I am working on an application with memory constraints. We are getting vectors from python Gensim models but need to transmit copies of them to react native mobile app and potentially in-browser JS. I need to get word2vec word vectors using as much less memory as possible. So, I need some ways in which this can be achieved.

I already tried reducing floating-point precision to 9 floating points and got a 1.5 times improvement on memory. I am okay with compromisation on performance. Can anyone suggest some more alternatives?


1 Answer 1


you actually need to reduce the size of your embedding. Use a nonlinear dimensionality reduction algorithm such as LLE, UMAP or using an Autoencoder and reduce the size of Word2Vec from $n$ to $m$.

Choosing $m$ is done through a simple hyperparametr tuning for your model


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.