Thyme: Think Beyond Images with Code-Executing MLLMs

This source introduces Thyme, a novel AI paradigm designed to enhance multimodal language models by integrating autonomous code generation and execution for image manipulation and complex calculations. Thyme enables models to dynamically process images through operations like cropping, rotation, and contrast enhancement, and to solve mathematical problems by converting them into executable code within a secure sandbox environment. The paper details Thyme's training methodology, which combines supervised fine-tuning and reinforcement learning, to achieve significant performance improvements across a wide range of perception, reasoning, and general AI tasks. The authors emphasize Thyme's high autonomy in deciding when and how to apply these operations, along with its efficient end-to-end training and consistent gains in benchmark evaluations. The research highlights the development of specialized datasets and training strategies to overcome challenges in code generation and improve the model's ability to reason with and beyond visual information.