LAGOS, Nigeria (CHATNEWSTV) — In a bid to close Nigeria’s gap in developing large language models tailored for local contexts, Saheed Azeez, a machine learning engineer, announced the release of YarnGPT in a LinkedIn post on Thursday.
“YarnGPT is a major step toward solving this challenge,” Azeez wrote. “It’s a family of open-source text-to-speech models designed specifically for Nigerian-accented English (YarnGPT) and native languages like Igbo, Yoruba, and Hausa (YarnGPT-local).”
Built on SmolLM2-360M by Hugging Face, the models were entirely trained on Google Colab. Azeez said the project aims to encourage collaboration by making the models and training codes fully open source.
“I encourage developers to experiment with these models, share feedback, and contribute to improving them,” he wrote, adding that the repositories are now available on Hugging Face and GitHub.
Azeez credited the Hugging Face team and the global open-source community for their contributions, saying, “This project stands on the shoulders of many incredible open-source efforts.”
Developers are encouraged to experiment with the models and provide feedback to refine and enhance their capabilities.
