Reasoning on MMLU STEM (out-of-distribution)
93.3AccuracyQwen3-4B-Original
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen3-4B-OriginalCondition=Original2026.02 | 93.3 | 2,862 | — | — | |
| Qwen3-4B-ConPressCondition=ConPress2026.02 | 92.9 | 1,816 | -0.4 | -36.6 | |
| R1-7B-OriginalCondition=Original2026.02 | 83.5 | 1,548 | — | — | |
| R1-7B-ConPressCondition=ConPress2026.02 | 82.4 | 1,035 | -1.1 | -33.1 |