Hi, yes, this is being actively used.
We do not need this particular configuration (it has the serious flaw that it is not thread-safe), but we need something that: - allows to specify weights on a per-sentence basis - allows to turn off features completely on a per sentence basis (e.g., unused domain-specific language models) - caches translation options correctly (same weight/feature setting -> re-use cached options). The alternative weight setting is currently doing that, so whatever you want to replace it with, should have at least these features. Being thread-safe would be desirable as well. -phi On Wed, May 13, 2015 at 3:12 AM, Leusch, Gregor <[email protected]> wrote: > So, to clarify: > * If tracking “weights per translation task” is per xmlrpc request, and > different concurrent requests can have different weights, then yes, please, > go ahead! > * We are currently not using AWS because it only works in single-threaded > mode. > > — Gregor > > > From: <Leusch>, "Leusch, Gregor" <[email protected]> > Date: Wednesday 13 May 2015 10:01 > To: "[email protected]" <[email protected]>, " > [email protected]" <[email protected]> > Subject: Re: [Moses-support] Is anyone still using the alternate weights > feature in Moses? > > Hi Uli, > > We would love to use it (or a similar mechanism to set sentence-specific > weights), but since it cannot be used in multithreaded moses (and thus in > mosesserver), we currently have to start individual mosesserver instances > for each individual weigths setting, and switch on dispatcher level. So a > mechanism like AWS that also works in mosesserver would be highly > appreciated. > > Best, > > Gregor > > > From: Ulrich Germann <[email protected]> > Reply-To: "[email protected]" <[email protected]> > Date: Tuesday 12 May 2015 22:04 > To: "[email protected]" <[email protected]> > Subject: [Moses-support] Is anyone still using the alternate weights > feature in Moses? > > Hi all, > > I'd like to retire the alternate weight settings mechanism in Moses in > exchange for keeping track of weights per translation task. Is anyone still > using that mechanism, or is that a thing of the past? > > - Uli > > -- > Ulrich Germann > Senior Researcher > School of Informatics > University of Edinburgh > > > > > > > _______________________________________________ > 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
