Book Style-Transfer SFT Pipeline

This skill provides an end-to-end framework for building style-transfer models through sophisticated context engineering and data preparation. It guides users through extracting clean text from ePub files, performing semantic segmentation to maintain prose flow, and generating diverse synthetic instructions to prevent model memorization. Designed for use with platforms like Tinker and base models like Qwen, the pipeline ensures that the resulting fine-tuned models capture the rhythm, vocabulary, and stylistic nuances of an author rather than just parroting plot points. It includes built-in validation steps to test style generalization in modern scenarios.

主な機能

ユースケース