chartreuse-verte/prose-rewriter-1.7b-v1.4

Model Overview

The prose-rewriter-1.7b-v1.4 is a specialized 1.7 billion parameter model, based on Qwen/Qwen3-1.7B-Base, engineered to rewrite and humanize prose generated by large language models. It aims to preserve the original meaning while enhancing the naturalness and stylistic variety of the text. This version, a successor to v1.2, demonstrates improved restructuring and reduced verbatim copying, particularly in sentence-length variation.

Key Capabilities

• Paragraph-level Prose Rewriting: Transforms LLM-generated text into more human-like prose.
• Semantic Preservation: Maintains the original meaning and content during rewriting.
• Stylistic Enhancement: Significantly increases sentence-length variety, mimicking human writing patterns.
• Controlled Rewriting: Supports match, inflate, and compress modes to control output length relative to the input.
• Efficient Performance: Achieves substantial rewriting with a relatively small parameter count (1.7B).

Use Cases and Limitations

This model is primarily designed for fictional prose in English, suitable for refining narrative registers (third and first person fiction, dialogue). It operates on one paragraph per call; longer inputs should be split. The model is not an instruct model and has a specific prompt format. It is not intended for technical documentation and will not bypass AI detectors. Input paragraphs should be at least 15 words; shorter inputs may lead to padding and fabrication. The model tends to reuse nouns more frequently than typical LLMs, reflecting human prose habits.