Automate ePub-to-JSONL dataset creation and LoRA fine-tuning.
npx skills add https://github.com/hoangnb24/skills --skill book-sft-pipeline-hoangnb24
Please help me install this Agent Skill. Skill: book-sft-pipeline Source: https://github.com/hoangnb24/skills/tree/main/skills/book-sft-pipeline Command: npx skills add https://github.com/hoangnb24/skills --skill book-sft-pipeline-hoangnb24
Book SFT Pipeline automates turning books into fine-tuning datasets and training style-transfer models, enabling author-voice style adaptation from literature.
Provide an ePub book to start generating 150-400 word training chunks and 2 variants per chunk for a LoRA fine-tuning run.
High-intent search queries and answers about installing and using this skill.
To create an SFT dataset from an ePub for LoRA fine-tuning, the pipeline extracts text, segments it into 150-400 word blocks with overlap, generates diverse prompts, and outputs a JSONL file ready for training.
The best way to segment book text for author-style transfer is using paragraph-based segmentation that creates 150-400 word blocks with overlap, ensuring robust style replication across training chunks.
Yes, this pipeline is explicitly suitable for small-to-medium dataset fine-tuning workflows, applying LoRA rank 32 to a base model like Qwen/Qwen3-8B-Base for author-voice replication.
Yes, the pipeline supports diverse instruction generation by applying multiple system prompts and templates to each segmented chunk, producing multiple training variants for LoRA fine-tuning.
The pipeline is compatible with base models such as Qwen/Qwen3-8B-Base, applying LoRA rank 32 for fine-tuning to achieve author-style transfer from the extracted ePub text.
To prepare an ePub file for JSONL dataset construction, provide the ePub to the pipeline which handles text extraction, paragraph-based segmentation, prompt generation, and outputs the final JSONL dataset.
FVossebeldConvert books into SFT datasets and orchestrate LoRA training pipelines.
phonowellConvert books into SFT-ready JSONL datasets and train LoRA models.
enigmaicon-engConvert ePub books into SFT datasets with JSONL formatting and LoRA training.
fanguyunAutomates LLM fine-tuning with LoRA/QLoRA, dataset prep, hyperparameters, and deployment optimization.
