chartreuse-verte/prose-rewriter-1.7b-v1.2

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 21, 2026License:agpl-3.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The chartreuse-verte/prose-rewriter-1.7b-v1.2 is a 2 billion parameter model based on Qwen3-1.7B-Base, specifically designed for paragraph-level prose rewriting. It re-renders LLM-generated text to sound more human while preserving original semantics. This model excels at transforming fictional prose, offering distinct modes for length control, and is optimized for single-paragraph inputs.

Loading preview...

Overview

The chartreuse-verte/prose-rewriter-1.7b-v1.2 is a specialized 2 billion parameter language model, built upon Qwen/Qwen3-1.7B-Base with a rank-16 LoRA merge. Its core function is to rewrite prose, particularly text generated by other large language models, to make it sound more natural and human-like, without altering the original meaning. This model is the smaller of two checkpoints trained on the same pipeline, with a larger 4B version also available.

Key Capabilities

  • Prose Rewriting: Transforms LLM-generated prose into more human-sounding text.
  • Semantic Preservation: Designed to maintain the original meaning of the input paragraph.
  • Length Control Modes: Supports three edit modes (match, inflate, compress) to control the output length relative to the input, with match being the recommended default for in-place rewriting.
  • Fictional Prose Focus: Optimized for fictional narratives, including first and third-person fiction and dialogue.
  • Input Handling: Performs best with inputs between 15 and 120 words; shorter inputs (below ~15 words) may result in padding and invention of new material.
  • Memory Efficiency: The Q8_0 GGUF variant requires approximately 2.08 GB of VRAM for weights, with each 1280-token slot adding 140 MiB of KV cache.

Limitations

  • Not an Instruct Model: It has a single, specific task and prompt format; it is not designed for general instruction following.
  • Fictional Prose Only: Not suitable for technical documentation or other non-fictional content.
  • Single Paragraph Input: Designed for one paragraph per call; longer inputs should be split.
  • English Only: Primarily supports English narrative register.
  • AI Detector Vulnerability: Will not bypass AI content detectors as it preserves certain word choices and sentence structures.

Finetunes
1 models