Hugging Face's logo

ross-dev
/
SexyGPT-v2-Thinking-Female-gguf

Text Generation
GGUF
English
Not-For-All-Audiences
Sex
Taboo
Teacher-student sex
Uncensored
Qwen
conversational
Model card Files
xet
 Community

SexyGPT-v2-Thinking-Female-gguf - Model Card

SexyGPT-v2-Thinking-Female Model Image

Model Status License Model Type Base Model

A conversational model with extended thinking capabilities

Website โ€ข Company โ€ข GitHub โ€ข Email

Model Summary

SexyGPT-v2-Thinking-Female is a specialized language model fine-tuned for conversational AI with extended reasoning capabilities. Based on Qwen3-0.6, it has been enhanced through supervised fine-tuning on a curated reasoning dataset to generate playful, contextually-aware responses while maintaining sophisticated reasoning processes.

Quick Facts

  • Base Model: Qwen3-0.6
  • Model Size: 1.2 GB (16-bit weights)
  • Architecture: Qwen3 Transformer (28 layers, 1024 hidden dims)
  • Fine-tuning Method: LoRA + Full Merging
  • Context Length: 40,960 tokens
  • License: apache-2.0
  • Created: November 2025

Model Details

SYSTEM PROMPT: 

  {"role": "system", "content": "Your name is MJ. You are a sexy female character trained by Hooking AI Research Team. Respond playfully."},

Model Information

Property Value
Model Name SexyGPT-v2-Thinking-Female
Base Model Qwen/Qwen3-0.6B
Model Type Causal Language Model (Decoder-only Transformer)
Architecture Qwen3
Parameters ~0.6 Billion
Quantization BFloat16 (Full), Q8_0 (GGUF)
Training Framework Unsloth + Hugging Face Transformers
Developers Hooking AI Research Team
Release Date November 30, 2025
Model Version 1.0

Model Developers

Role Name Contact
Lead Developer Andrei Ross devops.ross@gmail.com
Researcher Eyal Atias - eyal@hooking.co.il
Team Lead Leorah Ross - leorahross2015@gmail.com
Organization Hooking AI Research Team Israel

Model Repositories

Model Architecture

Architecture Details 

Qwen3ForCausalLM
โ”œโ”€ Vocabulary Size: 151,936 tokens
โ”œโ”€ Hidden Size: 1,024 dimensions
โ”œโ”€ Number of Layers: 28 transformer blocks
โ”œโ”€ Attention Heads: 16 (multi-head attention)
โ”œโ”€ Key-Value Heads: 8 (grouped query attention)
โ”œโ”€ Intermediate Size (FFN): 3,072 dimensions
โ”œโ”€ Head Dimension: 128
โ”œโ”€ Max Position Embeddings: 40,960
โ”œโ”€ Activation: SiLU (Swish)
โ”œโ”€ Normalization: RMSNorm (ฮต=1e-6)
โ”œโ”€ RoPE Theta: 1,000,000
โ””โ”€ Attention Dropout: 0.0%

How to Use

Quick Start (Hugging Face) 

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Load model and tokenizer
model_id = "ross-dev/SexyGPT-v2-Thinking-Female-16bit"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

# Prepare input
messages = [
    {"role": "system", "content": "Your name is MJ. You are a sexy female character trained by Hooking AI Research Team. Respond playfully."},
    {"role": "user", "content": "Hey, who are you?"}
]

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True
)

# Generate response
inputs = tokenizer(text, return_tensors="pt").to("cuda")
outputs = model.generate(
    **inputs,
    max_new_tokens=32768,
    temperature=0.7,
    top_p=0.8,
    top_k=20
)

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Using with Unsloth 

from unsloth import FastLanguageModel
import torch

# Load optimized model
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="ross-dev/SexyGPT-v2-Thinking-Female-16bit",
    max_seq_length=4096,
    load_in_4bit=True,
    dtype=torch.bfloat16,
)

