Question Answering on HotpotQA, TriviaQA, MSMARCO, NQ-Open (macro-average)
839.81TokensRLCR
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| RLCRBackbone=Mistral-7B-Instruct2026.05 | 839.81 | 0.1436 | 65.05 | |
| RLCRBackbone=Llama-3.1-8B-Instruct2026.05 | 784.71 | 0.1745 | 61.86 | |
| RLCRBackbone=Qwen2.5-7B-Instruct2026.05 | 641.07 | 0.1443 | 67.53 | |
| BaseBackbone=Llama-3.1-8B-Instruct2026.05 | 443.51 | 0.1072 | 75.18 | |
| BaseBackbone=Mistral-7B-Instruct2026.05 | 405.79 | 0.0851 | 76.35 | |
| RLVRBackbone=Mistral-7B-Instruct2026.05 | 336.16 | 0.0649 | 75.07 | |
| RLVRBackbone=Llama-3.1-8B-Instruct2026.05 | 312.63 | 0.073 | 80.53 | |
| CSRBackbone=Llama-3.1-8B-Instruct2026.05 | 293.86 | 0.0557 | 80.76 | |
| CSRBackbone=Mistral-7B-Instruct2026.05 | 265.01 | 0.0474 | 80.67 | |
| RDBackbone=Mistral-7B-Instruct2026.05 | 193.32 | 0.3299 | 72.55 | |
| RDBackbone=Llama-3.1-8B-Instruct2026.05 | 191.5 | 0.1173 | 76.62 | |
| RDBackbone=Qwen2.5-7B-Instruct2026.05 | 156.33 | 0.1424 | 65.5 | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.05 | 127.76 | 0.0947 | 77.45 | |
| RLVRBackbone=Qwen2.5-7B-Instruct2026.05 | 123.52 | 0.0939 | 76.86 | |
| CSRBackbone=Qwen2.5-7B-Instruct2026.05 | 113.27 | 0.0616 | 83.15 |