Mathematical Reasoning on GSM8K In-Domain (test)
94.39AccuracyDynaSteer
Evaluation Results
| Method | Links | |
|---|---|---|
| DynaSteerBackbone=Qwen3-8B2026.06 | 94.39 | |
| ACTBackbone=Qwen3-8B2026.06 | 92.8 | |
| DRESSBackbone=Qwen3-8B2026.06 | 92.72 | |
| "Wait"Backbone=Qwen3-8B2026.06 | 92.57 | |
| PlainBackbone=Qwen3-8B2026.06 | 92.49 | |
| ITIBackbone=Qwen3-8B2026.06 | 92.27 | |
| DynaSteerBackbone=Qwen3-1.7B2026.06 | 82.87 | |
| ACTBackbone=Qwen3-1.7B2026.06 | 81.73 | |
| DRESSBackbone=Qwen3-1.7B2026.06 | 81.65 | |
| ITIBackbone=Qwen3-1.7B2026.06 | 81.5 | |
| DynaSteerBackbone=Llama3.2-3B-Instruct2026.06 | 81.35 | |
| "Wait"Backbone=Qwen3-1.7B2026.06 | 81.05 | |
| ACTBackbone=Llama3.2-3B-Instruct2026.06 | 80.29 | |
| PlainBackbone=Llama3.2-3B-Instruct2026.06 | 80.14 | |
| "Wait"Backbone=Llama3.2-3B-Instruct2026.06 | 80.06 | |
| DRESSBackbone=Llama3.2-3B-Instruct2026.06 | 80.06 | |
| ITIBackbone=Llama3.2-3B-Instruct2026.06 | 79.98 | |
| PlainBackbone=Qwen3-1.7B2026.06 | 79.61 |