Inducing Elasticity in Foundation Models: Post-Training Techniques for Adaptable Inference