Error when running this tensorflow-1.1.0 version

[x-zho14]

``$ python -m basic.cli --mode train --noload --debug Traceback (most recent call last): File "anaconda2/envs/tensorflow/lib/python3.5/runpy.py", line 193, in _run_module_as_main "main", mod_spec) File "anaconda2/envs/tensorflow/lib/python3.5/runpy.py", line 85, in _run_code exec(code, run_globals) File "/bi-att-flow/basic/cli.py", line 5, in from basic.main import main as m File "/bi-att-flow/basic/main.py", line 14, in from basic.model import get_multi_gpu_models File "/bi-att-flow/basic/model.py", line 6, in from tensorflow.python.ops.rnn_cell import BasicLSTMCell ImportError: No module named 'tensorflow.python.ops.rnn_cell'

I have installed the tensorflow-gpu 1.1.0 version and python 3.5.3 and all the required libraries.

It may seem some error with the code version and tensorflow, how can I solve this problem?

Well, it's hard to find a tensorflow-gpu 0.11 version available now so I have to run the 1.1.0 version.

Thanks a lot!

[neufang]

You should use dev branch to run the tensorflow 1.1 version; for master branch, use tensorflow 0.12

[x-zho14]

Thanks for your help! From this I come to know that I should not use git clone for the other branch.

[x-zho14]

I encounter another error when turning to the dev branch. (tensorflow-1.1.0) /bi-att-flow-1.1.0/bi-att-flow-dev$ python -m basic.cli --mode train --noload --debug Traceback (most recent call last): File "envs/tensorflow-1.1.0/lib/python3.5/runpy.py", line 193, in _run_module_as_main "main", mod_spec) File "anaconda2/envs/tensorflow-1.1.0/lib/python3.5/runpy.py", line 85, in _run_code exec(code, run_globals) File "/bi-att-flow-1.1.0/bi-att-flow-dev/basic/cli.py", line 5, in from basic.main import main as m File "/bi-att-flow-1.1.0/bi-att-flow-dev/basic/main.py", line 14, in from basic.model import get_multi_gpu_models File "/bi-att-flow-1.1.0/bi-att-flow-dev/basic/model.py", line 10, in from my.tensorflow.nn import softsel, get_logits, highway_network, multi_conv1d File "/bi-att-flow-1.1.0/bi-att-flow-dev/my/tensorflow/nn.py", line 1, in from tensorflow.python.ops.rnn_cell_impl import _linear ImportError: cannot import name '_linear' And there isn't a _linear module in tensorflow 1.1.0 version, as is shown in https://github.com/tensorflow/tensorflow/blob/r1.1/tensorflow/python/ops/rnn_cell_impl.py. How can I solve this problem?

[neufang]

check the tensorflow version, you might have installed 1.2, which is different from 1.1 (nice job google). change the first line of requirements.txt to tensorflow-gpu ==1.1.0

[daisyjack]

I met this problem too. I indeed used tensorflow 1.1.0, but there isn't a _linear module in tensorflow 1.1.0 version, as is shown in https://github.com/tensorflow/tensorflow/blob/r1.1/tensorflow/python/ops/rnn_cell_impl.py

[neufang]

@x-zho14 maybe it's because of anaconda package? I use default python3.5 with tensorflow 1.1.0 and virtualenv. It doesnt have this problem.

[usakey]

@x-zho14 @daisyjack As per this commit: https://github.com/allenai/bi-att-flow/commit/3fb0943fefd71eea6886dd247d373f46054e922f , it seems this _linear was changed for tf-1.2 compatibility, but on tf-1.1 it doesn't work for me either.

So my workaround is to revert back to previous one:

from tensorflow.contrib.rnn.python.ops.core_rnn_cell_impl import _linear
...
flat_out = _linear(flat_args, output_size, bias, bias_start=bias_start)

this works for me.

[huan]

@usakey Your solution not work under tensorflow v1.4

This fix it for me: from tensorflow.contrib.rnn.python.ops.rnn_cell import _Linear

[abeer-khan]

For anyone starting to try to make sure this works, the following steps did for me:

1. Create conda env with python version 3.5

2. Install tensorflow version 0.12.1 after activating env using: export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.12.1-cp35-cp35m-linux_x86_64.whl (or any appropriate version) pip install $TF_BINARY_URL

3. Edit the basic/run_single.sh file to use the created env when running by adding this line: source activate tensorflow35

4. Run the program within the env using this bash command bash basic/run_single.sh $HOME/data/squad/dev-v1.1.json single.json

[KevinBetterQ]

from tensorflow.contrib.rnn.python.ops import core_rnn_cell linear = core_rnn_cell._linear fix my problem!! windows10 + tensorflow1.6

[spate141]

@abeerunscore96 Thank you! Finally it worked!

[HuiBinR]

@abeerunscore96 Thank you! It works well!

[valsworthen]

Using the solution of @abeerunscore96 are you sure that the code runs on GPU?

[HuiBinR]

No, the code can't run on GPU, even I use CUDA_VISIBLE_DEVICES. But it can run on CPU. @valsworthen

[valsworthen]

Finally I managed to run the code correctly on GPU

• pip install tensorflow-gpu==1.2 (the version of @abeerunscore96 only installs tensorflow for cpu)
• make sure you have CUDA 8 and cuDNN 5 (or 5.1) and that you correctly setup the .bashrc
• use the dev branch of the repo

Run CUDA_VISIBLE_DEVICES=devicenumber python -m basic.cli --mode train --noload --debug

I didn't have the issue with _linear mentioned in this thread.

I think this would also work for the master branch with tensorflow 0.11 / 0.12 provided you install the gpu version.

[RehamAbobeah]

@valsworthen Please help me! I am new in this area and I try to follow your steps to succeed to run the code, what do you mean by setup the .bashrc'? Waiting to hear from you, I really appreciate your help!

[RehamAbobeah]

@valsworthen I am not sure if it works well or not. Every time I have the following warning which means sensor flow may be unable to work with GPU.
From search, someone suggests solving this issue by following this link: https://github.com/mind/wheels/releases?after=tf1.3-gpu However, the GPU wheel for TensorFlow 1.2 is only working with python 3.6 and our version here for this project is 3.5, How to solve this issue, please?

[shimafoolad]

You can run this branch for the newer version of tensorflow (all errors have been fixed): https://github.com/Vimos/bi-att-flow/tree/tf1.8

[1910066023]

tensorflow版本改为0.12.1