Open smallsmallwood opened 6 years ago
I had similar issue. In my case I've run setup_all.sh
then pip install -r requirements.txt
, that installed different version of tensorflow-gpu
.
Solution: I've run setup_all.sh
again (which creates coref_kernels.so
). Now setup and demo had the same version of tensorflow-gpu
, and linking is correct.
Traceback (most recent call last): File "demo.py", line 14, in
import coref_model as cm
File "/home/wlk/PycharmProjects/e2e/coref_model.py", line 10, in
import coref_ops
File "/home/wlk/PycharmProjects/e2e/coref_ops.py", line 4, in
coref_op_library = tf.load_op_library("./coref_kernels.so")
File "/home/wlk/anaconda3/envs/e2e/lib/python2.7/site-packages/tensorflow/python/framework/load_library.py", line 64, in load_op_library
None, None, error_msg, error_code)
tensorflow.python.framework.errors_impl.NotFoundError: ./coref_kernels.so: undefined symbol: _ZN10tensorflow8internal21CheckOpMessageBuilder9NewStringB5cxx11Ev