Hi, can you look at the *.STDERR.digest files? They should contain some patterns that indicate errors that were found in the *.STDERR output.
If the *.STDERR.digest are empty then indeed no errors were found, but then experiment.perl should not have stopped. -phi On Mon, Sep 5, 2011 at 9:21 AM, marco turchi <[email protected]> wrote: > Hi Philipp, > I have checked the STDERR log files but there are not errors. Giza++ has > generated both the A3.final.gz files that the experiment.perl crashed. > > Training data has been cleaned using the clean perl script > > Thanks a lot > Marco > > On Sat, Sep 3, 2011 at 2:54 AM, Philipp Koehn <[email protected]> wrote: >> >> Hi, >> >> GIZA++ operates by default with a sentence length limit of 100, >> which may be a problem here. >> >> But you have to look at the steps/1/TRAINING_run-giza*.1.STDERR >> files to see what triggered experiment.perl to conclude that there is >> an error. >> >> -phi >> >> On Wed, Aug 31, 2011 at 8:47 AM, marco turchi <[email protected]> >> wrote: >> > Dear All, >> > I'm training a model using 6M en-ar parallel sentences. Mgiza produces >> > the >> > two alignement files A3.final.gz, then the experimental enviroment >> > crashes >> > reporting only these messages: >> > number of steps doable or running: 2 >> > executing >> > >> > /nfs/staging/turchmo/WorkingFolder/TranslationModels/en-ar/Descr//steps/1/TRAINING_run-giza.1 >> > via sh (1) >> > executing >> > >> > /nfs/staging/turchmo/WorkingFolder/TranslationModels/en-ar/Descr//steps/1/TRAINING_run-giza-inverse.1 >> > via sh (2) >> > step TRAINING:run-giza-inverse crashed >> > number of steps doable or running: 1 >> > step TRAINING:run-giza crashed >> > number of steps doable or running: 0 >> > . >> > >> > I checked: >> > 1) all the log files but any error is reported; >> > 2) the data looking for "|", but it is not there. >> > 3) maximum length of a sentence, and it is less than 999. >> > 4) possible dos characters >> > 5) both the training files are cleaned with the moses script >> > >> > I have trained other models without any problems. I do not understand >> > which >> > is the problem. Any advice? >> > >> > Thanks a lot >> > Marco >> > >> > _______________________________________________ >> > Moses-support mailing list >> > [email protected] >> > http://mailman.mit.edu/mailman/listinfo/moses-support >> > >> > > > _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
