List your company
ResearchFormerly WizWand

Knowledge-Intensive Reasoning on MuSiQue (F1 score)

34.8F1 ScoreLlama3.1-8B + ARPO
2.35210.77619.227.624Dec 11, 2025Jan 8, 2026Feb 6, 2026Mar 7, 2026Apr 4, 2026May 3, 2026Jun 1, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Llama3.1-8B + ARPOBase Model=Llama3.1-8B, Algorithm=ARPO2025.1234.8——
EAPOBackbone Model=Qwen2.5-7B-Instruct2026.0633.1222.46
EAPOBackbone=Llama3.1-8B-Instruct2026.0632.8162.58
GRPOBackbone Model=Qwen2.5-7B-Instruct2026.0632.117.53.1
Qwen2.5-7B + ARPOBase Model=Qwen2.5-7B, Algorithm=ARPO2025.1231.1——
Llama3.1-8B + GRPOBase Model=Llama3.1-8B, Algorithm=GRPO2025.1231——
Soft-NBCEMem=O(L^2/n), zero-shot=false, distillation=true2026.0531——
Reinforce++Backbone Model=Qwen2.5-7B-Instruct2026.0630.715.42.26
Qwen2.5-7B + GRPOBase Model=Qwen2.5-7B, Algorithm=GRPO2025.1230.6——
GRPOBackbone=Llama3.1-8B-Instruct2026.0630.616.73.07
Reinforce++Backbone=Llama3.1-8B-Instruct2026.0630.215.92.37
Qwen2.5-3B + DAPOBase Model=Qwen2.5-3B, Algorithm=DAPO2025.1230——
Llama3.1-8B + Reinforce ++Base Model=Llama3.1-8B, Algorithm=Reinforce ++2025.1229.9——
Llama3.1-8B + DAPOBase Model=Llama3.1-8B, Algorithm=DAPO2025.1229.2——
Qwen2.5-3B + ARPOBase Model=Qwen2.5-3B, Algorithm=ARPO2025.1228.7——
Qwen2.5-7B + DAPOBase Model=Qwen2.5-7B, Algorithm=DAPO2025.1228.6——
Soft-NBCEMem=O(L^2/n), zero-shot=true, distillation=false2026.0528.3——
Qwen2.5-3B + Reinforce ++Base Model=Qwen2.5-3B, Algorithm=Reinforce ++2025.1227.9——
Vanilla NBCEMem=O(L^2/n)2026.0527.5——
Qwen2.5-7B + Reinforce ++Base Model=Qwen2.5-7B, Algorithm=Reinforce ++2025.1225.2——
AEPOBackbone Model=Qwen2.5-7B-Instruct2026.0625.113.54.71
Qwen2.5-3B + GRPOBase Model=Qwen2.5-3B, Algorithm=GRPO2025.1224.7——
EAPOBackbone Model=Qwen2.5-3B-Instruct2026.0623.417.52.26
ARPOBackbone Model=Qwen2.5-7B-Instruct2026.0622.5122.33
PCWMem=O(L^2/n)2026.0521.2——
Truncated (8K)Mem=O(L^2)2026.0520.9——
ARPOBackbone Model=Qwen2.5-3B-Instruct2026.0619.411.52.62
TIRBackbone=Llama3.1-8B-Instruct2026.0617.89.72.67
GRPOBackbone Model=Qwen2.5-3B-Instruct2026.0617.79.82.48
ToolStarBackbone Model=Qwen2.5-7B-Instruct2026.0616.8114.07
Reinforce++Backbone Model=Qwen2.5-3B-Instruct2026.06167.82.51
ToolStarBackbone Model=Qwen2.5-3B-Instruct2026.0615.78.52.52
Llama3.1-8B + TIR PromptingBase Model=Llama3.1-8B, Algorithm=TIR Prompting2025.1215.5——
Llama3.1-8BBase Model=Llama3.1-8B, Algorithm=Direct Reasoning2025.1210.4——
BaseBackbone=Llama3.1-8B-Instruct2026.069.87.4—
Qwen2.5-7B + TIR PromptingBase Model=Qwen2.5-7B, Algorithm=TIR Prompting2025.129.5——
TIRBackbone Model=Qwen2.5-7B-Instruct2026.067.72.52.56
Qwen2.5-7BBase Model=Qwen2.5-7B, Algorithm=Direct Reasoning2025.126.6——
BaseBackbone Model=Qwen2.5-7B-Instruct2026.066.52.5—
Qwen2.5-3B + TIR PromptingBase Model=Qwen2.5-3B, Algorithm=TIR Prompting2025.126.1——
TIRBackbone Model=Qwen2.5-3B-Instruct2026.066.13.21.59
Qwen2.5-3BBase Model=Qwen2.5-3B, Algorithm=Direct Reasoning2025.123.6——
BaseBackbone Model=Qwen2.5-3B-Instruct2026.063.60.5—