Just rewrite train function in gensim.models.word2vec.Word2Vec and gensim.models.doc2vec.Doc2Vec using Keras+Theano
like
class Word2VecKeras(gensim.models.word2vec.Word2Vec):
def train(...And can use GPU via Theano.
pip install word2veckerassame to gensim.models.word2vec.Word2Vec
vsk = Word2VecKeras(gensim.models.word2vec.LineSentence('test.txt'),iter=100)
print( vsk.most_similar('the', topn=5))
from nltk.corpus import brown
brk = Word2VecKeras(brown.sents(),iter=10)
print( brk.most_similar('the', topn=5))pip install --upgrade git+https://github.com/fchollet/keras