no module named enclib_gpu

[marshallwu1]

Hi,author. I have followed the instructions on your page. I installed PyTorch1.0 and do PIP INstall torch-encoding. Then I got your code by git clone, and run "python setup.py install" with no errors. But when I run "python scripts/prepare_pascal.py" or "import encoding", I got the error like this. It seems that the "enclib_gpu" cannot be built.

import encoding
Traceback (most recent call last):
File "", line 1, in
File "encoding/init.py", line 13, in
from . import nn, functions, parallel, utils, models, datasets, transforms
File "encoding/nn/init.py", line 12, in
from .encoding import *
File "encoding/nn/encoding.py", line 19, in
from ..functions import scaled_l2, aggregate, pairwise_cosine
File "encoding/functions/init.py", line 2, in
from .encoding import *
File "encoding/functions/encoding.py", line 14, in
from .. import lib
File "encoding/lib/init.py", line 27, in
build_directory=gpu_path, verbose=False)
File "/home/wmx4/anaconda3/envs/pytorch1.0/lib/python2.7/site-packages/torch/utils/cpp_extension.py", line 645, in load
is_python_module)
File "/home/wmx4/anaconda3/envs/pytorch1.0/lib/python2.7/site-packages/torch/utils/cpp_extension.py", line 825, in _jit_compile
return _import_module_from_library(name, build_directory, is_python_module)
File "/home/wmx4/anaconda3/envs/pytorch1.0/lib/python2.7/site-packages/torch/utils/cpp_extension.py", line 964, in _import_module_from_library
file, path, description = imp.find_module(module_name, [path])
ImportError: No module named enclib_gpu

The version of PyTorch is 1.0, python 2.7, cuda9.0, and I also installed ninja 1.8.2.

[henanjun]

same problem. The version of pytorch is 1.0, python 3.6, cuda 8.0, ninja 1.8.2. Ubuntu 16.04.5.
I am wondering if I need I try cuda 9.0

[zhanghang1989]

Please try using CUDA 9.2 and let me know if you still have the problem.

[henanjun]

Please try using CUDA 9.2 and let me know if you still have the problem.

Thank you very much. I solved the problem by re-installing CUDA 9.2.

[huanghoujing]

@marshallwu1 Do you still have problem with installation? I find it necessary to

• install pytorch 1.0 from source. This is to avoid Segmentation Fault (core dumped) error, in my trial.
• modify some include clauses of Pytorch-Encoding. This is learnt from this issue.

My installation guide is this and then this.