Question Answering with Abstention on ABSTAIN-QA
83.4Accuracy (A)DeepSeek-R1
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DeepSeek-R1Model=DeepSeek-R12026.04 | 83.4 | 0.2 | 9.1 | |
| DeepSeek-V3Model=DeepSeek-V32026.04 | 77.5 | 4.3 | 31.2 | |
| Qwen2.5 32B InstructModel=Qwen2.5 32B Instruct2026.04 | 70.7 | 8.5 | 34.7 | |
| Llama3.1 8B InstructModel=Llama3.1 8B Instruct2026.04 | 59.8 | 1 | 10.3 | |
| Qwen2.5 7B InstructModel=Qwen2.5 7B Instruct2026.04 | 58.9 | 8.8 | 35.5 | |
| Abstain-R1Model=Abstain-R12026.04 | 53.3 | 16.8 | 40.1 | |
| Qwen2.5 3B InstructModel=Qwen2.5 3B Instruct2026.04 | 52.9 | 15.3 | 30 |