Spacy: Changes to msgpack cause models to fail to load

[jnardone]

How to reproduce the behaviour

With release of msgpack 0.6.0 (earlier today) and this specifically:
https://github.com/msgpack/msgpack-python/issues/295

you can't do something like:
nlp = spacy.load('en_core_web_lg')

which gives:

File "msgpack/_unpacker.pyx", line 187, in msgpack._cmsgpack.unpackb
ValueError: 684830 exceeds max_map_len(32768)

which fails here in spacy:

File "vectors.pyx", line 370, in spacy.vectors.Vectors.from_disk.load_key2row

These lines that interact with webpack/webpack-numpy need to be passed non-default values to process larger models. At https://github.com/explosion/spaCy/blob/master/spacy/vectors.pyx#L370 and elsewhere.

OR
you need to restrict webpack to less than 0.6.0 until fixed.

Your Environment

• Operating System: MacOS, Ubuntu
• Python Version Used: 3.6.7
• spaCy Version Used: 2.0.17
• Environment Information:

[KaNuNSuZoFLu]

same issue here

[honnibal]

I pushed a new version of Thinc pinned to msgpack <0.6.0. I think that should take care of the problem?

[ines]

Also see #2995!

[imatiach-msft]

+1 seeing this as well

[daniel347x]

spacy.load(u'en_core_web_lg') gives the same error, triggered from within msgpack_numpy.py - I think maybe msgpack needs to be updated in spaCy as well?

Thanks ahead of time as I work around this myself temporarily right now so I can get back to work while waiting for an update.

[honnibal]

Reopening #2995 , as I think it's clearer. tl;dr: Fresh installs should work fine. If you're having problems, you can fix your installation with python -m pip install "msgpack<0.6.0"

[lock[bot]]

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.