ChisatoY/dolas-zh-finetuned

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 21, 2026License:cc-by-nc-sa-4.0Architecture:Transformer Open Weights Featherless Exclusive Cold

ChisatoY/dolas-zh-finetuned is a 4 billion parameter Qwen3-based model fine-tuned for naturalizing AI-generated Chinese text. Developed by ChisatoY, it specializes in rewriting AI-style Chinese into more natural human-like Chinese while preserving facts, tone, and key information. This model is optimized for Chinese text naturalization tasks, ensuring names, numbers, dates, and citations remain unchanged.

Loading preview...

Overview

dolas-zh-finetuned is a 4 billion parameter Qwen3-based model, specifically fine-tuned by ChisatoY for naturalizing AI-generated Chinese text. It aims to transform text with a distinct AI style into more natural human Chinese, meticulously preserving all factual content, tone, names, numbers, dates, units, and citations. The model was developed using a reverse distillation approach, building upon unsloth/Qwen3-4B-Instruct-2507 with Chinese QLoRA SFT.

Key Capabilities

  • AI Text Naturalization: Rewrites AI-generated Chinese text to sound more human and natural.
  • Information Preservation: Ensures complete retention of original meaning, tone, and all critical information, including names, titles, organizations, numbers, statistics, dates, units, and citations.
  • Chinese-focused: Designed exclusively for Chinese text rewriting tasks; it does not perform translation.

Training Details

The model was trained on 11,921 samples derived from a cleaned pool of 129,278 human-written Chinese texts from datasets like CASIA-LM/ChineseWebText2.0 and Hello-SimpleAI/HC3-Chinese. The training involved synthesizing AI-style Chinese from human originals and then training the model to revert them to the human original. This process included rigorous two-level quality control to ensure factual integrity and semantic consistency.

Good For

  • Improving Readability: Enhancing the naturalness and flow of AI-generated Chinese content.
  • Content Refinement: Polishing texts produced by large language models to remove robotic or overly formal tones.
  • Specific Use Cases: Ideal for applications where AI-generated Chinese needs to blend seamlessly with human-written content, such as in publishing, customer service, or creative writing assistance.

Finetunes
1 models