List your company

Multi-Turn Consistency on Role-playing dataset

2.65KE ScorePPR-GDE
1.78682.01092.2352.4591May 18, 2026

Evaluation results

MethodLinks
PPR-GDEBackbone=Qwen2.5-3B-Instruct2026.052.652.943.032.87
GRPOBackbone=Qwen2.5-3B-Instruct2026.052.613.052.892.85
PPR-GDEBackbone=Qwen2.5-1.5B-Instruct2026.052.573.012.992.86
PPOBackbone=Qwen2.5-3B-Instruct2026.052.483.12.932.84
GRPOBackbone=Qwen3-0.6B2026.052.452.862.642.65
PPOBackbone=Qwen3-0.6B2026.052.422.892.682.66
PPR-GDEBackbone=Qwen3-0.6B2026.052.393.063.042.83
PPOBackbone=Qwen2.5-1.5B-Instruct2026.052.372.932.682.66
GRPOBackbone=Qwen2.5-1.5B-Instruct2026.052.332.932.852.7
w/o RLBackbone=Qwen3-0.6B2026.052.232.682.612.51
w/o RLBackbone=Qwen2.5-3B-Instruct2026.052.172.832.952.65
w/o RLBackbone=Qwen2.5-1.5B-Instruct2026.051.822.72.792.44