Jailbreak Instruction-Tuned LLMs via end-of-sentence MLP Re-weighting