List your company

Persona Alignment on Consistent-LLMs education domain (test)

99Persona ConsistencyPG
40.2455.49570.7586.005Mar 6, 2026

Evaluation results

MethodLinks
PGLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.0399
PGLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.0397.6
PGLLM Backbone=Llama, Evaluation Methodology=Step-level2026.0397.4
4-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.0397.1
3-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.0396
4-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.0395.4
3-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.0394.8
5-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.0393.7
2-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.0393.6
2-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.0393
GreedyPGLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.0393
PGLLM Backbone=Llama, Evaluation Methodology=Rollout2026.0391.3
5-Step-PGLLM Backbone=Llama, Evaluation Methodology=Step-level2026.0389.6
PPOLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.0389.5
BaseLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.0388.6
5-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.0388.1
PPOLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.0387.5
4-Step-PGLLM Backbone=Llama, Evaluation Methodology=Step-level2026.0385.8
4-Step-PGLLM Backbone=Llama, Evaluation Methodology=Rollout2026.0383.7
GreedyPGLLM Backbone=Llama, Evaluation Methodology=Step-level2026.0383.3
BaseLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.0379.9
5-Step-PGLLM Backbone=Llama, Evaluation Methodology=Rollout2026.0377.5
2-Step-PGLLM Backbone=Llama, Evaluation Methodology=Step-level2026.0375.9
GreedyPGLLM Backbone=Llama, Evaluation Methodology=Rollout2026.0373.9
2-Step-PGLLM Backbone=Llama, Evaluation Methodology=Rollout2026.0372.1
3-Step-PGLLM Backbone=Llama, Evaluation Methodology=Step-level2026.0364.5
PPOLLM Backbone=Llama, Evaluation Methodology=Step-level2026.0362.4
GreedyPGLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.0362.3
PPOLLM Backbone=Llama, Evaluation Methodology=Rollout2026.0361.3
3-Step-PGLLM Backbone=Llama, Evaluation Methodology=Rollout2026.0357.3
BaseLLM Backbone=Llama, Evaluation Methodology=Rollout2026.0342.5
BaseLLM Backbone=Llama, Evaluation Methodology=Step-level2026.0342.5