FS#66790 - [python-tensorflow-cuda] Can't interface with cudnn

Attached to Project: Community Packages
Opened by Andrew (thelongdivider) - Tuesday, 26 May 2020, 16:41 GMT
Last edited by freswa (frederik) - Wednesday, 27 May 2020, 00:43 GMT
Task Type Bug Report
Category Packages
Status Closed
Assigned To Sven-Hendrik Haase (Svenstaro)
Konstantin Gizdov (kgizdov)
Architecture All
Severity Low
Priority Normal
Reported Version
Due in Version Undecided
Due Date Undecided
Percent Complete 100%
Votes 0
Private No

Details

Description: If you run any model that contains either convolutional or recurrent NNs, errors are raised:

Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR
W tensorflow/core/framework/op_kernel.cc:1753] OP_REQUIRES failed at cudnn_rnn_ops.cc:1510 : Unknown: Fail to find the dnn implementation.
Failed to find the dnn implementation


Additional info:
* package version(s)
* config and/or log files etc.
* link to upstream bug report, if any

Steps to reproduce:

model = Sequential()
model.add(Embedding(maxFeatures, 32))
model.add(LSTM(32))
# model.add(SimpleRNN(32, return_sequences=True))
# model.add(SimpleRNN(32, return_sequences=True))
model.add(Dense(1, activation='sigmoid'))

model.compile(optimizer='rmsprop', loss='binary_crossentropy', metrics=['acc'])

history = model.fit(inputTrain, yTrain, epochs=10, batch_size=128, validation_split=.2)
This task depends upon

Closed by  freswa (frederik)
Wednesday, 27 May 2020, 00:43 GMT
Reason for closing:  Works for me
Comment by Konstantin Gizdov (kgizdov) - Tuesday, 26 May 2020, 23:42 GMT
I cannot reproduce your error. I ran the attached code and it works fine.

You did not specify which version of 'python-tensorflow-cuda' you are using. Please give us more detail. Also, did you perhaps install 'tensorflow-gpu' through pip?
   test.py (0.7 KiB)
Comment by Andrew (thelongdivider) - Wednesday, 27 May 2020, 00:42 GMT
It seems to be working after a restart. May have been my mistake. Sorry.

Loading...