word2vec_to_keyedvectors.py

#!/usr/bin/env python3

"""
Convert Google News model to KeyedVectors format:
* We can load it and share among workers using mmap()
* Loads faster
* Uses at least 2x less memory
After generating the new model, zip both generated files into a single archive:
  zip -9 GoogleNews-vectors-negative300.keyedvectors.bin.zip \
      GoogleNews-vectors-negative300.keyedvectors.bin \
      GoogleNews-vectors-negative300.keyedvectors.bin.vectors.npy
Lastly, upload the archive to S3 and update download_models.py accordingly.
"""

from gensim.models import KeyedVectors

if __name__ == '__main__':
    model = KeyedVectors.load_word2vec_format('GoogleNews-vectors-negative300.bin', binary=True)
    model.save('GoogleNews-vectors-negative300.keyedvectors.bin')