nlpub/word2vec-pyro4 repository overview

⁠Word Vectors Served via Pyro4.

This is a simple network service that serves the Gensim⁠'s KeyedVectors via Pyro4⁠.

⁠Running

$ ./server.py
usage: server.py [--id ID] [--no-sims] [-h HOST] [-p PORT] w2v

The only mandatory parameter, w2v, specifies the path to the word vectors in the word2vec⁠ format. By default, the server listens to any network interface on the port 9090.

For instance, the following pre-trained vectors are available:

• English: Google News⁠, GloVe⁠, fastText⁠;
• Russian: Russian Distributional Thesaurus⁠, RusVectōrēs⁠.

Note that the Docker image is based on Anaconda 3⁠ for performance reasons. The image is designed under the assumption that the vectors are mounted as /usr/src/app/w2v.bin.

⁠Accessing

import Pyro4

Pyro4.config.SERIALIZER = 'pickle' # see the Disclaimer

w2v = Pyro4.Proxy('PYRO:w2v@localhost:9090')

w2v.word_vec('cat') # => array([…], dtype=float32)

⁠Disclaimer

Note that this service should be running in a trusted environment since it uses the pickle serializer to handle the NumPy⁠ arrays. This makes the system extremely vulnerable and even allows arbitrary code execution, which is not recommended⁠ by the Pyro4 developers.

⁠Copyright

Copyright (c) 2017–2019 Dmitry Ustalov⁠. See LICENSE for details.