Developed a dynamic backend selection feature for the MTIA framework within the pytorch/pytorch repository, focusing on the GPU vertical pass and pattern matcher. This work introduced a runtime backend parameter, allowing users to select different GPU backends while maintaining backward compatibility by defaulting to the inductor backend. The implementation, carried out in Python and leveraging GPU programming and backend development expertise, enables greater flexibility for experimentation and performance tuning in the MTIA GPU path. By establishing this foundation, the contribution supports future enhancements and aligns with the evolving MTIA roadmap, while ensuring stability for existing workflows and users.