Hi,
I recently added to experiment.perl an option to first copy all big model
files to local disk before running the decoder.
For this, you just need to set the parameter
cache-model = "/scratch/disk/path"
in the [GENERAL] section.
This works well in our GridEngine setup.
-phi
On Tue, Apr 12,
Hi,
is there any way to track down why it does not produce a translation for
the sentence?
This really should not happen in the phrase-based model...
-phi
On Thu, Apr 14, 2016 at 10:05 AM, Hieu Hoang wrote:
>
> if you're decoding with the normal pb algorithm, there's a an argument
> [stack
I'm not aware of any changes... I'll try to create a minimal example
(ttables + moses.ini) so that you can reproduce it on your machine
(hopefully it's reproducible :)).
Best,
Ales
On Thu, Apr 14, 2016 at 4:10 PM, Marcin Junczys-Dowmunt
wrote:
> Hm, this is not something I can fix out of my hea
Hm, this is not something I can fix out of my head. It's actually weird
this is happening at all, is there anything unusual with those machines?
New compiler?
W dniu 14.04.2016 o 15:08, Aleš Tamchyna pisze:
Hi Marcin,
I ran it several times and the backtrace looked identical in all cases:
#0
Hi Marcin,
I ran it several times and the backtrace looked identical in all cases:
#0 0x0081fd18 in Moses::BitWrapper::Seek
(this=0x7ffc23a5c7f0,
bitPos=18446744073708512633) at
moses/TranslationModel/CompactPT/CanonicalHuffman.h:303
#1 0x0081dd78 in Moses::BitWrapper::SeekF
if you're decoding with the normal pb algorithm, there's a an argument
[stack_diversity]
+positve number
I'm not exactly sure what this does but I think it makes sure that all
coverages has some hypos. This will reduce the risk of the decoder
creating only dead ends.
On 14/04/2016 17:31
load=parallel_read and moving to the local tmp-partition seems to have solved
my problem.
Thanks to all ...
Jörg
> On 12 Apr 2016, at 12:47, Kenneth Heafield wrote:
>
> Did you remove all "lazyken" arguments from moses.ini?
>
> Is the network filesystem Lustre? If so, mmap will perfo
You can probably do it in Manager.cpp line 1583. But you'll need to fill
out the columns for total scores, and individual ff scores, which is
gonna be arbitary.
Prob best if this is handled downstream.
btw, why is it getting no entries? Is there walls/zones in your input?
Are the pruning harsh