Reasoning on MMLU-Pro (Accuracy and Length)
2,993Output LengthBase
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BaseBackbone=DSQW-7B2026.05 | 2,993 | 56.61 | |
| BaseBackbone=Qwen3-8B2026.05 | 2,631 | 76.07 | |
| +GRPOBackbone=DSQW-7B2026.05 | 2,515 | 57.32 | |
| BaseBackbone=Qwen3-4B2026.05 | 2,312 | 72.86 | |
| +ICRBackbone=DSQW-7B2026.05 | 2,136 | 59.82 | |
| +LP-G (Kimi)Backbone=DSQW-7B2026.05 | 1,877 | 58.39 | |
| +LC-R1Backbone=DSQW-7B2026.05 | 1,693 | 57.67 | |
| +LP-FBackbone=DSQW-7B2026.05 | 1,688 | 57.85 | |
| +ShorterBetterBackbone=DSQW-7B2026.05 | 1,676 | 57.85 | |
| +ICR /w LP-FBackbone=DSQW-7B2026.05 | 1,666 | 59.28 | |
| +GRPOBackbone=Qwen3-8B2026.05 | 1,486 | 79.1 | |
| +GRPOBackbone=Qwen3-4B2026.05 | 1,268 | 72.32 | |
| +ICRBackbone=Qwen3-8B2026.05 | 1,195 | 79.46 | |
| +LP-FBackbone=Qwen3-4B2026.05 | 1,000 | 71.6 | |
| +LC-R1Backbone=Qwen3-8B2026.05 | 946 | 76.42 | |
| +ICRBackbone=Qwen3-4B2026.05 | 830 | 73.39 | |
| +LP-FBackbone=Qwen3-8B2026.05 | 811 | 79.28 | |
| +LC-R1Backbone=Qwen3-4B2026.05 | 805 | 72.14 | |
| +ICR /w LP-FBackbone=Qwen3-8B2026.05 | 754 | 78.03 | |
| +ICR /w LP-FBackbone=Qwen3-4B2026.05 | 752 | 73.39 | |
| +LP-G (Kimi)Backbone=Qwen3-8B2026.05 | 751 | 75.17 | |
| +ShorterBetterBackbone=Qwen3-8B2026.05 | 725 | 76.78 | |
| +LP-G (Kimi)Backbone=Qwen3-4B2026.05 | 720 | 71.25 | |
| +ShorterBetterBackbone=Qwen3-4B2026.05 | 566 | 71.25 |