RAG Reranking on SQuAD 2.0
69.08PrecisionRicci-Filtration
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ricci-FiltrationGeneration LLM=gpt-4o-mini2026.06 | 69.08 | 73.58 | 70.17 | 73 | |
| No RerankerGeneration LLM=gpt-4o-mini2026.06 | 65.34 | 70.89 | 66.46 | 69.5 | |
| Cross-Encoder RerankerGeneration LLM=gpt-4o-mini2026.06 | 63.6 | 69.46 | 64.86 | 68.2 | |
| Ricci-FiltrationGeneration LLM=Llama 3.1-8B Instruct2026.06 | 52.13 | 57.28 | 53 | 55.4 | |
| Cross-Encoder RerankerGeneration LLM=Llama 3.1-8B Instruct2026.06 | 47.74 | 52.14 | 48.49 | 50.9 | |
| No RerankerGeneration LLM=Llama 3.1-8B Instruct2026.06 | 47.64 | 52.25 | 48.48 | 52 | |
| LLM-based RerankerGeneration LLM=Llama 3.1-8B Instruct2026.06 | 37.86 | 41.2 | 38.25 | 49.5 | |
| LLM-based RerankerGeneration LLM=gpt-4o-mini2026.06 | 37.52 | 42.58 | 38.56 | 52.1 |