Search for a command to run...
DSDR: Dual-Scale Diversity Regularization for Exploration in LLM Reasoning