# Step-by-step optimization strategy

R1-VL-7B is an inference model based on Qwen2-VL-7B-Instruct, trained using the Stepwise Grouped Relative Policy Optimization (StepGRPO) method, focusing on the image-text to text task.