response-alignment

PersonaVLM introduces a personalized multimodal LLM framework that enables long-term user adaptation through memory retention, multi-turn reasoning, and response alignment, outperforming GPT-4o by 5.2% on the new Persona-MME benchmark.