Create Evaluation

Launch an Evaluation Round

An Evaluation is a background process that evaluates an Application using a set of prompt and reference answers. It checks the quality of the generated predictions using one or more metrics. The following metrics are supported:

• equivalence
• groundedness.

You can provide the evaluation data for an Evaluation in two ways:

1. CSV File
   Upload an evalset_file as a CSV with the following columns:

   • prompt: The question.
   • reference: The correct (gold) answer.
     If you want to evaluate with your own predictions you need to provide the extra columns:
   • response: Include this column to measure equivalence.
   • knowledge: Include this column to evaluate groundedness.

2. Dataset Name
   Use a evalset_name that refers to an evaluation_set Dataset created through the Dataset API.