = ds_train.map(
    normalize_img, num_parallel_calls=tf.data.experimental.AUTOTUNE)
ds_train = ds_train.cache()
ds_train = ds_train.shuffle(ds_info.splits['train'].num_examples)
ds_train = ds_train.batch(batch_size)
ds_train = ds_train.prefetch(tf.data.experimental.AUTOTUNE)


ds_test = ds_test.map(
    normalize_img, num_parallel_calls=tf.data.experimental.AUTOTUNE)
ds_test = ds_test.batch(batch_size)
ds_test = ds_test.cache()
ds_test = ds_test.prefetch(tf.data.experimental.AUTOTUNE)


model = tf.keras.models.Sequential([
  tf.keras.layers.Conv2D(32, kernel_size=(3, 3),
                 activation='relu'),
  tf.keras.layers.Conv2D(64, kernel_size=(3, 3),
                 activation='relu'),
  tf.keras.layers.MaxPooling2D(pool_size=(2, 2)),
#   tf.keras.layers.Dropout(0.25),
  tf.keras.layers.Flatten(),
  tf.keras.layers.Dense(128, activation='relu'),
#   tf.keras.layers.Dropout(0.5),
  tf.keras.layers.Dense(10, activation='softmax')
])
model.compile(
    loss='sparse_categorical_crossentropy',
    optimizer=tf.keras.optimizers.Adam(0.001),
    metrics=['accuracy'],
)

model.fit(
    ds_train,
    epochs=12,
    validation_data=ds_test,
)

실행한다:

python test.py

실행되는 동안 활성상태를 보면

GPU가속을 잘 활용하는것을 볼 수 있다.

본인은 다음과 같이 결과가 나왔는데, 다른 시스템하고 비교해보는것도 재밌을 것이다.

Init Plugin
Init Graph Optimizer
Init Kernel
Metal device set to: Apple M1
2021-07-20 17:32:24.807252: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:305] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support.
2021-07-20 17:32:24.807441: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:271] 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>)
2021-07-20 17:32:24.859240: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:305] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support.
2021-07-20 17:32:24.859261: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:271] 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>)
2021-07-20 17:32:24.865391: W tensorflow/core/platform/profile_utils/cpu_utils.cc:128] Failed to get CPU frequency: 0 Hz
2021-07-20 17:32:24.956177: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:112] Plugin optimizer for device_type GPU is enabled.
2021-07-20 17:32:24.965837: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:112] Plugin optimizer for device_type GPU is enabled.
2021-07-20 17:32:25.012885: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:112] Plugin optimizer for device_type GPU is enabled.
2021-07-20 17:32:25.026958: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:112] Plugin optimizer for device_type GPU is enabled.
2021-07-20 17:32:25.090473: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:112] Plugin optimizer for device_type GPU is enabled.
2021-07-20 17:32:25.104164: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:112] Plugin optimizer for device_type GPU is enabled.
2021-07-20 17:32:25.125363: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:112] Plugin optimizer for device_type GPU is enabled.
2021-07-20 17:32:25.142313: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:112] Plugin optimizer for device_type GPU is enabled.
2021-07-20 17:32:25.159820: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:176] None of the MLIR Optimization Passes are enabled (registered 2)
Train on 469 steps, validate on 79 steps
2021-07-20 17:32:25.175285: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:112] Plugin optimizer for device_type GPU is enabled.
Epoch 1/12
2021-07-20 17:32:25.186795: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:112] Plugin optimizer for device_type GPU is enabled.
468/469 [============================>.] - ETA: 0s - batch: 233.5000 - size: 1.0000 - loss: 0.1561 - accuracy: 0.9540/Users/qone/miniforge3/envs/tensorflow-test/lib/python3.9/site-packages/tensorflow/python/keras/engine/training.py:2426: UserWarning: `Model.state_updates` will be removed in a future version. This property should not be used in TensorFlow 2.0, as `updates` are applied automatically.
  warnings.warn('`Model.state_updates` will be removed in a future version. '
2021-07-20 17:32:35.514217: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:112] Plugin optimizer for device_type GPU is enabled.
469/469 [==============================] - 11s 22ms/step - batch: 234.0000 - size: 1.0000 - loss: 0.1559 - accuracy: 0.9541 - val_loss: 0.0478 - val_accuracy: 0.9848
Epoch 2/12
469/469 [==============================] - 11s 22ms/step - batch: 234.0000 - size: 1.0000 - loss: 0.0437 - accuracy: 0.9866 - val_loss: 0.0416 - val_accuracy: 0.9861
Epoch 3/12
469/469 [==============================] - 11s 22ms/step - batch: 234.0000 - size: 1.0000 - loss: 0.0269 - accuracy: 0.9917 - val_loss: 0.0353 - val_accuracy: 0.9878
Epoch 4/12
469/469 [==============================] - 11s 22ms/step - batch: 234.0000 - size: 1.0000 - loss: 0.0187 - accuracy: 0.9941 - val_loss: 0.0306 - val_accuracy: 0.9898
Epoch 5/12
469/469 [==============================] - 11s 22ms/step - batch: 234.0000 - size: 1.0000 - loss: 0.0133 - accuracy: 0.9958 - val_loss: 0.0389 - val_accuracy: 0.9885
Epoch 6/12
469/469 [==============================] - 11s 22ms/step - batch: 234.0000 - size: 1.0000 - loss: 0.0097 - accuracy: 0.9968 - val_loss: 0.0431 - val_accuracy: 0.9876
Epoch 7/12
469/469 [==============================] - 11s 22ms/step - batch: 234.0000 - size: 1.0000 - loss: 0.0104 - accuracy: 0.9966 - val_loss: 0.0334 - val_accuracy: 0.9899
Epoch 8/12
469/469 [==============================] - 11s 22ms/step - batch: 234.0000 - size: 1.0000 - loss: 0.0047 - accuracy: 0.9984 - val_loss: 0.0359 - val_accuracy: 0.9897
Epoch 9/12
469/469 [==============================] - 11s 22ms/step - batch: 234.0000 - size: 1.0000 - loss: 0.0060 - accuracy: 0.9981 - val_loss: 0.0414 - val_accuracy: 0.9890
Epoch 10/12
469/469 [==============================] - 11s 22ms/step - batch: 234.0000 - size: 1.0000 - loss: 0.0053 - accuracy: 0.9983 - val_loss: 0.0366 - val_accuracy: 0.9906
Epoch 11/12
469/469 [==============================] - 11s 22ms/step - batch: 234.0000 - size: 1.0000 - loss: 0.0042 - accuracy: 0.9987 - val_loss: 0.0415 - val_accuracy: 0.9899
Epoch 12/12
469/469 [==============================] - 11s 22ms/step - batch: 234.0000 - size: 1.0000 - loss: 0.0050 - accuracy: 0.9984 - val_loss: 0.0462 - val_accuracy: 0.9890