List your company

Question Answering on 2WikiMultiHopQA, HotpotQA, and MuSiQue Aggregate (test)

24.5Average AccuracyFedGRPO
6.311.02515.7520.475Feb 12, 2026

Evaluation results

MethodLinks
FedGRPOModel=Qwen2.5-3B2026.0224.5——
Central-GRPOModel=Qwen2.5-3B2026.0223.9——
DPSDA-FL+SFTModel=Qwen2.5-3B2026.0222.4——
Central-GRPOModel=Qwen2.5-1.5B2026.0220.6——
FedGRPOModel=Qwen2.5-1.5B2026.0220.6——
DPSDA-FL+GRPOModel=Qwen2.5-3B2026.0219.6——
Fedpetuning+SFTModel=Qwen2.5-3B2026.0216.5——
DPSDA-FL+SFTModel=Qwen2.5-1.5B2026.0216.3——
DPSDA-FL+GRPOModel=Qwen2.5-1.5B2026.0215——
Fedpetuning+GRPOModel=Qwen2.5-3B2026.0213.2——
Fedpetuning+SFTModel=Qwen2.5-1.5B2026.028.4——
Zero-shotModel=Qwen2.5-3B2026.028——
Fedpetuning+GRPOModel=Qwen2.5-1.5B2026.027.2——
Zero-shotModel=Qwen2.5-1.5B2026.027——
GRPOFramework=LLM + RL2025.06—18.5216.917
Llama3.1-8BFramework=LLM2025.06—19.3569.087
Llama3.1-8BFramework=LLM + MAS2025.06—18.75112.561
MAPPOFramework=LLM + MAS + MARL2025.06—39.68924.499
MHGPO-FoFFramework=LLM + MAS + MARL2025.06—42.15325.248
MHGPO-FoF(os)Framework=LLM + MAS + MARL2025.06—43.33126.109
MHGPO-ISFramework=LLM + MAS + MARL2025.06—38.98923.369
MHGPO-RRFramework=LLM + MAS + MARL2025.06—42.54424.99
PPOFramework=LLM + RL2025.06—17.0526.158
Qwen2.5-72BFramework=LLM2025.06—30.68419.522
Qwen2.5-72BFramework=LLM + MAS2025.06—30.46817.379
R1-Searcher(GRPO)Framework=LLM + RAG (+RL)2025.06—38.32823.483
R1-Searcher(PPO)Framework=LLM + RAG (+RL)2025.06—40.23622.297
Search-o1Framework=LLM + RAG (+RL)2025.06—22.21516.23