Multi-hop reasoning on 2WikiMultihopQA
48.44Exact Match (EM)Prompt-R1
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Prompt-R1Backbone=GPT-4o-mini2025.11 | 48.44 | 54.41 | |
| CoT ReasoningBackbone=GPT-4o-mini2025.11 | 43.75 | 49.13 | |
| SFTBackbone=Qwen3-4B2025.11 | 41.41 | 42.62 | |
| GEPAOptimization Framework=APO, Backbone=GPT-4o-mini2025.11 | 41.41 | 46.27 | |
| GRPOBackbone=Qwen3-4B2025.11 | 34.38 | 35.05 | |
| BaselineBackbone=GPT-4o-mini2025.11 | 33.59 | 36.57 | |
| BaselineBackbone=Qwen3-4B2025.11 | 28.13 | 29.32 | |
| OPROOptimization Framework=APO, Backbone=GPT-4o-mini2025.11 | 25 | 35.96 | |
| CoT ReasoningBackbone=Qwen3-4B2025.11 | 21.88 | 24.17 | |
| TextGradOptimization Framework=APO, Backbone=GPT-4o-mini2025.11 | 18.75 | 27.5 |