book-sft-pipeline

What problem does it solve?

This Skill automates turning full-length books into SFT datasets and author-style training pipelines. It handles extraction from books, segmentation into training chunks, generation of diverse prompts, and packaging of data for LoRA fine-tuning on base models.

Core Features & Use Cases

• Text extraction: Convert books (ePub/text) into clean, model-ready text.
• Intelligent segmentation: Produce 150-400 word chunks with overlaps to preserve narrative style.
• Diverse instruction generation: Use multiple templates and system prompts to prevent memorization of source text.
• Dataset construction: Build JSONL datasets compatible with Tinker-style LoRA fine-tuning workflows.
• Use Case: Train a small to mid-size model to imitate a specific author’s voice using a modular SFT pipeline.

Quick Start

Copy the SKILL.md into your project skills folder and run the conceptual pipeline example to validate the flow and artifacts.