Developed the LiteRAWFormer model architecture for the OmniGen2 repository, focusing on raw image processing with a lightweight, transformer-based approach. The work involved implementing core transformer components such as LayerNorm, Attention, and FeedForward blocks, as well as utilities for efficient tensor manipulation and channel shuffling. Using Python and PyTorch, the architecture established a scalable foundation for advanced image processing features, including future support for raw image super-resolution. The contribution emphasized modular model design and efficient computation, enabling enhanced user-facing capabilities in image processing workflows. YAML was also used for configuration, supporting reproducibility and maintainability within the project’s codebase.