Spacy 3.0 not working in colab pro

[satbalak]

How to reproduce the behaviour

I have google colab pro and when I do !nvidia-smi, I see that I have been allocated Tesla P100-PCIE.

After installing Spacy 3.0 and installing cupy library, I get the following error when I even try to get the spacy version

!python -m spacy info
Traceback (most recent call last):
  File "/usr/lib/python3.7/runpy.py", line 183, in _run_module_as_main
    mod_name, mod_spec, code = _get_module_details(mod_name, _Error)
  File "/usr/lib/python3.7/runpy.py", line 142, in _get_module_details
    return _get_module_details(pkg_main_name, error)
  File "/usr/lib/python3.7/runpy.py", line 109, in _get_module_details
    __import__(pkg_name)
  File "/usr/local/lib/python3.7/dist-packages/spacy/__init__.py", line 10, in <module>
    from thinc.api import prefer_gpu, require_gpu, require_cpu  # noqa: F401
  File "/usr/local/lib/python3.7/dist-packages/thinc/__init__.py", line 5, in <module>
    from .config import registry
  File "/usr/local/lib/python3.7/dist-packages/thinc/config.py", line 14, in <module>
    import srsly
  File "/usr/local/lib/python3.7/dist-packages/srsly/__init__.py", line 4, in <module>
    from ._msgpack_api import read_msgpack, write_msgpack, msgpack_dumps, msgpack_loads
  File "/usr/local/lib/python3.7/dist-packages/srsly/_msgpack_api.py", line 3, in <module>
    from . import msgpack
  File "/usr/local/lib/python3.7/dist-packages/srsly/msgpack/__init__.py", line 13, in <module>
    from ._msgpack_numpy import encode_numpy as _encode_numpy
  File "/usr/local/lib/python3.7/dist-packages/srsly/msgpack/_msgpack_numpy.py", line 19, in <module>
    import cupy
  File "/usr/local/lib/python3.7/dist-packages/cupy/__init__.py", line 12, in <module>
    _environment._preload_libraries()  # NOQA
  File "/usr/local/lib/python3.7/dist-packages/cupy/_environment.py", line 286, in _preload_libraries
    if (config is None) or (config['packaging'] == 'conda'):
KeyError: 'packaging'

This error did not occur till yesterday when I used the free colab pro :-( ... I suspect this is something

Your Environment

• Operating System: Google Colab with Tesla P100 GPU
• Python Version Used: 3.20.7
• spaCy Version Used: 3
• Environment Information:

!wget https://developer.nvidia.com/compute/cuda/9.2/Prod/local_installers/cuda-repo-ubuntu1604-9-2-local_9.2.88-1_amd64 -O cuda-repo-ubuntu1604–9–2-local_9.2.88–1_amd64.deb

!dpkg -i cuda-repo-ubuntu1604–9–2-local_9.2.88–1_amd64.deb
!apt-key add /var/cuda-repo-9–2-local/7fa2af80.pub
!apt-get update
!apt-get install cuda-9.2

!pip install torch==1.7.1+cu92 torchvision==0.8.2+cu92 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html

!pip install -U spacy[cuda92,transformers]
!export CUDA_PATH=”/usr/local/cuda-9.2"
!export LD_LIBRARY_PATH=$CUDA_PATH/lib64:$LD_LIBRARY_PATH

!pip install -U spacy
!python -m spacy download en_core_web_trf

!pip install cupy

After installing cupy, I see that that Spacy commands start giving the above error. Please help!
It worked fine till yesterday when I was on the free google colab when it was taking a different type of GPU

[satbalak]

Ok, you can ignore this. I figured out the issue. Probably there were some changes at google colab (nothing to do with colab pro or free version). I just followed the instructions at https://spacy.io/usage and upgraded spacy in colab and I was able to train.

pip install -U pip setuptools wheel
pip install -U spacy
python -m spacy download en_core_web_trf

Closing this issue

[polm]

Thanks for following up!

[satbalak]

FWIW, I am recording my findings here
When you start a new google colab, cuda 11 is already installed. But this shows the foll when you choose GPU

pip install -U pip setuptools wheel
pip install -U spacy[cuda102,transformers,lookups]
python -m spacy download en_core_web_trf

But I realized that adding cuda102 (or whatever version) in this command causes an issue. So just exclude this cuda102 from the second command. After this, I did confirm that we are indeed using the GPU by using spacy.prefer_gpu().

[yanqianglu]

Hi, just curious if you are able to enable GPU with spacy v3. I added cuda110 to the pip install command, but for some reason it just failed to enable.

It's strange that if I reset the runtime, Colab seems have spacy v2 and it works well with GPU. So I uninstall and install with v3 with cuda and it just won't work

[polm]

@yanqianglu Based on the above description of the problem, I believe that you can use a GPU on colab with spaCy just fine, but you have to not install cupy yourself. If you don't add cudaXXX to the pip install then the pre-installed cupy, which is special for colab, should be used and just work. If that's not the case let us know.

[github-actions]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.