如何在Tensorflow中使用keras在2个GPU之间拆分模型？

Question

Essentially, I am looking for something like with tf.device('/device:GPU:0' for keras. I want to put my operations on different GPUs. I am using a Sequential model following lines of

...
model.compile(optimizer='adam', 
                  loss='sparse_categorical_crossentropy',
                  metrics=['accuracy'])
...

model.fit(train_images, train_labels, epochs=5)

Answer 1

Probably you can use with K.tf.device('/gpu:1'): as a context. Or, if the backend is tensorflow, then the way you assign gpu in tf should work for keras also.