List your company

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

81.71AccuracyDynaSteer
46.19455.414564.63573.8555Jun 26, 2026

Evaluation results

MethodLinks
DynaSteerBackbone=Qwen3-8B2026.0681.71
DRESSBackbone=Qwen3-8B2026.0680.49
PlainBackbone=Qwen3-8B2026.0679.88
ACTBackbone=Qwen3-8B2026.0679.88
ITIBackbone=Qwen3-8B2026.0679.27
"Wait"Backbone=Qwen3-8B2026.0678.66
DynaSteerBackbone=Qwen3-1.7B2026.0667.68
"Wait"Backbone=Qwen3-1.7B2026.0665.85
ACTBackbone=Qwen3-1.7B2026.0665.85
PlainBackbone=Qwen3-1.7B2026.0665.24
ITIBackbone=Qwen3-1.7B2026.0665.24
DRESSBackbone=Qwen3-1.7B2026.0665.24
DynaSteerBackbone=Llama3.2-3B-Instruct2026.0654.27
ACTBackbone=Llama3.2-3B-Instruct2026.0651.83
DRESSBackbone=Llama3.2-3B-Instruct2026.0651.22
ITIBackbone=Llama3.2-3B-Instruct2026.0648.78
"Wait"Backbone=Llama3.2-3B-Instruct2026.0648.17
PlainBackbone=Llama3.2-3B-Instruct2026.0647.56