Parallel execution on a single local machine has been here all along, since the time Martin Majlis implemented it (the parameter --local is even documented in Treex::Core::Run)... I only never knew.
The only problem was that killing the Head process did not kill all the processing Nodes. This tiny change fixes the problem, so we may now run Treex in parallel on a single machine (using e.g. treex -p -j 4 --local).
Note that no memory checking is done, so the user must make sure that the whole thing will fit into memory (e.g. parsers/translation models need to load n times).
Parallel execution on a single local machine has been here all along, since the time Martin Majlis implemented it (the parameter
--local
is even documented inTreex::Core::Run
)... I only never knew.The only problem was that killing the Head process did not kill all the processing Nodes. This tiny change fixes the problem, so we may now run Treex in parallel on a single machine (using e.g.
treex -p -j 4 --local
).Note that no memory checking is done, so the user must make sure that the whole thing will fit into memory (e.g. parsers/translation models need to load n times).