ServiceNow report mapper

This crate creates a OpenAI fine-tuning file to enable the model to categorize ServiceNow incidents into assignment groups.

Usage

USAGE:
    snow_report_mapper [OPTIONS] <FILE_INCIDENTS> <FILE_ASSIGNMENT_GROUPS> <FILE_OUTPUT>

ARGS:
    <FILE_INCIDENTS>            Filepath to the SNOW incidents export
    <FILE_ASSIGNMENT_GROUPS>    Filepath to the SNOW export of the assignment groups
    <FILE_OUTPUT>               Filepath where the mapped training file should be stored to

OPTIONS:
    -h, --help       Print help information
    -s, --stats      Prints additional statistics
    -v, --verbose    Verbose output
    -V, --version    Print version information

To get this help, run:

$ snow_report_mapper --help

For more details, check the rust documentation of this crate.

Fine tune the model

Check the official docs on how to fine-tune the model.

Before the generated files can be used, they have to be mapped to the JSONL format. For that it's recommended to use the official utility tools:

$ openai tools fine_tunes.prepare_data -f <LOCAL_FILE>

It also shows suggestions if there are any and provides you with the proper command, to actual fine tune the model, in the end.

Troubleshooting

If you encounter an error message like:

The number of classes in file-7zfLf0xUhfTomlH7XPs5OGUL does not match the number of classes specified in the
hyperparameters

After triggering the fine tuning job, you can omit the parameters --compute_classification_metrics and --classification_n_classes (additionally to its numeric value). You will lose the Classification specific metrics but at least it works then.

Motivation

The main reasons why I've built this, is to learn Rust and to get a bit more familiar with OpenAI. As a "starter project" I've decided to train a model so that it's able to determine an appropriate assignment group for a ServiceNow incident, based on the incident title. In order to that, I need training data in the appropriate format.