Persona Alignment on Consistent-LLMs education domain (test)
99Persona ConsistencyPG
Evaluation Results
| Method | Links | |
|---|---|---|
| PGLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.03 | 99 | |
| PGLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.03 | 97.6 | |
| PGLLM Backbone=Llama, Evaluation Methodology=Step-level2026.03 | 97.4 | |
| 4-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.03 | 97.1 | |
| 3-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.03 | 96 | |
| 4-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.03 | 95.4 | |
| 3-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.03 | 94.8 | |
| 5-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.03 | 93.7 | |
| 2-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.03 | 93.6 | |
| 2-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.03 | 93 | |
| GreedyPGLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.03 | 93 | |
| PGLLM Backbone=Llama, Evaluation Methodology=Rollout2026.03 | 91.3 | |
| 5-Step-PGLLM Backbone=Llama, Evaluation Methodology=Step-level2026.03 | 89.6 | |
| PPOLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.03 | 89.5 | |
| BaseLLM Backbone=Qwen, Evaluation Methodology=Step-level2026.03 | 88.6 | |
| 5-Step-PGLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.03 | 88.1 | |
| PPOLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.03 | 87.5 | |
| 4-Step-PGLLM Backbone=Llama, Evaluation Methodology=Step-level2026.03 | 85.8 | |
| 4-Step-PGLLM Backbone=Llama, Evaluation Methodology=Rollout2026.03 | 83.7 | |
| GreedyPGLLM Backbone=Llama, Evaluation Methodology=Step-level2026.03 | 83.3 | |
| BaseLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.03 | 79.9 | |
| 5-Step-PGLLM Backbone=Llama, Evaluation Methodology=Rollout2026.03 | 77.5 | |
| 2-Step-PGLLM Backbone=Llama, Evaluation Methodology=Step-level2026.03 | 75.9 | |
| GreedyPGLLM Backbone=Llama, Evaluation Methodology=Rollout2026.03 | 73.9 | |
| 2-Step-PGLLM Backbone=Llama, Evaluation Methodology=Rollout2026.03 | 72.1 | |
| 3-Step-PGLLM Backbone=Llama, Evaluation Methodology=Step-level2026.03 | 64.5 | |
| PPOLLM Backbone=Llama, Evaluation Methodology=Step-level2026.03 | 62.4 | |
| GreedyPGLLM Backbone=Qwen, Evaluation Methodology=Rollout2026.03 | 62.3 | |
| PPOLLM Backbone=Llama, Evaluation Methodology=Rollout2026.03 | 61.3 | |
| 3-Step-PGLLM Backbone=Llama, Evaluation Methodology=Rollout2026.03 | 57.3 | |
| BaseLLM Backbone=Llama, Evaluation Methodology=Rollout2026.03 | 42.5 | |
| BaseLLM Backbone=Llama, Evaluation Methodology=Step-level2026.03 | 42.5 |