Fairseq: RoBERTa example crashes on libnat import

[nazarov-yuriy]

There is example how to use RoBERTa model on page: https://github.com/pytorch/fairseq/tree/master/examples/roberta

import torch
roberta = torch.hub.load('pytorch/fairseq', 'roberta.large')
roberta.eval()  # disable dropout (or leave in train mode to finetune)

It doesn't work after commit 86857a58bf2919c7bec3c29c58234aa4c434d566 with error

ImportError                               Traceback (most recent call last)
<ipython-input-1-9914d0fa65af> in <module>
      1 import torch
----> 2 roberta = torch.hub.load('pytorch/fairseq', 'roberta.large')
      3 roberta.eval()  # disable dropout (or leave in train mode to finetune)

/usr/local/lib/python3.6/dist-packages/torch/hub.py in load(github, model, *args, **kwargs)
    334     sys.path.insert(0, repo_dir)
    335 
--> 336     hub_module = import_module(MODULE_HUBCONF, repo_dir + '/' + MODULE_HUBCONF)
    337 
    338     entry = _load_entry_from_hubconf(hub_module, model)

/usr/local/lib/python3.6/dist-packages/torch/hub.py in import_module(name, path)
     68         spec = importlib.util.spec_from_file_location(name, path)
     69         module = importlib.util.module_from_spec(spec)
---> 70         spec.loader.exec_module(module)
     71         return module
     72     elif sys.version_info >= (3, 0):

/usr/lib/python3.6/importlib/_bootstrap_external.py in exec_module(self, module)

/usr/lib/python3.6/importlib/_bootstrap.py in _call_with_frames_removed(f, *args, **kwds)

~/.cache/torch/hub/pytorch_fairseq_master/hubconf.py in <module>
      6 import functools
      7 
----> 8 from fairseq.hub_utils import BPEHubInterface as bpe  # noqa
      9 from fairseq.hub_utils import TokenizerHubInterface as tokenizer  # noqa
     10 from fairseq.models import MODEL_REGISTRY

~/.cache/torch/hub/pytorch_fairseq_master/fairseq/__init__.py in <module>
      8 
      9 import fairseq.criterions  # noqa
---> 10 import fairseq.models  # noqa
     11 import fairseq.modules  # noqa
     12 import fairseq.optim  # noqa

~/.cache/torch/hub/pytorch_fairseq_master/fairseq/models/__init__.py in <module>
    126     if not file.startswith('_') and not file.startswith('.') and (file.endswith('.py') or os.path.isdir(path)):
    127         model_name = file[:file.find('.py')] if file.endswith('.py') else file
--> 128         module = importlib.import_module('fairseq.models.' + model_name)
    129 
    130         # extra `model_parser` for sphinx

/usr/lib/python3.6/importlib/__init__.py in import_module(name, package)
    124                 break
    125             level += 1
--> 126     return _bootstrap._gcd_import(name[level:], package, level)
    127 
    128 

~/.cache/torch/hub/pytorch_fairseq_master/fairseq/models/insertion_transformer.py in <module>
      7 import torch
      8 import torch.nn.functional as F
----> 9 from fairseq import libnat
     10 from fairseq.models import register_model, register_model_architecture
     11 from fairseq.models.levenshtein_transformer import (

ImportError: cannot import name 'libnat'

[tuhinjubcse]

Kinda getting same

en2de = torch.hub.load('pytorch/fairseq', 'transformer.wmt19.en-de', checkpoint_file='model1.pt:model2.pt:model3.pt:model4.pt',tokenizer='moses', bpe='fastbpe')
Using cache found in /home/tuhinusc577/.cache/torch/hub/pytorch_fairseq_master
Traceback (most recent call last):
File "", line 1, in
File "/home/tuhinusc577/miniconda3/lib/python3.7/site-packages/torch/hub.py", line 359, in load
hub_module = import_module(MODULE_HUBCONF, repo_dir + '/' + MODULE_HUBCONF)
File "/home/tuhinusc577/miniconda3/lib/python3.7/site-packages/torch/hub.py", line 72, in import_module
spec.loader.exec_module(module)
File "", line 728, in exec_module
File "", line 219, in _call_with_frames_removed
File "/home/tuhinusc577/.cache/torch/hub/pytorch_fairseq_master/hubconf.py", line 8, in
from fairseq.hub_utils import BPEHubInterface as bpe # noqa
File "/home/tuhinusc577/.cache/torch/hub/pytorch_fairseq_master/fairseq/__init__.py", line 10, in
import fairseq.models # noqa
File "/home/tuhinusc577/.cache/torch/hub/pytorch_fairseq_master/fairseq/models/__init__.py", line 128, in
module = importlib.import_module('fairseq.models.' + model_name)
File "/home/tuhinusc577/miniconda3/lib/python3.7/importlib/__init__.py", line 127, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "/home/tuhinusc577/.cache/torch/hub/pytorch_fairseq_master/fairseq/models/insertion_transformer.py", line 9, in
from fairseq import libnat
ImportError: cannot import name 'libnat' from 'fairseq' (/home/tuhinusc577/.cache/torch/hub/pytorch_fairseq_master/fairseq/__init__.py)

[hypnoai]

Same for me.
It's only been happening for the last 2-3 days

  7 import torch
  8 import torch.nn.functional as F

----> 9 from fairseq import libnat
10 from fairseq.models import register_model, register_model_architecture
11 from fairseq.models.levenshtein_transformer import (

ImportError: cannot import name 'libnat'

[lematt1991]

Hmm, I'm unable to repro this. How are you installing fairseq? I did

$ git pull
$ pip install --editable .
$ python
Python 3.6.6 |Anaconda, Inc.| (default, Jun 28 2018, 17:14:51)
[GCC 7.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from fairseq import libnat
>>> from fairseq import models
>>>

[ebagdasa]

I just cleaned the cache .cache/torch/hub/pytorch_fairseq_master and installed upstream version.

[pasikon]

Same for me, crashing on libnat import

[myleott]

The problem is that torch.hub doesn’t build the extensions causing the import to fail.

This should be fixed now: https://github.com/pytorch/fairseq/commit/acb6fba005f45e363a6da98d7ce79c36c011d473

[nazarov-yuriy]

I can confirm that example works correctly after clearing ~/.cache/torch

[vince-lynch]

This issue has returned.

[allhelllooz]

facing the same issue ... removing cache file doesn't help.

[RingoS]

same issue here. It happened after I re-installed fairseq with some changes making on the source code.

[RingoS]

I found the solution. Just run python setup.py build_ext --inplace in the fairseq source folder and all extensions will be built. Besides, if you cannot use libnat_cuda, you need to set a CUDA_HOME path.

[zhenpingli]

I found the solution. Just run python setup.py build_ext --inplace in the fairseq source folder and all extensions will be built. Besides, if you cannot use libnat_cuda, you need to set a CUDA_HOME path.

3Q for you help.This have take hours of my time .
FIRST shoud do python setup.py build_ext --inplace and pip install --editable . at index
Second MUST set a CUDA_HOME path
the order is:
put
export PATH=/usr/lib/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/lib/cuda/bin/lib64:$LD_LIBRARY_PATH
in ./bashrc
export CUDA_HOME=/usr/lib/cuda
lib/cuda is nvida driver path
Final source ~/.bashrc