Fordentinc/book-builder-bookwriter-v1
Fordentinc/book-builder-bookwriter-v1 is a 7.6 billion parameter prose-writing model, fine-tuned on Qwen 2.5 7B. It was trained on 10,211 human-authored novels (1.82 billion tokens) to generate novel chapters from structured story bibles. This model excels at producing readable prose within a specific story bible-to-chapter format, making it ideal for automated long-form content generation in creative writing pipelines.
Loading preview...
Overview
Fordentinc/book-builder-bookwriter-v1 is an early release (v1) of a 7.6 billion parameter model, built on Qwen 2.5 7B, specifically designed for prose generation. It has been fine-tuned on a massive corpus of 10,211 human-authored novels, totaling approximately 1.82 billion tokens, to generate novel chapters from structured story bibles.
Key Capabilities
- Prose Generation: Produces coherent and readable chapter prose based on a provided story bible.
- Structured Input: Expects a specific input format including title, author, genre, synopsis, and chapter title.
- Pipeline Integration: Best utilized as a prose-style backbone within structured pipelines (like BookBuilder) that provide per-chapter beat sheets and plot anchors.
Limitations (v1 Early Release)
- Plot Drift: May treat bibles as style prompts rather than strict plot instructions, leading to drift from the synopsis in one-shot generation.
- Instruction Following: Does not reliably follow "FORBIDDEN" rules, character role assignments, or per-character constraints.
- Repetition: Can drift into paragraph-level repetition loops on longer generations without tuned sampling defaults.
Good For
- Developers building automated novel writing tools or creative content generation pipelines.
- Generating long-form prose content where a structured story bible can guide the output.
- Experimenting with a model specifically trained for narrative chapter generation, understanding its current capabilities and limitations.