UniBias: Unveiling and Mitigating LLM Bias through Internal Attention and FFN Manipulation | Read Paper on Bytez