Recent advancements in training large multimodal models have been driven by efforts to eliminate modeling constraints and unify architectures across domains. Despite these strides, many existing ...