Question Answering on TriviaQA Out-of-distribution (test)
72.64AccuracyRLVR
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| RLVRBackbone=Llama-3.1-8B-Instruct2026.05 | 72.64 | 2.43 | 80.42 | |
| RDBackbone=Llama-3.1-8B-Instruct2026.05 | 71.1 | 6.67 | 80.3 | |
| RLVRBackbone=Mistral-7B-Instruct2026.05 | 70.54 | 3.12 | 78.51 | |
| CSRBackbone=Mistral-7B-Instruct2026.05 | 69.76 | 1.74 | 84.36 | |
| BaseBackbone=Mistral-7B-Instruct2026.05 | 69.66 | 5.85 | 81.94 | |
| BaseBackbone=Llama-3.1-8B-Instruct2026.05 | 69.16 | 5.87 | 82.09 | |
| CSRBackbone=Llama-3.1-8B-Instruct2026.05 | 69.05 | 1.96 | 85.77 | |
| RLCRBackbone=Mistral-7B-Instruct2026.05 | 68.7 | 6.66 | 66.61 | |
| RLCRBackbone=Llama-3.1-8B-Instruct2026.05 | 68.2 | 18.42 | 56.87 | |
| RDBackbone=Mistral-7B-Instruct2026.05 | 64.2 | 58.2 | 76.68 | |
| RLCRBackbone=Qwen2.5-7B-Instruct2026.05 | 63.6 | 4.01 | 71.19 | |
| RLVRBackbone=Qwen2.5-7B-Instruct2026.05 | 59.4 | 3.86 | 80.13 | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.05 | 58.17 | 4.35 | 81.81 | |
| RDBackbone=Qwen2.5-7B-Instruct2026.05 | 57.8 | 18.59 | 66.68 | |
| CSRBackbone=Qwen2.5-7B-Instruct2026.05 | 56.85 | 1.84 | 87.02 |