Knowledge-intensive Dialogue on OpendialKG
88.84Factual AccuracyFine-Refine
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Fine-RefineBackbone=Llama-3.1-8B-Instruct, Refinement Iterations=32026.02 | 88.84 | 36.51 | 2.859 | 2.512 | |
| Self-RefineBackbone=Llama-3.1-8B-Instruct, Refinement Iterations=32026.02 | 88.83 | 48.91 | 2.915 | 2.545 | |
| Fine-RefineBackbone=Qwen3-8B, Refinement Iterations=32026.02 | 87.91 | 56.99 | 2.962 | 2.949 | |
| Fine-RefineBackbone=Mistral-7B-Instruct-v0.3, Refinement Iterations=32026.02 | 86.81 | 46.52 | 2.945 | 2.627 | |
| Mistral-7B-Instruct-v0.3Backbone=Mistral-7B-Instruct-v0.32026.02 | 86.79 | 49.6 | 2.988 | 2.914 | |
| Self-RefineBackbone=Qwen3-8B, Refinement Iterations=32026.02 | 85.14 | 60.65 | 2.952 | 2.902 | |
| Llama-3.1-8B-InstructBackbone=Llama-3.1-8B-Instruct2026.02 | 85 | 46.27 | 2.919 | 2.69 | |
| G-Retriever2026.02 | 84.79 | 62.65 | 2.895 | 2.927 | |
| Qwen3-8BBackbone=Qwen3-8B2026.02 | 84.16 | 59.36 | 2.973 | 2.965 | |
| Self-RAG2026.02 | 84.06 | 49.53 | 2.462 | 2.418 | |
| Self-RefineBackbone=Mistral-7B-Instruct-v0.3, Refinement Iterations=32026.02 | 77.12 | 59.2 | 2.893 | 2.652 |