keras: TypeError: Fail to find the dnn implementation.

Platform: Windows10 Tensorflow Version: 1.7.0（GPU） Cuda compilation tools, release 9.0, V9.0.176 CUDNN： 7.1.2 Graphic processor: Nvidia Geforce GTX 1050

My code:

from keras.layers import CuDNNLSTM,Bidirectional
lstmsize=6
lstm0 = CuDNNLSTM(lstmsize,return_sequences = True)

Error:

UnknownError (see above for traceback): Fail to find the dnn implementation.
[[Node: cu_dnngru_1/CudnnRNN = CudnnRNN[T=DT_FLOAT, direction="unidirectional", dropout=0, input_mode="linear_input", is_training=true, rnn_mode="gru", seed=87654321, seed2=0, _device="/job:localhost/replica:0/task:0/device:GPU:0"](cu_dnngru_1/transpose, cu_dnngru_1/ExpandDims_1, cu_dnngru_1/Const_1, cu_dnngru_1/concat)]]
[[Node: loss/mul/_73 = _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_618_loss/mul", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"]]

Hopefully for help!

About this issue

Original URL
State: closed
Created 6 years ago
Reactions: 6
Comments: 24

Links to this issue

Having trouble in running RNN LSTM model with tensorflow 2.0

Most upvoted comments

Try this. It works

gpus = tf.config.experimental.list_physical_devices('GPU')
for gpu in gpus:
        tf.config.experimental.set_memory_growth(gpu, True)
tf.config.experimental.set_virtual_device_configuration(gpus[0], [tf.config.experimental.VirtualDeviceConfiguration(memory_limit=1024)])

+20

Shekhrozx on Apr 3, 2020

I ended up fixing this issue with the allow_growth = True comment on https://github.com/tensorflow/tensorflow/issues/24496

+11

kyleabeauchamp on Mar 6, 2019

maybe u can solve it by “tf.config.experimental.set_memory_growth()”！！！

tricyzhou on Mar 23, 2020

for cuda 7.1.1 and cudnn 9.0 : CuDNNLSTM, or CuDNNGRU ran successfully, then after some days gave the same error. Fixed : reinstalling cuda and cudnn.

There has to be some other better solution. This way is too tiresome and lengthy!!

ASH1998 on Nov 9, 2018

unsubscribe please

zyu511008@hotmail.com

From: Krishna Bhogaonkermailto:notifications@github.com Date: 2019-04-27 02:33 To: keras-team/kerasmailto:keras@noreply.github.com CC: Subscribedmailto:subscribed@noreply.github.com Subject: Re: [keras-team/keras] TypeError: Fail to find the dnn implementation. (#10634)

I got this error last night while working on the tensorflow tutorial “https://www.tensorflow.org/alpha/tutorials/load_data/text”. I was using tensorflow-gpu 2.0alpha on an Ubuntu 18.04x64 machine and python version 3.6. I updated my Cudnn from 7.4 to 7.5.1 and tried up upgrade tensorflow too–but that did not change anything. I was able to compile the Cudnn samples Mnist network–which is the usual test for a successful install. Just wanted to let you know about the continuing issue.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/keras-team/keras/issues/10634#issuecomment-487157424, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AKXHW4SYRLENBQXLRNXPYU3PSNDGDANCNFSM4FJDZWHA.

zyu511008 on Apr 27, 2019

I have the same problem under Linux Ubuntu 16.04

Maybe this helps:

https://devtalk.nvidia.com/default/topic/1030610/cuda-setup-and-installation/fail-to-find-the-dnn-implementation-/

ChristofHenkel on Jul 14, 2018

I solve this problem using this way: physical_devices = tf.config.list_physical_devices('GPU') tf.config.experimental.set_memory_growth(physical_devices[0], enable=True)

sergio12S on Jun 17, 2020

Same problem on Ubuntu 18.04.1 LTS running Cuda V9.0.176 and cuDNN 7.2.1. Ditto on RHEL 7.4 with Cuda V9.0.176 and cuDNN 9.0-v7

duhaime on Oct 20, 2018

I fixed this issue by upgrading cuddn from 7.0 to 7.5. I am using cuda10.1 and tf-gpu1.14 on Ubuntu 16.04.

VertexC on Jun 10, 2019