MohamedRashad/Arabic-Orpo-Llama-3-8B-Instruct

MohamedRashad/Arabic-Orpo-Llama-3-8B-Instruct is an 8 billion parameter instruction-tuned causal language model fine-tuned by MohamedRashad from Meta-Llama-3-8B-Instruct using the ORPO method. This model is specifically optimized for generating coherent Arabic text, despite formal evaluations showing a slight decrease in Arabic MMLU scores compared to its base model. It aims to better align the Llama 3 architecture for Arabic language tasks, offering a practical improvement in Arabic text generation.