Question Answering on ARC-C (Acc, Length, η)
82.9AccuracyAdaptThink
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AdaptThinkModel Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 82.9 | 1,088 | 1.198 | |
| NCoTS (Ours)Model Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 82.6 | 979 | 1.322 | |
| OriginalModel Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 80.6 | 1,232 | 1 | |
| NoWaitModel Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 80.3 | 1,082 | 1.13 | |
| LaserModel Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 79.9 | 944 | 1.283 | |
| MeanModel Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 77.1 | 1,147 | 0.983 | |
| NCoTS (Ours)Model Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 54.9 | 778 | 1.405 | |
| ThinkPruneModel Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 51.5 | 768 | 1.252 | |
| NoWaitModel Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 50.8 | 812 | 1.152 | |
| LaserModel Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 50.8 | 870 | 1.075 | |
| AdaptThinkModel Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 50.5 | 897 | 1.031 | |
| OriginalModel Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 49.8 | 899 | 1 | |
| MeanModel Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 47.2 | 856 | 0.943 |