vurtnesaerdna/Qwen3.8-27B-Uncensored-Chinese

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 25, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

vurtnesaerdna/Qwen3.8-27B-Uncensored-Chinese is a 27 billion parameter fine-tuned variant of the Qwen/Qwen3.8-27B model, specifically optimized for generating sexually explicit Chinese adult fiction. This model was trained using LoRA on approximately 4.2k instruction examples derived from Chinese adult fiction, with the LoRA weights merged into the base model. Its primary differentiator is its specialized capability in creative writing for adult content, making it suitable for generating realistic and detailed erotic narratives in Chinese.

Loading preview...

Overview

vurtnesaerdna/Qwen3.8-27B-Uncensored-Chinese is a 27 billion parameter language model, fine-tuned from the Qwen/Qwen3.8-27B base model. Its core purpose is to generate sexually explicit Chinese adult fiction, making it distinct from general-purpose LLMs. The model was developed by vurtnesaerdna through a LoRA (Low-Rank Adaptation) SFT process, utilizing about 4.2k single-turn instruction examples from Chinese adult fiction.

Key Capabilities

  • Specialized Content Generation: Excels at creating detailed and realistic sexually explicit narratives in Chinese.
  • Fine-tuned for Creative Writing: Optimized for long-form prose generation within its specific domain.
  • Instruction-Following: Trained on instruction-response pairs to guide content generation, though instruction following outside creative writing may be weaker than the base model.

Training Details

The model was trained using LoRA SFT (r=16, alpha=32) with loss calculated only on assistant tokens. The training dataset consisted of approximately 4.2k instruction/response pairs, each with a maximum length of 1536 tokens. The training schedule involved 2 epochs with a learning rate of 1e-4 and cosine decay, using an effective batch size of 8.

Usage Considerations

  • 18+ Content: This model is explicitly designed for generating sexually explicit fiction and is intended for adult audiences only.
  • Hardware Requirements: Requires significant resources, with ~50 GB of bf16 weights necessitating multiple GPUs or CPU offload for efficient operation.
  • Sampling Recommendations: Suggested sampling parameters (e.g., temperature 0.9, top_p 0.9, top_k 40, min_p 0.05) are provided to avoid repetitive generations, which can occur with greedy decoding in Qwen models.

Limitations

  • Content is fictional and for adult readers only.
  • Instruction following for tasks outside creative writing may be less robust compared to the base model.
  • May occasionally exhibit repetition during very long generations.

Finetunes
2 models