Search-E1: Self-Distillation Drives Self-Evolution in Search-Augmented Reasoning
Researchers have introduced Search-E1, a novel self-evolution method for search-augmented reasoning agents that bypasses complex external supervision. This approach utilizes vanilla GRPO combined with offline self-distillation (OFSD) to enable agents to improve independently. The method achieved a $0.440$ average EM score on seven QA benchmarks using the Qwen2.5-3B model, outperforming existing open-source baselines. AI
IMPACT Simplifies training for search-augmented reasoning agents, potentially making them more accessible and efficient.