Question Answering on CRAG
20.1Finance ScoreTTARAG
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| TTARAGBackbone=Llama-3.1-8b-it, Evaluation Protocol=Test-time adaptation2026.01 | 20.1 | 29.5 | 37.7 | 34.6 | 41.5 | 31.9 | |
| Chain-of-ThoughtBackbone=Llama-3.1-8b-it, Evaluation Protocol=Prompting technique2026.01 | 17.9 | 30.2 | 37.6 | 31.5 | 45.8 | 31.6 | |
| naive-ragBackbone=Llama-3.1-8b-it, Evaluation Protocol=Standard retrieval-augmentation (top 20 passages)2026.01 | 17.4 | 27.6 | 34.9 | 31.3 | 42.4 | 29.8 | |
| TTARAGBackbone=Llama-2-7b-chat, Evaluation Protocol=Test-time adaptation2026.01 | 16.4 | 25.8 | 40.7 | 33.8 | 41.1 | 30.5 | |
| In-Context LearningBackbone=Llama-3.1-8b-it, Evaluation Protocol=Prompting technique2026.01 | 16.1 | 24.8 | 33.5 | 29.4 | 40.4 | 28 | |
| In-Context LearningBackbone=Llama-2-7b-chat, Evaluation Protocol=Prompting technique2026.01 | 16 | 24.2 | 36.1 | 31.2 | 39.2 | 28.4 | |
| Chain-of-ThoughtBackbone=Llama-2-7b-chat, Evaluation Protocol=Prompting technique2026.01 | 15.7 | 26.7 | 34.3 | 31.4 | 41.5 | 29.1 | |
| naive-ragBackbone=Llama-2-7b-chat, Evaluation Protocol=Standard retrieval-augmentation (top 20 passages)2026.01 | 14.7 | 23.2 | 36.5 | 30.4 | 39.2 | 27.8 | |
| TTARAGBackbone=ChatGLM-3-6b, Evaluation Protocol=Test-time adaptation2026.01 | 14 | 22.1 | 33.5 | 25.5 | 38.1 | 25.7 | |
| Chain-of-ThoughtBackbone=ChatGLM-3-6b, Evaluation Protocol=Prompting technique2026.01 | 12.7 | 20.6 | 28.4 | 25.8 | 33.9 | 23.6 | |
| In-Context LearningBackbone=ChatGLM-3-6b, Evaluation Protocol=Prompting technique2026.01 | 9.9 | 18.2 | 30.8 | 22.1 | 33 | 21.8 | |
| naive-ragBackbone=ChatGLM-3-6b, Evaluation Protocol=Standard retrieval-augmentation (top 20 passages)2026.01 | 9.8 | 18.7 | 31.4 | 22.4 | 33.4 | 22 |