To run the decoder

  $ ./decoder.py --flagfile FLAGFILE

  Run

  $ ./decoder.py --help

  to see the help message for flags and default values.

  You can override flagfile options by using command line flags. A flagfile can
  import flags from another flagfile by using the --flagfile flag.

  See test/smoke.flag for an example flagfile.

The output

  You will find these in the directory where you run the decoder:

  - The translation output (the file "output" by default)
  - The kbest file, if --output_kbest=true (the file "kbest" by default)
  - A directory containing the detailed decoding results. By default the name of
    the directory is "run-dir", which is changeable using --run_dir.

To do parallel decoding

  If you have a cluster of machines running on NFS, you can do parallel decoding
  with the command

  $ ./decoder.py --flagfile FLAGFILE --parallel=true --nodes=f02,f03

  This will start multiple slave instances of the decoder on the hosts specified
  by --nodes.

Tips

  - With the --do option, the decoder decodes only one sentence ("--do=4") or
    a range of sentences in the input ("--do=8-10"). Currently this doesn't
    work in parallel mode.

  - Under running directory, file "run.flag" saves the options you have used to
    make this run.

  - Under running directory you will also find the hypergraph files and the
    1-best parse trees.