What is it?

The ml-interface repository is intended to eventually be a standalone library that games can use to interface with various text-generation endpoints in one "easy" to use library.

Currently, it's very hardcoded for my current project: Something Else (a mod for OpenMW), as it's still very early stages.

Usage

Note: Detailed instructions for installing Something Else / OpenMW can be found in the doc directory, or by clicking here.

Generically, the repository can be used as follows...

Warning: The following may be out of date, things are still in very early development. Check the scripts for the most accurate information.

Installation

  • Run ml-interface.sh init to create the venv and install dependencies.
    • Requires python3 to be on the path and venv to be installed.
    • This creates a virtualenv called 'openmw_ml' in ~/.venv/openmw_ml with a few dependencies installed.
    • To delete the venv, run ml-interface.sh clean.

Running

  • Create an input json for the model you wish to run.
    • The schema/layout of the json depends on the model being run, so there is no generic example available.
    • For model-specific examples, look inside the examples folder of the model you wish to run.
  • Set environment variable VENV_NAME to the name of the conda environment you created above.
    • TODO: Not yet impletmented, currently the script is hardcoded to run scripts in the openmw_ml environment.
  • Execute ml-interface.sh.
    • Example: ml-interface.sh openai_chat /path/to/input.json