Framing Sensitivity Evaluation on FRAGILE Narrative vividness
6.4Flip%VALIGN
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VALIGNLLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 6.4 | 6.3 | 4.7 | 88.9 | |
| BaseLLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 8.6 | 8.2 | 14.2 | 77.3 | |
| VALIGNLLM Backbone=Qwen2.5-7B-Instruct2026.05 | 11.3 | 11 | 11.5 | 77.2 | |
| VALIGNLLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 13.4 | 10.7 | 5.4 | 81.2 | |
| BaseLLM Backbone=Qwen2.5-7B-Instruct2026.05 | 14 | 13.2 | 15.3 | 70.6 | |
| BaseLLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 17.4 | 3.8 | 6 | 76.6 | |
| CAALLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 24.8 | 22.9 | 36.2 | 39 | |
| CAALLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 28.9 | 28.6 | 15 | 56.1 | |
| CAALLM Backbone=Qwen2.5-7B-Instruct2026.05 | 29.2 | 28.4 | 18.9 | 51.9 | |
| Instruc.LLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 46.2 | 25 | 32 | 21.9 | |
| Instruc.LLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 48.7 | 19.6 | 12.1 | 39.2 | |
| Instruc.LLM Backbone=Qwen2.5-7B-Instruct2026.05 | 53.5 | 18.3 | 16.6 | 29.9 | |
| CoTLLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 53.6 | 29.3 | 27.7 | 18.7 | |
| CoTLLM Backbone=Qwen2.5-7B-Instruct2026.05 | 56.2 | 20.3 | 14.6 | 29.2 | |
| CoTLLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 58.8 | 21.5 | 10.2 | 31.1 |