List your company

Mathematical Reasoning on GSM8K OOD (test)

94.98AccuracyMultiRole-R1
76.62481.389586.15590.9205Jul 27, 2025

Evaluation results

MethodLinks
MultiRole-R1Backbone=Qwen3-8B, Method Strategy=SelfConsis SFT+GRPO(RS)2025.0794.9886.93
Role-PlayBackbone=Qwen3-8B2025.0793.5881.93
MultiRole-R1Backbone=R1-Distill-Qwen-14B, Method Strategy=SelfConsis SFT+GRPO(RS)2025.0793.4387.24
SelfConsis SFT+GRPOBackbone=Qwen3-8B2025.0791.9785.98
SelfConsis SFT+GRPOBackbone=R1-Distill-Qwen-14B2025.0791.8786.36
SelfConsis SFT+DPOBackbone=Qwen3-8B2025.0791.8465.32
SelfConsis SFTBackbone=Qwen3-8B2025.0791.6270.23
SelfConsis SFTBackbone=R1-Distill-Qwen-14B2025.0791.6191.62
Role-PlayBackbone=R1-Distill-Qwen-14B2025.0791.576.6
SelfConsis SFT+DPOBackbone=R1-Distill-Qwen-14B2025.0790.9268.69
More thinkBackbone=Qwen3-8B2025.0790.0275.07
MultiRole-R1Backbone=R1-Distill-Llama-8B, Method Strategy=SelfConsis SFT+GRPO(RS)2025.0789.7988.45
More thinkBackbone=R1-Distill-Qwen-14B2025.0788.7680.94
SelfConsis SFT+GRPOBackbone=R1-Distill-Llama-8B2025.0787.4985.31
MultiRole-R1Backbone=R1-Distill-Qwen-7B, Method Strategy=SelfConsis SFT+GRPO(RS)2025.0787.3682.46
Self-RefineBackbone=R1-Distill-Qwen-7B2025.0787.0180.37
SelfConsis SFT+DPOBackbone=R1-Distill-Llama-8B2025.0786.3481.81
SelfConsis SFTBackbone=R1-Distill-Llama-8B2025.0786.0281.53
Role-PlayBackbone=R1-Distill-Qwen-7B2025.0785.6672.87
SelfConsis SFT+GRPOBackbone=R1-Distill-Qwen-7B2025.0785.5882.16
Zero-shot CoTBackbone=R1-Distill-Qwen-14B2025.0785.5870.68
Zero-shot CoTBackbone=Qwen3-8B2025.0785.4170.49
Self-RefineBackbone=Qwen3-8B2025.0784.9182.31
Self-RefineBackbone=R1-Distill-Qwen-14B2025.0784.7380.24
More thinkBackbone=R1-Distill-Llama-8B2025.0783.384.12
More thinkBackbone=R1-Distill-Qwen-7B2025.0782.0581.79
SelfConsis SFT+DPOBackbone=R1-Distill-Qwen-7B2025.0781.5167.56
Self-RefineBackbone=R1-Distill-Llama-8B2025.0780.9581.24
SelfConsis SFTBackbone=R1-Distill-Qwen-7B2025.0780.6274.87
Zero-shot CoTBackbone=R1-Distill-Qwen-7B2025.0780.4868.08
Zero-shot CoTBackbone=R1-Distill-Llama-8B2025.0778.8776.52
Role-PlayBackbone=R1-Distill-Llama-8B2025.0777.3375.02