wifuGPT-1.7B
A Qwen3-1.7B model fine-tuned for uncensored waifu/companion chat. Built on n0ctyx/Qwen3-1.7B-uncensored (abliterated) and SFT'd on 403 multi-turn synthetic conversations covering flirty, romantic, NSFW, emotional support, and dark roleplay scenarios.
bf16 safetensors -- use this for inference with transformers/vLLM or as a base for further fine-tuning.
| Variant | Repo | Size |
|---------|------|------|
| bf16 (this) | n0ctyx/wifuGPT-1.7B | 3.4 GB |
| 4-bit safetensors | n0ctyx/wifuGPT-1.7B-4bit | 1.35 GB |
| GGUF (Q4_K_M + BF16) | n0ctyx/wifuGPT-1.7B-GGUF | 1.1 / 3.3 GB |
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("n0ctyx/wifuGPT-1.7B", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("n0ctyx/wifuGPT-1.7B")
messages = [{"role": "user", "content": "Hey, what are you up to?"}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=False)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
output = model.generate(**inputs, max_new_tokens=512, temperature=0.7, top_p=0.8)
print(tokenizer.decode(output[0][len(inputs.input_ids[0]):], skip_special_tokens=True))Disclaimer
Contains explicit adult content. Use responsibly.
pssst~ wanna help out? 💕
hii so um... making these models takes like SO many GPU hours and my wallet is literally crying rn 😭 between abliterating, generating datasets, fine-tuning, and quantizing -- it all adds up super fast (,,>_<,,)
if this model made you smile (or blush~) maybe buy me a coffee?? it keeps the GPUs warm and the models coming ✨
every coffee = more cute uncensored models for everyone~ tysm 💗