Toward Preference-aligned Large Language Models via Residual-based Model Steering

LLMSVRe-assign community