fixed test for training (new fastText format)

[prakhar2b]

Fix #1370
For the dataset that we used (lee) vocab_size in older fasttext format was 1762, but for newer format it's 1763. For loading (PR #1319 ), we have two separate tests now (1762 for older, and 1763 for newer).
cc @menshikh-iv

[menshikh-iv]

Thank you, @prakhar2b

[piskvorky]

What was the reason for this bug, what happened?

Are there any more bugs like this?

[menshikh-iv]

@piskvorky It's ok, I speak with @prakhar2b in gitter about it.

for the dataset that we used (lee) vocab_size in older fasttext format was 1762, but for newer format it's 1763. For loading (PR 1319), we have two separate tests (1762 for older, and 1763 for newer).`

Now, FastText saved data in new format, this was a problem of current test.

[piskvorky]

That's good to know.

Try to keep the summary / gist of the explanation on Github too -- the repo is followed by many people, and also indexed by google. Knowledge that's Gitter-only is lost.