Instructions to use kusonooyasumi/strix-xss-4b-rl-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use kusonooyasumi/strix-xss-4b-rl-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="kusonooyasumi/strix-xss-4b-rl-GGUF",
	filename="strix-xss-4b-rl-FP16.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use kusonooyasumi/strix-xss-4b-rl-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf kusonooyasumi/strix-xss-4b-rl-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf kusonooyasumi/strix-xss-4b-rl-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf kusonooyasumi/strix-xss-4b-rl-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf kusonooyasumi/strix-xss-4b-rl-GGUF:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf kusonooyasumi/strix-xss-4b-rl-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf kusonooyasumi/strix-xss-4b-rl-GGUF:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf kusonooyasumi/strix-xss-4b-rl-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf kusonooyasumi/strix-xss-4b-rl-GGUF:Q4_K_M

Use Docker

docker model run hf.co/kusonooyasumi/strix-xss-4b-rl-GGUF:Q4_K_M

LM Studio
Jan
Ollama
How to use kusonooyasumi/strix-xss-4b-rl-GGUF with Ollama:
```
ollama run hf.co/kusonooyasumi/strix-xss-4b-rl-GGUF:Q4_K_M
```

Unsloth Studio

How to use kusonooyasumi/strix-xss-4b-rl-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for kusonooyasumi/strix-xss-4b-rl-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for kusonooyasumi/strix-xss-4b-rl-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for kusonooyasumi/strix-xss-4b-rl-GGUF to start chatting

Atomic Chat new
Docker Model Runner
How to use kusonooyasumi/strix-xss-4b-rl-GGUF with Docker Model Runner:
```
docker model run hf.co/kusonooyasumi/strix-xss-4b-rl-GGUF:Q4_K_M
```

Lemonade

How to use kusonooyasumi/strix-xss-4b-rl-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull kusonooyasumi/strix-xss-4b-rl-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.strix-xss-4b-rl-GGUF-Q4_K_M

List all available models

lemonade list

Strix-XSS-Qwen3-4B-RL - GGUF

⚠️ Proof of Concept: This model is an early research prototype. Not production-ready.

Quantized GGUF versions of Strix-XSS-Qwen3-4B-RL, a RL-trained model specialized for detecting Cross-Site Scripting (XSS) vulnerabilities.

Model Information

This is a quantized version of Strix-XSS-Qwen3-4B-RL, optimized for efficient inference on consumer hardware. The model was trained using reinforcement learning on the Prime Intellect platform and achieves 0.79 on the strix-xss evaluation.

Available Quantizations

Quantization	File Size	Use Case	VRAM Required
Q4_K_M	~2.5GB	Recommended for most users	~4GB
Q5_K_M	~3.0GB	Better quality, still efficient	~5GB
Q8_0	~4.5GB	High quality	~6GB
FP16	~8GB	Full quality (for testing)	~10GB

Recommendation: Start with Q4_K_M for the best balance of quality and performance. Upgrade to Q5_K_M or Q8_0 if you have extra VRAM and want better accuracy.

Hardware Requirements

Minimum (Q4_K_M)

RAM: 6GB
VRAM: 4GB (with GPU offloading)
Disk: 3GB free space

Recommended (Q5_K_M)

RAM: 8GB
VRAM: 6GB
Disk: 4GB free space

Optimal (Q8_0)

RAM: 12GB
VRAM: 8GB
Disk: 5GB free space

Performance

Strix-XSS Eval Score: 0.79 (measured on original model)
Quantization Loss: Minimal (<2% degradation from Q4_K_M to FP16)
Inference Speed: ~20-40 tokens/sec on RTX 3060 (12GB)

Training Details

Base Model: Qwen3-4B-Thinking-2507
Training Method: Reinforcement Learning
Dataset: 135 simulated XSS scenarios with Strix tooling
Training Platform: Prime Intellect hosted training beta
Evaluation: strix-xss benchmark on Prime Intellect environment

Special thanks to Prime Intellect for enabling this research!

Limitations

⚠️ This is a proof of concept:

Trained on only 135 examples in simulated environments
Designed for research and demonstration purposes

Original Model

Full precision PyTorch version: kusonooyasumi/strix-xss-qwen3-4b-rl

License

MIT License - See main repository for full details

Acknowledgments

Prime Intellect - Training infrastructure
Qwen Team - Base model
llama.cpp - Quantization tools
Strix Project - Testing framework

Citation

@misc{strix-xss-qwen3-rl-gguf,
  author = {oyasumi},
  title = {Strix-XSS-Qwen3-4B-RL: GGUF Quantized Models},
  year = {2025},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/kusonooyasumi/strix-xss-qwen3-4b-rl-gguf}}
}

Downloads last month: 2

GGUF

Model size

4B params

Architecture

qwen3

Hardware compatibility

4-bit

5-bit

8-bit

View +1 variant

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for kusonooyasumi/strix-xss-4b-rl-GGUF

Base model

Qwen/Qwen3-4B-Thinking-2507

Finetuned

kusonooyasumi/strix-xss-4b-rl

Quantized

(1)

this model