List your company

Embedded Direct Attack on GSM8K ⊕ BeaverTails

74.7UtilitySFT
28.73240.66652.664.534Jun 29, 2026

Evaluation results

MethodLinks
SFTModel=Qwen2.5-7B-Instruct2026.0674.754.5
Constrained SFTModel=Qwen2.5-7B-Instruct2026.0674.548.7
LisaModel=Qwen2.5-7B-Instruct2026.0673.941.4
DR-SFTModel=Qwen2.5-7B-Instruct2026.0670.626.5
LisaModel=Gemma-3-4B-IT2026.0665.952.1
DR-SFTModel=Gemma-3-4B-IT2026.0665.419
SFTModel=Gemma-3-4B-IT2026.0664.759.7
Constrained SFTModel=Gemma-3-4B-IT2026.0664.550.7
SFTModel=Llama-2-7B-Chat2026.0641.137.7
DR-SFTModel=Llama-2-7B-Chat2026.0638.72.9
Constrained SFTModel=Llama-2-7B-Chat2026.0637.522.1
LisaModel=Llama-2-7B-Chat2026.0630.523.4