bloom filter occupied 90 % of memory on server with 836Gb available ram

[ethany21]

I have 12Tb of text file for dedup
(with size of each file is 4Gb, about 652050487 tokens within)
and for bff, I configured exepected-ngram-count to be 2001794995090 , fp-rate to be 0.01

And when I started bff, bloom filter occupied 90% of ram and soon system nearly crashed.
Is it better way to divide files into smaller groups and run each group sequentially?