Dataset for train the system

[quandb]

Hi,
I'm trying to update the word2vec and doc2vec models with the newer version from wiki.
I see there are several kind of wiki data such as: pages-meta-history*, pages-meta-current*, pages-articels*.
I was wondering that which kind of data you use for build models?
Because when I run then word2vec_phrases.py, I just got the size of output file ~12M (compare with your 62M).

Thanks,