A new open-source project, poetry_dataset, offers a complete pipeline for fine-tuning large language models to generate classical Chinese poetry, specifically designed for Apple Silicon Macs. The project leverages the MLX framework, enabling developers with as little as 16GB of RAM to fine-tune models from the Qwen series using LoRA. It includes a curated poetry training dataset, data conversion scripts, and even a tone-checking tool for classical Chinese prosody. The creator also launched a demo site, shi-ci.cn, showcasing the fine-tuned model in action. This initiative lowers the barrier for developers interested in domain-specific fine-tuning, particularly for cultural and linguistic applications, without requiring expensive GPU clusters. The project is fully open-source and available on GitHub, making it a practical resource for NLP researchers, indie hackers, and anyone exploring AI-generated classical poetry.
A developer has released poetry_dataset, an open-source dataset and complete pipeline for fine-tuning LLMs to generate classical Chinese poetry, optimized for Apple Silicon Macs using the MLX framework. The project includes training data, conversion scripts, and a live demo site (shi-ci.cn), making it accessible to developers without expensive GPU hardware. This is a valuable resource for NLP practitioners interested in domain-specific fine-tuning and cultural AI applications.