Undefined symbol __atomic_load_8 error when downloading spacy models on RPI 4

[InputMoe]

When I install the module on Raspberry Pi 4 Linux raspberrypi 5.4.79-v7l+ #1373 SMP Mon Nov 23 13:27:40 GMT 2020 armv7l GNU/Linux it ended up with the same error multiple times:

undefined symbol __atomic_load_8

when downloading a model with

python -m spacy download ...

• Readme.md

  • Step a) ends with undefined symbol __atomic_load_8
  • Step b)
  • compiling with cortexa57 and cortexa53 ended with a gcc error: armv8-a does not support feature 'fp'
  • compiling with cortexa15 and cortex9 succeeded but also ran in the atomic_load_8 error when downloading a model

• stackoverflow (https://stackoverflow.com/questions/59927844/is-it-possible-to-install-spacy-to-raspberry-pi-4-raspbian-buster)

  • compiling with cortexa57 and cortexa53` ended with a gcc error: 'armv8-a does not support feature 'fp', did you mean +nofp'
  • compiling with cortexa15 and cortexa9 succeeded but also ran in the atomic_load_8 error when downloading a model

Are there other ways for a proper installation?

Add: the docker option does not work, because the container is not compatible with the arm architecture

[adrianeboyd]

I'm not sure exactly what leads to these particular errors, but you have a 32-bit distro (armv7l indicates that it's 32-bit) and you will need a 64-bit OS + 64-bit python for spaCy v2 to work correctly. See the notes at the top of this comment for installing / checking that you have a 64-bit OS:

#6021 (comment)

The install instructions in that comment are now a bit out-of-date, too. We recently improved the source distributions for our packages so that the build dependencies are correct and so that blis will choose cortexa57 by default for linux+aarch64, so as of spacy v2.3.5 you should now be able to install spacy in a venv with only:

python3 -m venv .venv
source .venv/bin/activate
pip install --upgrade pip setuptools
pip install spacy

It will take ~20 minutes to compile everything.

You can run the test suite after installing a few additional packages:

pip install pytest pytest-timeout mock jsonschema pyrsistent
pytest --pyargs spacy

The test suite will take ~5 minutes to run.

Alternatively, there are binary linux aarch64 packages on conda-forge. There's a linux aarch64 miniforge installer and the install command would be conda install spacy. If you're not using the miniforge installer, then you need to add -c conda-forge or otherwise add the conda-forge channel.

[InputMoe]

yes this worked!

[InputMoe]

thx!