A Pidgin Gpt-Like Chatbot Model Based on Llama 2

In the modern technology scene, GPT chatbot models serve as supportive tools for customer support, customized personal use, and task assistance in various domains, leveraging concepts such as natural language processing. The Pidgin GPT-like Chatbot Model addresses the issue of limited language technology resources for Nigerian Pidgin, which is widely spoken in the country but digitally underrepresented. In this research, we developed a chatbot model that can comprehend and generate text in Nigerian Pidgin Language. The methodology includes collecting a question-and-answer dataset and fine-tuning the Llama 2 language model, a state-of-the-art open-source pre-trained language model, on a carefully curated dataset of Nigerian Pidgin. This method allows the model to adapt to the variations and linguistic features of the Pidgin language. The developed chatbot model's key features include its ability to generate a context-related response to a prompted question in Pidgin, it serves as a stepping stone for future work, with its potential to be a valuable tool for language preservation, education, and broader communication. While evaluating the trained Pidgin GPT Model, it achieved a precision result of 0.85, recall of 0.87, F1-Score of 0.86 and Cosine-Similarity of 0.81. Based on these test and evaluation results, it was confirmed that the fine-tuned Llama 2 model generates accurate context-related responses for prompts passed, with the Pidgin variations maintained, providing a solid foundation for the development of language technology focused on this language with space for future improvements. The major limitations of this work include limited size of Pidgin language dataset, and lack of efficient computational resources which limits the capabilities of the model to handle more complex linguistic structures