Common Parameters

Whether using the command line (CLI) or Python interfaces, the solver parameters are for the most part the same. These are described below:

• Problem definition file

  • The full path to the problem definition (PD) JSON file.

  • CLI: --pf, Python: problem_file

• External file data path

  • The full path to the directory that contains all external files (e.g., sets, matrices, atoms) used by the PD. Just the file names (not full path) are referenced in the PD so this must be set here.

  • CLI: --dp, Python: data_path

• NVIDIA Device

  • The index of the NVIDIA device, typically 0 for a PC with a single GPU.

  • CLI: --device, Python: device

• Run mode

  • Whether to run in debug or release mode. All problems should be tested in debug mode as debug mode examines all variables, logic errors, etc. but it is MUCH slower as it does not run on the GPU. Thus it is best to test small samples in debug mode. In release mode, it is run on the GPU and will not offer debugging information and thus may just crash.

  • CLI: --run debug|release, Python: debug=true|false

• Thread count

  • The number of threads to utilize. In debug mode, these threads are run in serial - i.e., they are not true threads. In release mode of course they are. This should be set based on the capabilities of the GPU and the memory requirements of the problem.

  • CLI: --tc, Python: thread_count

• Stop seconds

  • The maximum number of seconds to run the solver.

  • CLI: --st, Python: stop_seconds

• Stop score

  • Stop the solver when it reaches this score.

  • CLI: --ss, Python: stop_score

• Stop if no change

  • Stop the solver when the score does not improve after this many seconds.

  • CLI: --sn, Python: stop_no_change