A word embedding algorithm that can fit into the memory without features loss: for better performance and memory optimization

[HebaGamalElDin]

what examples for a word embeddings algorithms that cares about the memory size without any features loss?
Hint: TFIDF with my data of the shape (500000 Row) gives an array of shape (500000, 44754) and it gives memory allocation error, if i specify a max_features paramter so i'll lose very important features in my case (i can't ignore even any word in the dataset)