Hi, I've successfully implemented a baseline system sv-fr on a box with 8 GB too, but it was a separate installation on a separate hard disk : dual boot Windows - Debian. Training was quite slow, though. Not to speak about tuning! Virtual memory was used a lot and that's really slow. Upgrading to 16 GB wasn't enough to load all files in physical memory, but improved performance. Pruning the translation table after training improved performance considerably.
I guess running on a virtual system will further slow things down, but I don't see any reason why it shouldn't work. I suggest you start building a really small system, with some thousand sentences to train on. Or why not try the synthetic Block Word Corpus, originally from Uppsala University. You'll find it at my site www.tunedal.nu/download. Comments in my blogg www.tradera.tunedal.nu Yours, Per Tunedal On Mon, Sep 23, 2013, at 9:47, Barry Haddow wrote: > Hi Wei > > The baseline system is relatively small, as MT systems go. I built it on > a laptop with 8G RAM, but it may well build with less RAM, > > cheers - Barry > > On 23/09/13 01:22, Wei JIANG [PT-COM] wrote: > > Thank you for your info, Ken. Should it help if I had a large and fast > > disk, say a SSD? > > > > On 9/23/2013 6:48 AM, Kenneth Heafield wrote: > >> Hi, > >> > >> The kernel kills process when it runs out on memory. Run dmesg and you > >> will likely see evidence to this effect. As to how much memory you > >> need, I don't know for the baseline, but I will say that 32 GB is the > >> lowest I've run experiments on in the past couple years. > >> > >> Kenneth > >> > >> On 09/22/13 18:35, Wei JIANG [PT-COM] wrote: > >>> Hello, > >>> > >>> In the last few weeks, I have been playing with the "Baseline", on a > >>> Ubuntu virtual machine hosted by Virtualbox on Windows 7 64bit with 8GB > >>> RAM and 4-core CPU, and 200G allocated HD space. How, I am stuck in the > >>> following places: > >>> > >>> ~/mosesdecoder/bin/processLexicalTable \ > >>> -in train/model/reordering-table.wbe-msd-bidirectional-fe.gz \ > >>> -out binarised-model/reordering-table > >>> > >>> > >>> This runs several minutes, and then it starts to draw dots, one by one, > >>> before its suddenly exits to the prompt, without any warning. Then I > >>> try the following > >>> > >>> ~/mosesdecoder/bin/moses -f ~/working/mert-work/moses.ini > >>> > >>> The program runs for quite a while and then declares something like > >>> "killed" and exits. > >>> > >>> I have even tried the binarised-model, with moses.ini modified > >>> accoringly, but it exits, warning about missing "reordering-table". > >>> > >>> I have searched online and it seems that my RAM is too low. Is that so. > >>> If so, how much RAM should I have? > >>> > >>> Kindest, > >>> Wei > >>> > >>> _______________________________________________ > >>> 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 > >> > > _______________________________________________ > > Moses-support mailing list > > [email protected] > > http://mailman.mit.edu/mailman/listinfo/moses-support > > > > > -- > The University of Edinburgh is a charitable body, registered in > Scotland, with registration number SC005336. > > _______________________________________________ > 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
