Translating ePub Ebooks with Local LLM and for Free. An Effective Approach (grigio.org)

A practical guide shows how to translate entire EPUB ebooks for free by combining a purpose-built tool (TranslateBookWithLLM) with local LLMs (via Ollama or llama.cpp) instead of paid APIs. This matters because ebook translation requires preserving structure—HTML/XML markup, code blocks, formulas, images, and links—while respecting LLM context-window limits. The approach emphasizes chunking the book into manageable pieces, handling markup to keep formatting, and special-casing technical content so translations remain accurate and readable. The guide also calls out resource constraints on free cloud tiers (Colab timeouts) and offers a Colab notebook with Ollama + a T4 GPU usable for short sessions. Technically, TranslateBookWithLLM automates EPUB unpacking, text chunking, and LLM interactions via a simple CLI (with an optional web UI). Recommended local models include mistral-nemo—noted as faster and robust for structured translation (about 2x faster than gpt-oss in llm-eval-simple tests)—and gpt-oss 20B (more capable but prone to Colab timeouts). Example usage: python translate.py --provider ollama --api_endpoint http://localhost:11434/api/generate -sl English -tl Italian -i ../example.epub -o ../example-it.epub -m mistral-nemo. The workflow offers a cost-free, privacy-friendly path to high-quality book translation, with options to scale up via paid compute for speed or higher fidelity.