15
votes

I can train a Keras network with Dense layer using keras.datasets.fashion_mnist dataset. However, when I tried to train a convolutional network, I got an error.

Here is some part of the code:

from tensorflow.keras.layers import *

model = keras.Sequential([
        Convolution2D(16, (3,3), activation='relu', input_shape=(28,28,1)),
        MaxPooling2D(pool_size=(2,2)),
        Flatten(),
        Dense(16, activation='relu'),
        Dense(10, activation='softmax')    
])
model.compile(optimizer=tf.train.AdamOptimizer(), 
          loss='sparse_categorical_crossentropy',
          metrics=['accuracy'])

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

and its error when I tried to fit.

UnknownError: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above. [[{{node conv2d/Conv2D}} = Conv2D[T=DT_FLOAT, data_format="NCHW", dilations=[1, 1, 1, 1], padding="VALID", strides=[1, 1, 1, 1], use_cudnn_on_gpu=true, _device="/job:localhost/replica:0/task:0/device:GPU:0"](training/TFOptimizer/gradients/conv2d/Conv2D_grad/Conv2DBackpropFilter-0-TransposeNHWCToNCHW-LayoutOptimizer, conv2d/Conv2D/ReadVariableOp)]] [[{{node loss/dense_1_loss/broadcast_weights/assert_broadcastable/AssertGuard/Assert/Switch_2/_69}} = _Recvclient_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_112_l...t/Switch_2", tensor_type=DT_INT32, _device="/job:localhost/replica:0/task:0/device:CPU:0"]]

I have cudnn64_7.dll in C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin and the PATH is already contain that folder.

1

1 Answers

8
votes

I think this link would solve your problem, its because the cnDNN version you installed is not compatible with the cuDNN version that compiled in tensorflow.