Closed zeroeightysix closed 2 days ago
spacy has been struggling to fix numpy issues since the v2 release a while ago (i.e. issue 13528).
A quick google search suggests that downgrading numpy should fix this, but numpy is already on 1.26.4 in my venv
From what I understand you should do the opposite of that, since the current spacy 3.12 python version supports numpy 2, iirc.
Have you tried to do a clean install of the virtual environment?
From what I understand you should do the opposite of that, since the current spacy 3.12 python version supports numpy 2, iirc.
I tried doing this, but unfortunately I can't install numpy 2, as the oldest numpy2-compatible version of thinc (==8.3.0
) is incompatible with the latest spaCy (thinc<8.3.0
). This is kind of puzzling, as that spaCy release advertises numpy 2 support..
I guess this is the problem described in https://github.com/explosion/spaCy/issues/13607 .
Have you tried to do a clean install of the virtual environment?
I have, but unfortunately this still installs the old version of numpy.
I suppose we need to wait for a new spaCy release to resolve the version conflicts :/
It's quite a mess, I don't envy the explosion devs right now...
I suppose we need to wait for a new spaCy release to resolve the version conflicts :/
If you don't want to wait, you could downgrade the spacy version to something that is compatible with the other packages. So, maybe something like this?
python3.12 -m venv spacyenv
source spacyenv/bin/activate
python -m pip install --upgrade pip setuptools wheel
python -m pip install numpy==1.26
python -m pip install thinc==8.2.5
python -m pip install spacy==3.7.5
...
Haven't tested that particular permutation though.
@zeroeightysix does it work with the latest spacy release?
@zeroeightysix does it work with the latest spacy release?
Hi, yes, it does!
I recreated the venv, and it worked out of the box again.
Awesome! Let me know if you have any other problems :)
Probably a downstream issue, but I'm having a hard time figuring out why this is happening specifically.
Trying to open the anki-morphs settings or doing a recalc will yield the following error:
A quick google search suggests that downgrading numpy should fix this, but
numpy
is already on1.26.4
in my venv:Unlike last time (#239), I do not have
python-spacy
orpython-thinc
installed system-wide. The python version of the venv is the same as the one Anki is running under.Any clues?