Multi-Answer Question Answering on WebQSP
50.3PrecisionSPADER
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SPADERBackbone=Llama3.1-8B2026.05 | 50.3 | 45.3 | 43.8 | |
| StepSearchBackbone=Llama3.1-8B2026.05 | 49.9 | 38.7 | 40.3 | |
| R3-RAGBackbone=Llama3.1-8B2026.05 | 45.2 | 41.3 | 40.3 | |
| SPADER (QAMPARI-only)Backbone=Llama3.1-8B2026.05 | 42.7 | 39.2 | 37.8 | |
| SPADERBackbone=Qwen3-8B2026.05 | 42.3 | 40.9 | 38.8 | |
| R3-RAGBackbone=Qwen3-8B2026.05 | 38.5 | 34.3 | 33.8 | |
| PPOBackbone=Qwen3-8B2026.05 | 38 | 34 | 33.6 | |
| StepSearchBackbone=Qwen3-8B2026.05 | 37.7 | 34 | 33.3 | |
| SPADER (QAMPARI-only)Backbone=Qwen3-8B2026.05 | 35.1 | 32.8 | 32.1 | |
| GRPOBackbone=Llama3.1-8B2026.05 | 34.9 | 37.4 | 35.2 | |
| ReActBackbone=Qwen3-8B2026.05 | 34.4 | 35.1 | 31.8 | |
| GRPOBackbone=Qwen3-8B2026.05 | 33.9 | 29 | 29 | |
| PPOBackbone=Llama3.1-8B2026.05 | 32.5 | 36.8 | 32.9 | |
| ReActBackbone=Llama3.1-8B2026.05 | 32.3 | 31.3 | 28.6 | |
| RAGBackbone=Qwen3-8B2026.05 | 31.4 | 28.4 | 27.3 | |
| RAGBackbone=Llama3.1-8B2026.05 | 28.9 | 30.4 | 26.8 |