Framing Sensitivity Evaluation on FRAGILE Temporal slice
13.3Flip RateVALIGN
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VALIGNLLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 13.3 | 10.2 | 5.8 | 80.9 | |
| VALIGNLLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 16.8 | 16.6 | 5.3 | 77.9 | |
| VALIGNLLM Backbone=Qwen2.5-7B-Instruct2026.05 | 19.5 | 19.1 | 10.5 | 69.9 | |
| BaseLLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 23 | 21.6 | 12.7 | 64.3 | |
| CAALLM Backbone=Qwen2.5-7B-Instruct2026.05 | 27.5 | 26.9 | 13.4 | 59.1 | |
| CAALLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 27.8 | 25.6 | 27.4 | 44.8 | |
| BaseLLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 30.2 | 10.5 | 3.9 | 65.9 | |
| CAALLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 36 | 35.7 | 7.2 | 56.8 | |
| BaseLLM Backbone=Qwen2.5-7B-Instruct2026.05 | 37.9 | 21.7 | 17.7 | 44.4 | |
| Instruc.LLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 52.7 | 29.4 | 24.2 | 23.1 | |
| Instruc.LLM Backbone=Qwen2.5-7B-Instruct2026.05 | 52.9 | 20.6 | 19.8 | 27.3 | |
| Instruc.LLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 55.1 | 26.6 | 12.4 | 32.6 | |
| CoTLLM Backbone=LLaMA-3.1-8B-Instruct2026.05 | 57.3 | 31.6 | 22 | 20.7 | |
| CoTLLM Backbone=Qwen2.5-7B-Instruct2026.05 | 57.9 | 23.8 | 16.6 | 25.5 | |
| CoTLLM Backbone=Mistral-7B-Instruct-v0.32026.05 | 61.7 | 26.3 | 12.6 | 25.7 |