SMALL REASONING MODELS WITH TWO-STAGE RULE-BASED RL

Summary from the TLDR AI newsletter

By breaking the reasoning training process into two steps, this work shows that small models are still able to generalize across tasks and even modalities.