Configure a solver YAML and run `driver_point.py` on a remote GPU node, then validate results with metrics.