List your company

Code Generation on MBPP+ Out-of-Domain (test)

65.61AccuracyDynaSteer
49.375653.590357.80562.0197Jun 26, 2026

Evaluation results

MethodLinks
DynaSteerBackbone=Qwen3-8B2026.0665.61
DRESSBackbone=Qwen3-8B2026.0663.76
ACTBackbone=Qwen3-8B2026.0662.43
ITIBackbone=Qwen3-8B2026.0662.17
PlainBackbone=Qwen3-8B2026.0661.9
"Wait"Backbone=Qwen3-8B2026.0661.64
DynaSteerBackbone=Qwen3-1.7B2026.0654.5
DynaSteerBackbone=Llama3.2-3B-Instruct2026.0654.5
ACTBackbone=Llama3.2-3B-Instruct2026.0652.65
DRESSBackbone=Llama3.2-3B-Instruct2026.0651.85
DRESSBackbone=Qwen3-1.7B2026.0651.32
ACTBackbone=Qwen3-1.7B2026.0651.06
ITIBackbone=Llama3.2-3B-Instruct2026.0651.06
PlainBackbone=Llama3.2-3B-Instruct2026.0650.53
"Wait"Backbone=Llama3.2-3B-Instruct2026.0650.53
PlainBackbone=Qwen3-1.7B2026.0650.26
"Wait"Backbone=Qwen3-1.7B2026.0650.26
ITIBackbone=Qwen3-1.7B2026.0650