Reasoning on GPQA (Avg Length, Modify Ratio, Estimated Speed, Acc, Acc Improv.)
6,763.8Avg LengthSpeculative Thinking
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Speculative ThinkingSpeculative Model=Qwen-2.5-7B-Instruct, Target Model=1.5B2025.04 | 6,763.8 | 43 | 45.6 | 31.8 | -2 | |
| Speculative ThinkingSpeculative Model=Qwen-2.5-7B-Instruct, Target Model=32B2025.04 | 6,652.8 | 31 | 33.6 | 48 | 14.2 | |
| Speculative ThinkingSpeculative Model=Qwen-2.5-7B-Instruct, Target Model=7B2025.04 | 4,739.7 | 42 | 36.8 | 40.9 | 7.1 | |
| Qwen-2.5-7B-InstructSpeculative Model=Qwen-2.5-7B-Instruct, Target Model=None2025.04 | 5.6 | — | 1.5 | 33.8 | — |