pps121/LLM-Cultural-Model-Merging

A research project on Model Merging. Two fine-tuned culturally aligned models (e.g., African and Latin American) to produce offspring that are culturally aligned and do finetuning using LoRA to get culturally finetuned offspring using Fisher Merging # LLM-Cultural-Model-Merging A research project on Model Merging. Two fine-tuned culturally aligned models (e.g., African and Latin American) by doing finetuning using LoRA to produce offspring that are culturally aligned offspring using Fisher Merging. My recent works are centered on the intersection areas of human value alignment on explainable AI and fairness, creative capabilities in the applications of LLMs on mental health, detecting and mitigating biases and stereotypes, perceived transparency, understandability with cognitive explainability, and increasing diversity in LLMs. Recently, I created a dataset and two fine-tuned models on Llama-3.1-Tulu-3-8B, available on HuggingFace, to build socially and culturally aligned LLMs across various dimensions. Such datasets are useful for working on low-resource languages, socially aware human-AI communication in social interactions, social well-being in everyday working life, and for adapting AI models to global cultures.