ControlMed: Adding Reasoning Control to Medical Language Model

The introduction of ControlMed marks a significant advancement in the application of reasoning Large Language Models (LLMs) within the medical field. By enabling users to control the length of reasoning processes, ControlMed addresses the critical need for efficiency in clinical decision-making. Traditional LLMs often generate unnecessarily lengthy outputs, leading to delays and increased computational demands. ControlMed's three-stage training process, which includes pre-training on a synthetic medical dataset, fine-tuning with multi-length reasoning data, and reinforcement learning, ensures that it achieves comparable or superior performance to existing models. This capability not only enhances the accuracy of medical reasoning but also allows healthcare professionals to balance the depth of analysis with the urgency of clinical situations, ultimately improving patient care.