you're running out of memory. Binarize your phrase table

On 09/09/2013 22:34, rohit dholakia wrote:
Hi,

Alright, my phrase table is just 100MB compressed. And after taking quite long to read the phrase table, i got an exception bad_alloc. I have quite a bit of RAM, so, i am not sure what I am doing wrong .

Start loading text SCFG phrase table. Moses  format : [51.000] seconds
Reading /global/scratch/rd/Thesis/TestingMoses/train/model/phrase-table.gz
----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100
****************************************************************************************************
Exception: std::bad_alloc

Are the changes in the master branch mostly related to an updated moses file format ? Are there any changes in the algorithms(mert/ibm models) that warrant a re-run of tests done using the previous version ?




On Mon, Sep 9, 2013 at 1:24 PM, rohit dholakia <[email protected] <mailto:[email protected]>> wrote:

    Hi

      I have recently managed to install moses from the master branch
    on github.  The file format has changed. The training and mert has
    gone without incident. But, i have observed that reading the
    phrase table is taking very very long compared to the previous
    moses. From mert.out and from translating test data log,

    /Loading table into memory...done./
    /Start loading text SCFG phrase table. Moses  format : [51.000]
    seconds/
    /Reading
    /global/scratch/rd/Thesis/TestingMoses/train/model/phrase-table.gz/
    
/----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100/


    Now, this step, this step is taking really long. Am I the only one
    who is observing this ? My phrase table is trained on only 110K
    sentences, quite noisy ones at that. Has only 4.8 million rules.

    /[rd@seawolf1 TestingMoses]$ zcat train/model/phrase-table.gz | wc -l/
    /4838810/

    To explain the time taken, its been more than 5 minutes and not
    even a single star has come up below that last line ( ---- 5 ---
    10 ) .

    Thanks.





_______________________________________________
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

Reply via email to