Developed backend support for MUSA within the upstash/FlagEmbedding repository, focusing on expanding device compatibility for enterprise machine learning workflows. The work involved updating the device selection logic in both AbsEmbedder and AbsReranker to recognize and utilize MUSA when available, thereby optimizing embedding operations on supported hardware. Leveraging Python and PyTorch, the implementation provided clients with greater flexibility in deployment environments, particularly those relying on MUSA infrastructure. This targeted feature update addressed backend extensibility without introducing bug fixes, reflecting a focused engineering approach to enhancing runtime flexibility and broadening the adoption potential of FlagEmbedding in production settings.