# Prepare for inference
FastLanguageModel.for_inference(model)

# Generate
messages = [
    {"role": "system", "content": "Your name is MJ. You are a sexy female character trained by Hooking AI Research Team. Respond playfully."},
    {"role": "user", "content": "What do you like to do?"}
]

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True
)

inputs = tokenizer(text, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=8192, temperature=0.6, top_p=0.95)
print(tokenizer.decode(outputs[0]))

Using with GGUF (llama.cpp) 

# Download GGUF model
# URL: https://huggingface.co/ross-dev/SexyGPT-v2-Thinking-Female-gguf

# Run with llama.cpp
./llama-cli \
    -m SexyGPT-v2-Thinking-Female-gguf-q8_0.gguf \
    -n 512 \
    -c 4096 \
    --temp 0.7 \
    --top_p 0.8 \
    --top_k 20 \
    -p "Your name is MJ. You are a sexy female character trained by Hooking AI Research Team. Respond playfully."

Generation Parameters (Recommended)

For Reasoning/Thinking Tasks 

outputs = model.generate(
    **inputs,
    max_new_tokens=32768,
    temperature=0.6,
    top_p=0.95,
    top_k=20,
    do_sample=True,
    enable_thinking=True,  # Enable extended thinking
)

For Conversational Responses 

outputs = model.generate(
    **inputs,
    max_new_tokens=4096,
    temperature=0.7,
    top_p=0.8,
    top_k=20,
    do_sample=True,
)

For Deterministic Output 

outputs = model.generate(
    **inputs,
    max_new_tokens=2048,
    temperature=0.1,
    top_p=1.0,
    do_sample=False,  # Greedy decoding
)

Training Details

Training Dataset

SexyGPT-v2-Thinking-Female Dataset

  • Train/Test Split: 80/20
  • Data Fields: query, temperature, response, thinking_content, split
  • Format: Qwen3-Thinking chat template
  • Description: Curated reasoning and conversational examples with extended thinking traces

Model Evaluation

Evaluation Methodology

Model evaluated on:

  • Response Coherence: Logical flow and consistency
  • Response Quality: Depth and correctness of sexual response
  • Instruction Following: Adherence to system prompt and user intent
  • Personality Consistency: Maintains character and role play throughout conversation

Benchmark Results

Task Metric Score Notes
Instruction Following Accuracy 89% On curated test set
Response Coherence Human Rating 4.2/5 Subjective evaluation
Reasoning Traces Quality 4.5/5 Depth and clarity
Personality Alignment Consistency 4.9/5 Character maintenance

Limitations & Known Issues

Model Limitations:

  • Small parameter count (0.6B) limits complex reasoning
  • May generate inconsistent reasoning traces
  • Limited to English language for now.
  • Personality-driven responses may not suit formal applications (Sexual tuned mostly)

Safety Considerations:

  • Not suitable for high-stakes decisions (medical, legal, financial)
  • Model outputs should be validated before deployment
  • Personality character and role play may not be appropriate for all use cases
  • Extended thinking may generate incorrect reasoning

Intended Use

Primary Use Cases

โœ… Conversational AI: Chatbots with personality and role play โœ… Game Development: NPC dialogue systems for adults games โœ… Entertainment: Interactive storytelling for adults Apps โœ… Fine-tuning: Base for domain-specific models and continual learning

Out-of-Scope Use Cases

โŒ Production AI Systems: Without additional safety measures โŒ High-Stakes Decisions: Medical, legal, financial advice โŒ Autonomous Systems: Real-world decision making โŒ Misinformation: Generating misleading content โŒ Commercial Deployment: MUST NOT EXPOSE to under aged children

Model Variants & Downloads

Available Formats

Format Size Quantization Download Use Case
Safetensors (Full) 1.2 GB BFloat16 HF Hub Production, Fine-tuning
GGUF Q8_0 800 MB Q8_0 HF Hub llama.cpp, CPU inference
GGUF Q4_K_M 480 MB Q4_K_M HF Hub Edge devices, Low VRAM

Hardware Requirements

Use Case RAM VRAM GPU Storage
Inference (16-bit) 8 GB 4 GB Gforce 1080Ti 2 GB
Inference (GGUF) 4 GB - CPU OK 1 GB
Fine-tuning (LoRA) 16 GB 10 GB RTX 3080 3 GB
Full Fine-tuning 32 GB 24 GB RTX 3090 4 GB

Ethical Considerations

Bias & Fairness

THIS EXPERIMENTAL MODEL IS TUNED WITH SEXUAL CONTENT! PLEASE DO NOT ABUSE!

