xml looks ok to me running:
echo "<zone> <wall/> ( <wall/> en <wall/> ) <wall/> </zone> es ist
beschamend fur die europaische union , dass sie die technische
zusammenarbeit mit russland trotz des volker mordes am tschetschenischen
volk unterstutzt , und das zu einem zeit punkt , da dieses haus die ende
fassung" |
~/workspace/github/mosesdecoder.perf/contrib/server/client-stdin.perl
I don't have a multithreaded client, but running the command line client
multiple times looks like it saturates the server cores. This is how I run
1 of my client code:
~/workspace/github/mosesdecoder.perf/contrib/server/client-stdin.perl <
in.200k
Hieu Hoang
http://moses-smt.org/ <http://www.hoang.co.uk/hieu>
On 27 December 2016 at 03:21, Mike Ladwig <mdlad...@gmail.com> wrote:
> I have made no progress on the multithreading of moses2 in server mode.
> When processing a file specified via -i it appears to respect the threads
> setting in the .ini file
>
> I am fairly confident that the crash is related to decoding zone and wall
> tags.
>
> If I eliminate that element from my pipeline moses2 seems to work
> correctly. An example is:
>
> <zone> <wall/> ( <wall/> en <wall/> ) <wall/> </zone> es ist beschamend
> fur die europaische union , dass sie die technische zusammenarbeit mit
> russland trotz des volker mordes am tschetschenischen volk unterstutzt ,
> und das zu einem zeit punkt , da dieses haus die ende fassung
>
> Here is some information on how I've setup moses2:
>
> moses2 -server -search-algorithm 1 -cube-pruning-pop-limit 2000 -s 2000
> -placeholder-factor 1 -xml-input exclusive --server-port 8082 -config
> /opt/models/xlate.ger-eng/decoder.ini
>
>
> # input factors
> [input-factors]
> 0
>
> # mapping steps
> [mapping]
> 0 T 0
>
> [distortion-limit]
> 6
>
> # feature functions
> [feature]
> UnknownWordPenalty
> WordPenalty
> PhrasePenalty
> ProbingPT name=TranslationModel0 num-features=4
> path=/opt/models/xlate.ger-eng/pt2 input-factor=0 output-factor=0
> LexicalReordering name=LexicalReordering0 num-features=6
> type=wbe-msd-bidirectional-fe-allff input-factor=0 output-factor=0
> property-index=0
> Distortion
> KENLM name=LM0 factor=0 path=/opt/models/xlate.ger-eng/languagemodel-en
> order=5
>
> # dense weights for feature functions
>
> [placeholder-factor]
> 1
>
> [threads]
> 8
>
> [xml-input]
> exclusive
> [weight]
>
> LexicalReordering0= 0.095698 0.0346659 0.111703 0.0987009 0.112696 0.153312
> Distortion0= 0.00145445
> LM0= 0.0998281
> WordPenalty0= -0.105109
> PhrasePenalty0= 0.0459613
> TranslationModel0= 0.0322297 0.0625015 0.0378206 0.00831904
> UnknownWordPenalty0= 1
>
>
> On Mon, Dec 26, 2016 at 6:45 PM, Mike Ladwig <mdlad...@gmail.com> wrote:
>
>> Unfortunately I can't do that easily. The client code has worked with
>> mosesserver for years, as has the wall/zone code.
>>
>> Since I'm not certain that these are moses2 issues yet, I just wanted to
>> confirm that I'm not missing something on either issue -- to confirm that
>> the moses2 server mode is supposed to work as mosesserver did (allocate a
>> new thread to each incoming service request) without any additional
>> configuration and that the wall/zone features are implemented and expected
>> to work.
>>
>> Thanks,
>> mike.
>>
>> On Mon, Dec 26, 2016 at 4:11 PM, Hieu Hoang <hieuho...@gmail.com> wrote:
>>
>>> Hiya
>>>
>>> Can you provide a model, inputs, and your client code so I can reproduce
>>> the issues.
>>>
>>> Hieu
>>> Sent while bumping into things
>>>
>>> On 26 Dec 2016 20:29, "Mike Ladwig" <mdlad...@gmail.com> wrote:
>>>
>>> Hello. I was experimenting a bit with moses2 as a replacement for
>>> mosesserver and have a couple of issues I am looking at.
>>>
>>> First, I don't seem to get multithreading; max cpu is 100% even though
>>> the XML client is using 8 threads. Is there something I need to change to
>>> enable multithreading?
>>>
>>> Second, it seems to crash on sentences with zones and walls. Are they
>>> implemented in moses2?
>>>
>>> Thanks, mike.
>>>
>>> _______________________________________________
>>> Moses-support mailing list
>>> Moses-support@mit.edu
>>> http://mailman.mit.edu/mailman/listinfo/moses-support
>>>
>>>
>>>
>>
>
_______________________________________________
Moses-support mailing list
Moses-support@mit.edu
http://mailman.mit.edu/mailman/listinfo/moses-support
