Grammar-Constrained Decoding Can Jailbreak LLMs into Generating Malicious Code
Open paper
Benchmark
Model
Rank
Results
code-generation-on-mbpp
CodeShield (Qwen2.5-Coder-7B)
#68
Accuracy: 66.4
code-generation-on-humaneval
CodeShield (Qwen2.5-Coder-7B)
#90
Pass@1: 67.48
Rank counts only results with a code link.