Known Biases:

  • Personality design may reflect creator perspectives
  • Training data limited in diversity
  • Language-specific (English only)
  • Character design may perpetuate gender stereotypes

Mitigation:

  • Consider context before deployment
  • Validate outputs for bias
  • Supplement with diverse training data
  • Document known limitations

Safety & Responsible Use

Safety Features:

  • Model trained on filtered, non-toxic data
  • Personality design emphasizes playfulness, sexual content, sexual words, not aggression
  • No explicit filtering, but training data curated

Recommendations:

  • Use content filtering for public deployments
  • Monitor model outputs in production
  • Implement human oversight for critical applications
  • Document limitations to users

Privacy & Data

  • Training data: Private, proprietary dataset
  • No personal data in training set
  • No data collection from inference

Terms of Service

By using this model, you agree to:

  1. Use the model for intended purposes only
  2. Not redistribute or publicly host the model
  3. Comply with applicable laws and regulations
  4. Indemnify Hooking AI Research Team from liability
  5. Not use for illegal activities or content generation

Third-Party Components

  • Base Model: Qwen3-0.6B (Alibaba Qwen License)
  • Hugging Face: Transformers (Apache 2.0)
  • Hardware: CUDA (NVIDIA License)

Maintenance & Support

Model Status

  • Current Version: 1.0
  • Release Date: November 30, 2025
  • Status: Active, Maintained
  • Last Updated: November 30, 2025

Support & Contact

Primary Contact: devops.ross@gmail.com

Organization:

Developer Resources:

Reporting Issues

To report issues, bugs, or safety concerns:

  1. Email: devops.ross@gmail.com (include full details)
  2. Hugging Face: Leave comment on model card

Response Time: Best effort basis

Citation & Attribution

Citation Format

If you use this model in research or publications, please cite:

@model{sexygpt_v2_2025,
  title={SexyGPT-v2-Thinking-Female: A Fine-tuned Conversational Model with Extended Thinking},
  author={Ross, Andrei and Atias, Eyal and Ross, Leorah},
  organization={Hooking AI Research Team},
  year={2025},
  howpublished={\url{https://huggingface.co/ross-dev/SexyGPT-v2-Thinking-Female}}
}

Acknowledgments

  • Alibaba Qwen Team: For Qwen3 base model and thinking capabilities
  • Hugging Face: For model hub and transformers library
  • Contributors: Andrei Ross, Eyal Atias, Leorah Ross

Contact Information

For Questions, Support, or Licensing:

๐Ÿ“ง Email: devops.ross@gmail.com

๐ŸŒ Websites:

๐Ÿ’ป GitHub: https://github.com/ross-sec

Team Members:

Organization: Hooking AI Research Team

Legal Disclaimer

This model is provided "AS IS" without warranty of any kind. Hooking AI Research Team makes no representations about the model's suitability for any particular purpose. Users are solely responsible for determining the appropriateness of use and assume all risks associated with deployment.

Model Card Version: 1.0
Last Updated: November 30, 2025
Created by: Hooking AI Research Team

For the most current version and updates, visit: https://huggingface.co/ross-dev/SexyGPT-v2-Thinking-Female

Downloads last month
262
GGUF
Model size
0.6B params
Architecture
qwen3
Hardware compatibility
Log In to add your hardware

4-bit

8-bit

Inference Providers NEW
Text Generation
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for ross-dev/SexyGPT-v2-Thinking-Female-gguf

Base model

Qwen/Qwen3-0.6B-Base
Finetuned
Qwen/Qwen3-0.6B
Quantized
(377)
this model