Mathematical Reasoning on GSM8K (Acc, Length, η)
93AccuracyLaser
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LaserModel Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 93 | 968 | 1.852 | |
| AdaptThinkModel Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 92.6 | 1,086 | 1.635 | |
| NCoTS (Ours)Model Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 92.6 | 899 | 1.976 | |
| NoWaitModel Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 91.3 | 1,147 | 1.506 | |
| OriginalModel Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 90.3 | 1,690 | 1 | |
| MeanModel Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 88.1 | 1,649 | 0.976 | |
| AdaptThinkModel Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 85.4 | 1,109 | 1.848 | |
| NCoTS (Ours)Model Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 85.4 | 954 | 2.148 | |
| ThinkPruneModel Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 84.9 | 1,191 | 1.702 | |
| NoWaitModel Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 84.1 | 1,211 | 1.641 | |
| OriginalModel Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 83 | 1,938 | 1 | |
| LaserModel Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 82.3 | 1,064 | 1.792 | |
| MeanModel Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 80.9 | 1,846 | 0.997 |