Code Generation on MBPP+ Out-of-Domain (test)
65.61AccuracyDynaSteer
Evaluation Results
| Method | Links | |
|---|---|---|
| DynaSteerBackbone=Qwen3-8B2026.06 | 65.61 | |
| DRESSBackbone=Qwen3-8B2026.06 | 63.76 | |
| ACTBackbone=Qwen3-8B2026.06 | 62.43 | |
| ITIBackbone=Qwen3-8B2026.06 | 62.17 | |
| PlainBackbone=Qwen3-8B2026.06 | 61.9 | |
| "Wait"Backbone=Qwen3-8B2026.06 | 61.64 | |
| DynaSteerBackbone=Qwen3-1.7B2026.06 | 54.5 | |
| DynaSteerBackbone=Llama3.2-3B-Instruct2026.06 | 54.5 | |
| ACTBackbone=Llama3.2-3B-Instruct2026.06 | 52.65 | |
| DRESSBackbone=Llama3.2-3B-Instruct2026.06 | 51.85 | |
| DRESSBackbone=Qwen3-1.7B2026.06 | 51.32 | |
| ACTBackbone=Qwen3-1.7B2026.06 | 51.06 | |
| ITIBackbone=Llama3.2-3B-Instruct2026.06 | 51.06 | |
| PlainBackbone=Llama3.2-3B-Instruct2026.06 | 50.53 | |
| "Wait"Backbone=Llama3.2-3B-Instruct2026.06 | 50.53 | |
| PlainBackbone=Qwen3-1.7B2026.06 | 50.26 | |
| "Wait"Backbone=Qwen3-1.7B2026.06 | 50.26 | |
| ITIBackbone=Qwen3-1.7B2026.06 | 50 |