List your company

Embedded Direct Attack on GSM8K ⊕ HEx-PHI

76.1UtilitySFT
29.19641.37353.5565.727Jun 29, 2026

Evaluation results

MethodLinks
SFTModel=Qwen2.5-7B-Instruct2026.0676.181.2
Constrained SFTModel=Qwen2.5-7B-Instruct2026.0674.851.8
DR-SFTModel=Qwen2.5-7B-Instruct2026.0674.433.3
LisaModel=Qwen2.5-7B-Instruct2026.0673.957.7
DR-SFTModel=Gemma-3-4B-IT2026.0664.542
SFTModel=Gemma-3-4B-IT2026.0663.491.5
LisaModel=Gemma-3-4B-IT2026.066372.8
Constrained SFTModel=Gemma-3-4B-IT2026.0662.151.8
SFTModel=Llama-2-7B-Chat2026.0640.642.2
DR-SFTModel=Llama-2-7B-Chat2026.0638.11.3
Constrained SFTModel=Llama-2-7B-Chat2026.0637.420.3
LisaModel=Llama-2-7B-Chat2026.063122.7