List your company

Multi-hop Question Answering on 2WikiMultihopQA OOD (Exact Match)

39.2Exact Match AccuracyRECON
9.97617.56325.1532.737Oct 12, 2025

Evaluation results

MethodLinks
RECONBackbone=Qwen2.5-7B-Base, Training Protocol=PPO, RECON Stage=Stage 1 + Stage 22025.1039.2
Search-R1Backbone=Qwen2.5-7B-Base, Training Protocol=PPO2025.1038.2
RECONBackbone=Qwen2.5-3B-Base, Training Protocol=PPO, RECON Stage=Stage 1 + Stage 22025.1031.9
Rejection Sampling2025.1029.6
R1-instruct2025.1029.2
RECONBackbone=Qwen2.5-7B-Base, Training Protocol=PPO, RECON Stage=Stage 2 Distillation-only2025.1028.6
R1-base2025.1027.3
Search-R1Backbone=Qwen2.5-3B-Base, Training Protocol=PPO2025.1027.3
SFT2025.1025.9
Direct Inference2025.1025
RAG2025.1023.5
RECONBackbone=Qwen2.5-3B-Base, Training Protocol=PPO, RECON Stage=Stage 2 Distillation-only2025.1023
Search-o12025.1017.6
IRCoT2025.1014.9
CoT2025.1011.1