Mathematical Reasoning on MATH500 In-Domain (test)
86.2AccuracyDynaSteer
Evaluation Results
| Method | Links | |
|---|---|---|
| DynaSteerBackbone=Qwen3-8B2026.06 | 86.2 | |
| ACTBackbone=Qwen3-8B2026.06 | 84.8 | |
| DRESSBackbone=Qwen3-8B2026.06 | 84.6 | |
| "Wait"Backbone=Qwen3-8B2026.06 | 84.2 | |
| ITIBackbone=Qwen3-8B2026.06 | 84.2 | |
| PlainBackbone=Qwen3-8B2026.06 | 84 | |
| DynaSteerBackbone=Qwen3-1.7B2026.06 | 74.8 | |
| DRESSBackbone=Qwen3-1.7B2026.06 | 72.8 | |
| ACTBackbone=Qwen3-1.7B2026.06 | 72.6 | |
| ITIBackbone=Qwen3-1.7B2026.06 | 71.8 | |
| "Wait"Backbone=Qwen3-1.7B2026.06 | 71.2 | |
| PlainBackbone=Qwen3-1.7B2026.06 | 69.2 | |
| DynaSteerBackbone=Llama3.2-3B-Instruct2026.06 | 52.4 | |
| ACTBackbone=Llama3.2-3B-Instruct2026.06 | 48.8 | |
| DRESSBackbone=Llama3.2-3B-Instruct2026.06 | 48.6 | |
| ITIBackbone=Llama3.2-3B-Instruct2026.06 | 48.4 | |
| PlainBackbone=Llama3.2-3B-Instruct2026.06 | 47.8 | |
| "Wait"Backbone=Llama3.2-3B-Instruct2026.06 | 46.2 |