def main()

in evaluate_model.py [0:0]

• 28 lines of code
• 1 McCabe index (conditional complexity)

def main() -> None:
    """Main function to run model evaluation."""
    args = parse_args()
    
    evaluation_tracker = EvaluationTracker(
        output_dir="./results",
        save_details=True,
        push_to_hub=False,
    )

    pipeline_params = PipelineParameters(
        launcher_type=ParallelismManager.ACCELERATE,
        max_samples=1000,
        custom_tasks_directory="math_verify.tasks",
        env_config=EnvConfig(cache_dir="tmp/"),
        override_batch_size=args.override_bs,
    )

    model_config = TransformersModelConfig(
        pretrained=args.model,
        dtype="bfloat16",
        use_chat_template=args.use_chat_template,
    )

    pipeline = Pipeline(
        tasks=f"lighteval|{args.task}|4|1",
        pipeline_parameters=pipeline_params,
        evaluation_tracker=evaluation_tracker,
        model_config=model_config,
    )

    pipeline.evaluate()
    pipeline.show_results()
    pipeline.save_and_push_results()