diff options
| author | Chris Dyer <cdyer@cs.cmu.edu> | 2011-07-10 23:00:21 -0400 |
|---|---|---|
| committer | Chris Dyer <cdyer@cs.cmu.edu> | 2011-07-10 23:00:21 -0400 |
| commit | 95deb840699f9b6f8fe499b374bd726bce97365c (patch) | |
| tree | 64d0299b2da95122d18654f254ae41c14bcb456d /pro-train/README.shared-mem | |
| parent | f68aaa588a2e1eb7cef5a57d57dd7e86fd4b0c9a (diff) | |
starting implementation of Hopkins&May (2011) optimizer
Diffstat (limited to 'pro-train/README.shared-mem')
| -rw-r--r-- | pro-train/README.shared-mem | 9 |
1 files changed, 9 insertions, 0 deletions
diff --git a/pro-train/README.shared-mem b/pro-train/README.shared-mem new file mode 100644 index 00000000..7728efc0 --- /dev/null +++ b/pro-train/README.shared-mem @@ -0,0 +1,9 @@ +If you want to run dist-vest.pl on a very large shared memory machine, do the +following: + + ./dist-vest.pl --use-make I --decode-nodes J --weights weights.init --source-file=dev.src --ref-files=dev.ref.* cdec.ini + +This will use I jobs for doing the line search and J jobs to run the decoder. Typically, since the +decoder must load grammars, language models, etc., J should be smaller than I, but this will depend +on the system you are running on and the complexity of the models used for decoding. + |