tensorflow：似乎没有使用 GPU 进行计算

Question

I have downloaded and am using a tensorflow hub model.

The model creation code is:

model_url = "https://tfhub.dev/google/imagenet/mobilenet_v3_large_075_224/feature_vector/5"  # "https://tfhub.dev/google/imagenet/resnet_v2_152/feature-vector/4"

model = tf.keras.Sequential([
    tf.keras.layers.InputLayer(input_tensor=input_tensor),
    hub.KerasLayer(model_url, trainable=do_fine_tuning),
    tf.keras.layers.Dropout(rate=0.2),
    tf.keras.layers.Dense(len(class_names),
                          kernel_regularizer=tf.keras.regularizers.l2(0.0001))
])
model.build((None,) + IMAGE_SIZE + (3,))
model.summary()

model.compile(
    optimizer=tf.keras.optimizers.SGD(learning_rate=0.005, momentum=0.9),
    loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True, label_smoothing=0.1),
    metrics=['accuracy'])

I know I've configured my system correctly as other models use the GPU, and my tensorflow output includes the following:

2022-02-07 14:39:24.415320: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1720] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: NVIDIA GeForce GTX 1080 computeCapability: 6.1
coreClock: 1.7715GHz coreCount: 20 deviceMemorySize: 8.00GiB deviceMemoryBandwidth: 298.32GiB/s
2022-02-07 14:39:24.415404: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.10.1
2022-02-07 14:39:24.425868: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublas.so.10
2022-02-07 14:39:24.426021: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublasLt.so.10
2022-02-07 14:39:24.439115: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcufft.so.10
2022-02-07 14:39:24.441114: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcurand.so.10
2022-02-07 14:39:24.448086: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcusolver.so.10
2022-02-07 14:39:24.452371: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcusparse.so.10
2022-02-07 

However, when looking at my resource usage in task manager my GPU memory is being filled, but there's nothing in my GPU compute (3D) activity. CPU however is doing a lot of work:

Is there something I need to do for this model specifically so that it uses the GPU?

Answer 1

You have to use a MirroredStrategy to tell TF to use the GPU. See Distributed training with Keras and perhaps Solve GLUE tasks using BERT on TPU .