Keras: run imdb_lstm.py error: 'callkernel_node' is undefined

[xidongbo]

error info:
mod.cu(299): error: identifier ''callkernel_node_3a084fd032db206a67ca8121a24aa32e_0'' is undefined
1 error detected in the compilation of ''C:Users/aa/AppData/Local/Temp/tmpxft_000041d4_00000000-10_mod.cpp1.ii''

when I run mnist_cnn.py, it worked well. But when I run the imdb_lstm.py, It didn't work.

numpy1.12.0
theano0.8.2
pip8.1.2
Anconda2.7
CUDA7.5
cudnn-7.5-windows7-x64-v5.0-ga
keras1.2.1
GCC4.7.0

.theanorc.txt:
[global]
openmp=False
device = gpu
optimizer_including=cudnn
floatX = float32
allow_input_downcast=True
[lib]
cnmem = 0.8
[blas]
ldflags=
[gcc]
cxxflags=-ID:\Anaconda2\MinGW
[nvcc]
fastmath = True
flags=-LD:\Anaconda2\libs
compiler_bindir=D:\VS2012\VC\bin

when I run mnist_cnn.py, it worked well. But when I run the imdb_lstm.py, It didn't work.
Please help me!

[rodrigomfw]

    model.add(LSTM(n_hidden[0], input_shape=(seq_length, X.shape[2]),
                           return_sequences=False,W_regularizer=l2(w_reg)))
    model.fit(X_train, y_train, batch_size=batch_size, nb_epoch=nb_epoch,
              class_weight = {0:1, 1:2*n_int/n_ict},validation_data=(X_test,y_test),callbacks=callbacks)``

This snippet is giving me the same error:

mod.cu(299): error: identifier ''callkernel_node_3a084fd032db206a67ca8121a24aa32e_0'' is undefined
1 error detected in the compilation of ''C:Users/aa/AppData/Local/Temp/tmpxft_000041d4_00000000-10_mod.cpp1.ii''

I am also running on windows 7, Anconda2.7.

[ghost]

I run into the same error with theano backend and latest keras1.2.1. But no error if using keras 1.2.0
running on window10 64bits and Anaconda4.3.0

[xidongbo]

when I add 'optimizer=None' to .theanorc.txt, it worked.But it maybe run too slow.
Maybe it related to CUDNN?
My OS:win10

[xidongbo]

@iamkkk Thank you,it worked when I used keras 1.2.0

[rodrigomfw]

@iamkkk Also worked for me! Thank you so much

[helloabc99]

Thank you so much.