crf_learn interrupts after first iteration

[overnastuhed]

I've been using CRF++ quite for a while and never faced such a problem before.

Training fails on some training files. They are successfully read and the training process interrupts suddenly after "iter = 0..." appears. For some training files it helps just to launch training once again and then everything works as expected. But for some files it doesn't help. From what I can see, it happens when the amount of features is about 7 millions. Is such a behaviour somewhat OK for CRF++ or is there something wrong?

UPD. Just checked: the problem does not occur on Windows 7. I face it only with Windows 10.