Jan Helak pisze: > Hello, >> Hi, >> >> 4 minutes sounds like too much for translating a sentence. >> A little bit more detail would be helpful to detect where the >> time savings could be: >> - does the time include the start-up time (loading translation >> and language models)? >> - do you use IRSTLM, SRILM, binarized phrase and reordering tables? >> > I just went through a tutorial http://www.statmt.org/moses_steps.html > into end of Part II. > I used SRILM to build a language model. > > > More information about time: > Finished loading LanguageModels : [46.000] seconds > Finished loading phrase tables : [213.000] seconds > End. : [213.000] seconds
You definitely need to use IRSTLM and binary models if you want to load faster. And binarize the phrase table. The numbers you quote are quite typical for relatively small text-based phrase tables and language models (tutorial doesn't cover binarization, afair). Most of the time text-based phrase tables don't even fit in the whole virtual memory, so you are lucky not to get a segfault. To test preliminary results, you can filter your table re: your input data. regards Marcin _______________________________________________ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support
