Mathematical Reasoning on AMC23 In-Domain (test)
72.5AccuracyDynaSteer
Evaluation Results
| Method | Links | |
|---|---|---|
| DynaSteerBackbone=Qwen3-8B2026.06 | 72.5 | |
| ACTBackbone=Qwen3-8B2026.06 | 70 | |
| ITIBackbone=Qwen3-8B2026.06 | 67.5 | |
| DRESSBackbone=Qwen3-8B2026.06 | 67.5 | |
| PlainBackbone=Qwen3-8B2026.06 | 65 | |
| "Wait"Backbone=Qwen3-8B2026.06 | 65 | |
| DynaSteerBackbone=Qwen3-1.7B2026.06 | 47.5 | |
| ACTBackbone=Qwen3-1.7B2026.06 | 42.5 | |
| DRESSBackbone=Qwen3-1.7B2026.06 | 42.5 | |
| PlainBackbone=Qwen3-1.7B2026.06 | 40 | |
| "Wait"Backbone=Qwen3-1.7B2026.06 | 40 | |
| ITIBackbone=Qwen3-1.7B2026.06 | 37.5 | |
| DynaSteerBackbone=Llama3.2-3B-Instruct2026.06 | 35 | |
| ACTBackbone=Llama3.2-3B-Instruct2026.06 | 32.5 | |
| DRESSBackbone=Llama3.2-3B-Instruct2026.06 | 30 | |
| PlainBackbone=Llama3.2-3B-Instruct2026.06 | 27.5 | |
| "Wait"Backbone=Llama3.2-3B-Instruct2026.06 | 25 | |
| ITIBackbone=Llama3.2-3B-Instruct2026.06 | 25 |