I faced a similar memory related issue as reported by Marjan Van de Kauter on 24 Jan 2012. I was runnning a lemmatization application for Bengali, highly inflected language. CRF++ started to consume huge memory content and then crashed. The training data had around 321 sentences and 1512609 features - as discovered by CRF++.
Please can somebody help? I am adding the training data and the template file. I ran CRF++ 0.57 and 0.55 - both encountered same problem.