Attribution Alignment on Curated Attribution Dataset (NarrativeQA + SciQ)
5.1DSA (Dependent Sentence Attribution)HETA
Evaluation results
| Method | Links | |
|---|---|---|
| HETAModel Architecture=LLaMA2026.04 | 5.1 | |
| HETAModel Architecture=GPT-J2026.04 | 4.8 | |
| HETAModel Architecture=Qwen2.52026.04 | 4.65 | |
| HETAModel Architecture=Phi-32026.04 | 4.25 | |
| ReAgentModel Architecture=LLaMA2026.04 | 3.78 | |
| ReAgentModel Architecture=GPT-J2026.04 | 3.6 | |
| ReAgentModel Architecture=Qwen2.52026.04 | 3.5 | |
| ReAgentModel Architecture=Phi-32026.04 | 3.35 | |
| SEA-CoTModel Architecture=LLaMA2026.04 | 3.15 | |
| SEA-CoTModel Architecture=GPT-J2026.04 | 2.92 | |
| Progressive InferenceModel Architecture=LLaMA2026.04 | 2.88 | |
| SEA-CoTModel Architecture=Qwen2.52026.04 | 2.85 | |
| SEA-CoTModel Architecture=Phi-32026.04 | 2.77 | |
| Progressive InferenceModel Architecture=Qwen2.52026.04 | 2.73 | |
| Progressive InferenceModel Architecture=GPT-J2026.04 | 2.65 | |
| Progressive InferenceModel Architecture=Phi-32026.04 | 2.4 | |
| fAMLModel Architecture=LLaMA2026.04 | 2.3 | |
| fAMLModel Architecture=Qwen2.52026.04 | 2.2 | |
| fAMLModel Architecture=GPT-J2026.04 | 2.1 | |
| fAMLModel Architecture=Phi-32026.04 | 2.05 | |
| ContextCiteModel Architecture=LLaMA2026.04 | -0.09 | |
| ContextCiteModel Architecture=GPT-J2026.04 | -0.12 | |
| ContextCiteModel Architecture=Qwen2.52026.04 | -0.14 | |
| ContextCiteModel Architecture=Phi-32026.04 | -0.18 | |
| Peering (PML)Model Architecture=LLaMA2026.04 | -0.21 | |
| Peering (PML)Model Architecture=Qwen2.52026.04 | -0.22 | |
| Peering (PML)Model Architecture=GPT-J2026.04 | -0.25 | |
| TDD-backwardModel Architecture=LLaMA2026.04 | -0.27 | |
| IGModel Architecture=LLaMA2026.04 | -0.28 | |
| TDD-backwardModel Architecture=Qwen2.52026.04 | -0.29 | |
| Peering (PML)Model Architecture=Phi-32026.04 | -0.3 | |
| IGModel Architecture=Qwen2.52026.04 | -0.31 | |
| TDD-backwardModel Architecture=GPT-J2026.04 | -0.31 | |
| IGModel Architecture=GPT-J2026.04 | -0.34 | |
| TDD-backwardModel Architecture=Phi-32026.04 | -0.36 | |
| Attention RolloutModel Architecture=LLaMA2026.04 | -0.39 | |
| IGModel Architecture=Phi-32026.04 | -0.41 | |
| Attention RolloutModel Architecture=Qwen2.52026.04 | -0.41 | |
| Attention RolloutModel Architecture=GPT-J2026.04 | -0.44 | |
| Attention RolloutModel Architecture=Phi-32026.04 | -0.52 |