Hi Tom.

        lmplz doesn't need libxmlrpc_xmltok.  Looks like a case of
over-aggressive dependencies, resulting in a binary that needs a library
it doesn't use.

        One could install xmlrpc-c (a third-party library used by Moses server)
in the same path.  Or I guess substitute lmplz from the kenlm repo which
won't have this dependency and which should compile fine on OS X.

Kenneth

On 01/12/2018 03:08 PM, Tom Hoar wrote:
> I'm testing the Release 4.0 (5th Oct, 2017) binaries for macOS
> (macOS.zip) on a Mac High Sierra (10.13) virtual machine from
> TechsViewer
> (https://techsviewer.com/install-macos-high-sierra-virtualbox-windows/).
> I'm using the homebrew version of GNU sort/split utilities. The host
> machine is an 8 virtual core i7 4 Ghz with 32GB RAM and 500GB SSD. The
> virtual guest is configured with 24 GB RAM and six cores.
> 
> I've found two problems and I hope someone can help.
> 
>  1. train-model.perl works all of the way through. The sort/split
>     utilities work fine. The problem is mgiza in step 2 took almost 3
>     hours to complete with a small 15,000 segment parallel corpus. Step
>     2 using the same corpus on the native Windows host takes less than
>     10 minutes. That step with the Ubuntu 16.04 binaries running on
>     Windows Subsystem for Linux (WSL) takes 9 1/2 minutes.
> 
>     Observing the Mac's Activity Monitor during step 2 shows that mgiza
>     was running 6 threads, but the total CPU load was only 115%. All the
>     other macOS binaries ran at a comparable speed to the other
>     platforms' binaries. for example, when the moses binary runs with 6
>     threads, the monitor shows it's consuming nearly 550%.
> 
>     Last year, we cross-compiled our own macOS binaries on Linux host
>     using circa Moses 3 source. We experienced the same slow performance
>     with mgiza binary on an old Mac notebook hardware. At the time, we
>     wrote delayed work because we thought slow performance was due to
>     the cross-compile settings and/or the slow hardware. After slow
>     results with the Moses4 mgiza binary, I re-cross-compiled the mgiza
>     binary on our system and it was also dead slow on the 10.13 guest
>     taking almost 3 hours.
> 
>     These tests were run on a very fast host machine with a virtual
>     guest. Using Hieu's binaries or mine seems to run against those
>     assumptions that hardware/cross-compiler are totally at fault. Has
>     anyone else experienced this kind of slow performance?
> 
>  2. The lmplz binary fails on this host with this message. Was this not
>     compiled with static links for a reason? Where can I find the
>     necessary libraries?
>     $ ./lmplz
>     dyld: Library not loaded:
>     
> /User/hieu/workspace/xmlrpc-c/xmlrpc-c-1.39.07/lib/libxmlrpc_xmltok.3.39.dylib
>       Referenced from: /opt/macOS/bin/./lmplz
>       Reason: image not found
>     Abort trap: 6
> 
> 
> 
> 
> _______________________________________________
> 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