Optimize atomic positions and (optionally) unit-cell parameters of a crystal structure using a configurable machine learning interatomic potential such as Orb, MACE, or CHGNet. Upload a CIF file and receive the relaxed structure as a new CIF; set include_trajectory to also get the optimization path as an ASE .traj file. Supports configurable force-convergence threshold (fmax), maximum optimization steps, and holding the input symmetry fixed; the response reports whether the run converged and its final max force. Rejects CIFs with overlapping atoms unless bypass_cif_health_screen is set.