List your company
ResearchFormerly WizWand

Question Answering on 2WikiMultihopQA (EM only)

36.8EMTeacher-3B
-1.294168.5956718.485528.37533Aug 27, 2025Oct 10, 2025Nov 24, 2025Jan 7, 2026Feb 21, 2026Apr 6, 2026May 21, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Teacher-3BModel Family=Qwen 2.5, Teacher Model=Qwen 2.5 3B2025.0836.8
DGPOModel Family=Qwen 2.5, Teacher Model=Qwen 2.5 3B, Student Model=Qwen 2.5 0.5B2025.0830.3
KDModel Family=Qwen 2.5, Teacher Model=Qwen 2.5 3B, Student Model=Qwen 2.5 0.5B2025.0828.4
SFT → PPOModel Family=Qwen 2.5, Teacher Model=Qwen 2.5 3B, Student Model=Qwen 2.5 0.5B2025.0827.5
SeqKDModel Family=Qwen 2.5, Teacher Model=Qwen 2.5 3B, Student Model=Qwen 2.5 0.5B2025.0827.3
DistiLLMModel Family=Qwen 2.5, Teacher Model=Qwen 2.5 3B, Student Model=Qwen 2.5 0.5B2025.0827
TAIDModel Family=Qwen 2.5, Teacher Model=Qwen 2.5 3B, Student Model=Qwen 2.5 0.5B2025.0827
PPOModel Family=Qwen 2.5, Teacher Model=Qwen 2.5 3B, Student Model=Qwen 2.5 0.5B2025.0821.8
GKDModel Family=Qwen 2.5, Teacher Model=Qwen 2.5 3B, Student Model=Qwen 2.5 0.5B2025.0821.7
Student-0.5BModel Family=Qwen 2.5, Student Model=Qwen 2.5 0.5B2025.081.5
Search-E1-InstructRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Process-supervision RL2026.050.436
AutoRefine-BaseRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Outcome-reward RL2026.050.393
AutoRefine-InstructRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Outcome-reward RL2026.050.38
GiGPO-InstructRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Process-supervision RL2026.050.37
StepSearch-BaseRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Process-supervision RL2026.050.339
StepSearch-InstructRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Process-supervision RL2026.050.32
Search-R1-InstructRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Outcome-reward RL2026.050.31
R1-InstructRetrieval Strategy=None2026.050.275
Search-R1-BaseRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Outcome-reward RL2026.050.274
ReSearch-BaseRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Outcome-reward RL2026.050.272
ReSearch-InstructRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Outcome-reward RL2026.050.272
R1-BaseRetrieval Strategy=None2026.050.268
SFTRetrieval Strategy=None2026.050.248
Direct GenerationRetrieval Strategy=None2026.050.244
Naive RAGRetrieval Strategy=Single-Hop Retrieval2026.050.226
Search-o1Retrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Prompting2026.050.218
IRCoTRetrieval Strategy=Multi-Hop Retrieval, Supervision Paradigm=Prompting2026.050.171