Hieu, 

Does processPhraseTable binarize hierarchical models? We've
had excellent success running binarized models from multiple SSDs
configured as RAID0. Since the binrized models are read-only, we
configure the environment appropriately to avoid the problems of limited
I/O writes to the SSDs. 

On Thu, 29 Mar 2012 07:53:35 +0700, Hieu Hoang
wrote:  the other issue is that the phrase table for hierarchical
models is 5 to 10 times larger than for pb models. So unless you have a
server with a lot of memory (probably over 20GB), the decoder will spend
most of its time doing IO lookups to find the rules.

 On 28/03/2012
23:48, dongxinghua0213 wrote:  

hi, 
 all,  
 how can I speed up the
decoding process in hierarchical phrase model ? 
 I change the
"pop-limit" from 1000 to 10 in moses.ini , but compared with the
phrase-based model, decoding a sentence with a length of 10 is still
very slow . Are there other methods to speed up the decoding ? 
 the
second question, in ini file : 
[max-chart-span]
 20
 1000 

the first
value 20 is the max length of a span which a rule can be applied to ,
but what is the second value 1000 ? Is it also a length of span ?


_______________________________________________
Moses-support mailing
list
[email protected]
[1]
http://mailman.mit.edu/mailman/listinfo/moses-support [2]

 


Links:
------
[1] mailto:[email protected]
[2]
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