We propose X-Fusion, a framework that extends pretrained Large Language Models (LLMs) for multimodal tasks while preserving their language capabilities. X-Fusion employs a dual-tower design with modality-specific weights, keeping the LLM's parameters frozen while integrating vision-specific information for both understanding and generation. Our experiments demonstrate that X-Fusion consistently outperforms alternative architectures on both image-to-text and text-to-image tasks. We find that incorporating understanding-focused data improves generation quality, reducing image data noise enhances overall performance, and feature alignment accelerates convergence for smaller models but has minimal impact on larger ones. Our findings provide valuable insights into building efficient unified multimodal models.

Abstract. This work aims to accelerate the convergence of proximal gradient methods used to solve regularized linear inverse problems. This is achieved by designing a polynomial-based preconditioner that targets the eigenvalue spectrum of the normal operator derived from the linear operator. The preconditioner does not assume any explicit structure on the linear function and thus can be deployed in diverse applications of interest. The efficacy of the preconditioner is validated on three different Magnetic Resonance Imaging applications, where it is seen to achieve faster iterative convergence (around [Formula: see text] faster, depending on the application of interest) while achieving similar reconstruction quality.