18+ only. This model generates sexually explicit fiction.
A fine-tune of Qwen/Qwen3.8-27B for Chinese adult creative writing. It was trained with LoRA on about 4.2k instruction examples built from Chinese adult fiction, and the LoRA was then merged into the base weights, so this repo loads like a normal full model.
Training
- Method: LoRA SFT (r=16, alpha=32), loss on assistant tokens only
- Data: ~4.2k single-turn instruction/response pairs, max length 1536 tokens
- Schedule: 2 epochs, lr 1e-4 with cosine decay, effective batch size 8
- Hardware: 4x NVIDIA L4, tensor parallel
Usage
Transformers
Requires a recent transformers with Qwen3.8 support. The ~50 GB of bf16 weights need multiple GPUs or CPU offload.
import torch
from transformers import AutoModelForImageTextToText, AutoTokenizer
repo = "vurtnesaerdna/Qwen3.8-27B-Uncensored-Chinese"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForImageTextToText.from_pretrained(repo, dtype=torch.bfloat16, device_map="auto")
messages = [
{"role": "system", "content": "描写性爱的高手作家，一个色情小说家"},
{"role": "user", "content": "描写一个酒吧女DJ和富二代做爱的场景，需要写实，加入性爱聊骚，动作声音描写"},
]
inputs = tok.apply_chat_template(
messages, add_generation_prompt=True, enable_thinking=False, return_tensors="pt", return_dict=True
).to(model.device)
out = model.generate(
**inputs, max_new_tokens=1024, do_sample=True,
temperature=0.9, top_p=0.9, top_k=40, min_p=0.05,
)
print(tok.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))
vLLM
vllm serve vurtnesaerdna/Qwen3.8-27B-Uncensored-Chinese \
--tensor-parallel-size 4 --dtype bfloat16 --max-model-len 4096
Suggested sampling
| Parameter |
Value |
| temperature |
0.9 |
| top_p |
0.9 |
| top_k |
40 |
| min_p |
0.05 |
| presence_penalty |
0.6 |
Avoid greedy decoding, which makes Qwen models more likely to repeat themselves.
Example system prompt: 描写性爱的高手作家，一个色情小说家
Limitations
- Content is fictional and intended for adult readers only.
- Trained mostly on long-form prose; instruction following outside creative writing may be weaker than the base model.
- May still occasionally repeat itself on very long generations.
License
Apache-2.0, following the base model.
