List your company

Persona Alignment on Consistent-LLMs chatting domain

0.995Persona Consistency (PC)3-Step-PG
0.457320.596910.73650.87609Mar 6, 2026

Evaluation results

MethodLinks
3-Step-PGBackbone=Qwen, Evaluation Methodology=Step-level2026.030.995
3-Step-PGBackbone=Qwen, Evaluation Methodology=Rollout2026.030.985
5-Step-PGBackbone=Qwen, Evaluation Methodology=Step-level2026.030.964
5-Step-PGBackbone=Qwen, Evaluation Methodology=Rollout2026.030.958
2-Step-PGBackbone=Llama, Evaluation Methodology=Step-level2026.030.939
2-Step-PGBackbone=Llama, Evaluation Methodology=Rollout2026.030.929
4-Step-PGBackbone=Qwen, Evaluation Methodology=Step-level2026.030.925
GreedyPGBackbone=Qwen, Evaluation Methodology=Step-level2026.030.881
PGBackbone=Qwen, Evaluation Methodology=Rollout2026.030.867
GreedyPGBackbone=Llama, Evaluation Methodology=Step-level2026.030.85
PPOBackbone=Qwen, Evaluation Methodology=Step-level2026.030.845
3-Step-PGBackbone=Llama, Evaluation Methodology=Step-level2026.030.835
4-Step-PGBackbone=Qwen, Evaluation Methodology=Rollout2026.030.831
5-Step-PGBackbone=Llama, Evaluation Methodology=Step-level2026.030.824
3-Step-PGBackbone=Llama, Evaluation Methodology=Rollout2026.030.821
PPOBackbone=Qwen, Evaluation Methodology=Rollout2026.030.818
2-Step-PGBackbone=Qwen, Evaluation Methodology=Step-level2026.030.813
PGBackbone=Qwen, Evaluation Methodology=Step-level2026.030.813
5-Step-PGBackbone=Llama, Evaluation Methodology=Rollout2026.030.802
GreedyPGBackbone=Llama, Evaluation Methodology=Rollout2026.030.799
2-Step-PGBackbone=Qwen, Evaluation Methodology=Rollout2026.030.782
BaseBackbone=Qwen, Evaluation Methodology=Step-level2026.030.775
PGBackbone=Llama, Evaluation Methodology=Step-level2026.030.771
GreedyPGBackbone=Qwen, Evaluation Methodology=Rollout2026.030.762
BaseBackbone=Qwen, Evaluation Methodology=Rollout2026.030.76
PGBackbone=Llama, Evaluation Methodology=Rollout2026.030.727
4-Step-PGBackbone=Llama, Evaluation Methodology=Step-level2026.030.715
4-Step-PGBackbone=Llama, Evaluation Methodology=Rollout2026.030.692
PPOBackbone=Llama, Evaluation Methodology=Step-level2026.030.558
PPOBackbone=Llama, Evaluation Methodology=Rollout2026.030.556
BaseBackbone=Llama, Evaluation Methodology=Rollout2026.030.478
BaseBackbone=Llama, Evaluation Methodology=Step-level2026.030.478