Concept

Evaluation harness

The software that runs a model on a benchmark: it builds the prompts, calls the model, executes or scores the outputs and aggregates the results.