Skip to main content
Generaldraphonix

book-sft-pipeline

Use when the user asks to fine-tune on books, create an SFT dataset from books, train a style-transfer or author-voice model, extract ePub text, segment long-form book content, or prepare literary data for LoRA or small-model training.

Stars
218
Source
draphonix/skills
Updated
2026-05-20
Slug
draphonix--skills--book-sft-pipeline
View on GitHubRaw SKILL.md

// install — copy + paste into any project

mkdir -p .claude/skills && curl -fsSL https://raw.githubusercontent.com/draphonix/skills/HEAD/plugins/khuym/skills/book-sft-pipeline/SKILL.md -o .claude/skills/book-sft-pipeline.md

Drops the SKILL.md into .claude/skills/book-sft-pipeline.md. Works with Claude Code, Cursor, and any agent that loads SKILL.md files from .claude/skills/.

Book SFT Pipeline

Convert long-form books into supervised fine-tuning data for literary style transfer. Keep the entrypoint lightweight: use this file to route the work, then open only the references needed for the current phase.

Activate When

  • Building fine-tuning datasets from literary works
  • Creating author-voice or style-transfer models
  • Preparing training data for Tinker or similar SFT platforms
  • Designing text segmentation pipelines for long-form content
  • Training small models on limited literary data

Fast Path

  1. Confirm the source is suitable: prefer ePub over PDF, remove front/back matter, and preserve paragraph breaks.
  2. Segment into coherent 150-400 word chunks. Never break mid-sentence.
  3. Generate synthetic scene descriptions without quoting the source text.
  4. Build JSONL conversation examples with varied system prompts and user templates.
  5. Train a LoRA on a base model, not an instruction-tuned model.
  6. Validate on modern scenarios and grep training data for suspicious phrases.

Default Parameters

Setting Default
Chunk size 150-400 words
Prompt diversity 15+ templates, 5+ system prompts
Variants 2 per chunk
Model Qwen/Qwen3-8B-Base or another base 8B-class model
LoRA rank 32
Epochs 3
Test set 50 examples minimum

Core Rules

  1. Source ePub before PDF because OCR noise becomes learned behavior.
  2. Keep chunks semantically complete and paragraph-bounded where possible.
  3. Teach style, not plot: instructions should describe scenes without quoting.
  4. Rotate prompt and system templates to reduce memorization.
  5. Use base models for malleable style transfer.
  6. Validate originality before claiming the model learned the author's voice.

Progressive Disclosure

Open these only when the task reaches that layer:

Implementation Starter

Use the sample script when the user wants executable scaffolding:

python plugins/khuym/skills/book-sft-pipeline/scripts/pipeline_example.py

The script demonstrates the same pipeline semantics as this skill: segmentation, diverse prompt construction, Tinker datum construction, and originality checks.

Validation Checklist

  • Chunks end at natural grammatical boundaries.
  • JSONL rows contain system, user, and assistant messages.
  • Prompt variants are distributed across chunks.
  • Held-out test examples are excluded from training.
  • Modern scenario outputs contain style markers without original plot content.
  • Exact output phrases do not appear in the training JSONL.

References

Internal references:

External resources:

Skill Metadata

Created: 2025-12-26 Last Updated: 2025-12-28 Author: Muratcan Koylan Version: 2.0.0 Standalone: Yes