Multi-Hop Question Answering on Bamb
54.4Exact Match (EM)Tree-GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| Tree-GRPOBackbone=Qwen2.5-14b2025.09 | 54.4 | |
| GRPOBackbone=Qwen2.5-14b2025.09 | 53.6 | |
| GSPOBackbone=Qwen2.5-14b2025.09 | 52.8 | |
| Tree-GRPOBackbone=Llama3.2-3b2025.09 | 46.4 | |
| GSPOBackbone=Qwen2.5-7b2025.09 | 44 | |
| Tree-GRPOBackbone=Qwen2.5-7b2025.09 | 44 | |
| Tree-GRPOBackbone=Qwen2.5-3b2025.09 | 43.2 | |
| GRPOBackbone=Qwen2.5-7b2025.09 | 43.2 | |
| ReActBackbone=Qwen2.5-14b2025.09 | 43.2 | |
| Search-o1Backbone=Qwen2.5-14b2025.09 | 41.6 | |
| GRPOBackbone=Qwen2.5-3b2025.09 | 36.8 | |
| GSPOBackbone=Qwen2.5-3b2025.09 | 36.8 | |
| Search-o1Backbone=Llama3.2-3b2025.09 | 32 | |
| GRPOBackbone=Llama3.2-3b2025.09 | 32 | |
| GSPOBackbone=Llama3.2-3b2025.09 | 32 | |
| Search-o1Backbone=Qwen2.5-7b2025.09 | 32 | |
| Search-o1Backbone=Qwen2.5-3b2025.09 | 29.6 | |
| ReActBackbone=Qwen2.5-7b2025.09 | 28.8 | |
| ReActBackbone=Llama3.2-3b2025.09 | 23.2 | |
| ReActBackbone=Qwen2.5-3b2025.09 | 20 | |
| Direct InferenceBackbone=Qwen2.5-14b2025.09 | 16 | |
| Tree-GRPOBackbone=Qwen2.5-1.5b2025.09 | 13.6 | |
| Search-o1Backbone=Qwen2.5-1.5b2025.09 | 13 | |
| Direct InferenceBackbone=Qwen2.5-7b2025.09 | 12 | |
| ReActBackbone=Qwen2.5-1.5b2025.09 | 11.2 | |
| Direct InferenceBackbone=Qwen2.5-1.5b2025.09 | 8 | |
| Direct InferenceBackbone=Llama3.2-3b2025.09 | 8 | |
| GSPOBackbone=Qwen2.5-1.5b2025.09 | 4.8 | |
| GRPOBackbone=Qwen2.5-1.5b2025.09 | 4 | |
| Direct InferenceBackbone=Qwen2.5-3b2025.09 | 2.4 |