Error running pre-trained: coref_kernels.so: undefined symbol

amir-zeldes commented 6 years ago

I'm running into a problem running the pre-trained setup on Ubuntu, Python 2.7, TF==1.0.0:

e2e-coref$ python demo.py final
Traceback (most recent call last):
  File "demo.py", line 14, in <module>
    import coref_model as cm
  File "/home/ghost/e2e_coref/e2e-coref/coref_model.py", line 10, in <module>
    import coref_ops
  File "/home/ghost/e2e_coref/e2e-coref/coref_ops.py", line 4, in <module>
    coref_op_library = tf.load_op_library("./coref_kernels.so")
  File "/home/ghost/.local/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

mcjoshi commented 5 years ago

I am getting the same error. @amir-zeldes did you found any solution to this ??

Got it fixed :)

kentonl commented 5 years ago

The latest TF version requires a different set of commands to compile custom ops. The latest commit fixes that.

ereday commented 5 years ago

I am getting the same error. @amir-zeldes did you found any solution to this ??

Got it fixed :)

@mcjoshi, if you still remember, could you explain how you fixed it ? (I'm also trying it on TF 1.0.0 and Cuda8.0)

kentonl / e2e-coref

Error running pre-trained: coref_kernels.so: undefined symbol #25