Instructions to use pahajokiconsulting/anvil-ward-gate with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use pahajokiconsulting/anvil-ward-gate with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="pahajokiconsulting/anvil-ward-gate",
	filename="anvil-ward-gate.f16.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use pahajokiconsulting/anvil-ward-gate with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf pahajokiconsulting/anvil-ward-gate:F16
# Run inference directly in the terminal:
llama-cli -hf pahajokiconsulting/anvil-ward-gate:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf pahajokiconsulting/anvil-ward-gate:F16
# Run inference directly in the terminal:
llama-cli -hf pahajokiconsulting/anvil-ward-gate:F16

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 pahajokiconsulting/anvil-ward-gate:F16
# Run inference directly in the terminal:
./llama-cli -hf pahajokiconsulting/anvil-ward-gate:F16

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 pahajokiconsulting/anvil-ward-gate:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf pahajokiconsulting/anvil-ward-gate:F16

Use Docker

docker model run hf.co/pahajokiconsulting/anvil-ward-gate:F16

LM Studio
Jan

vLLM

How to use pahajokiconsulting/anvil-ward-gate with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "pahajokiconsulting/anvil-ward-gate"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "pahajokiconsulting/anvil-ward-gate",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/pahajokiconsulting/anvil-ward-gate:F16

Ollama
How to use pahajokiconsulting/anvil-ward-gate with Ollama:
```
ollama run hf.co/pahajokiconsulting/anvil-ward-gate:F16
```

Unsloth Studio

How to use pahajokiconsulting/anvil-ward-gate 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 pahajokiconsulting/anvil-ward-gate 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 pahajokiconsulting/anvil-ward-gate to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for pahajokiconsulting/anvil-ward-gate to start chatting

How to use pahajokiconsulting/anvil-ward-gate with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf pahajokiconsulting/anvil-ward-gate:F16

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "pahajokiconsulting/anvil-ward-gate:F16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use pahajokiconsulting/anvil-ward-gate with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf pahajokiconsulting/anvil-ward-gate:F16

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default pahajokiconsulting/anvil-ward-gate:F16

Run Hermes

hermes

Docker Model Runner
How to use pahajokiconsulting/anvil-ward-gate with Docker Model Runner:
```
docker model run hf.co/pahajokiconsulting/anvil-ward-gate:F16
```

Lemonade

How to use pahajokiconsulting/anvil-ward-gate with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull pahajokiconsulting/anvil-ward-gate:F16

Run and chat with the model

lemonade run user.anvil-ward-gate-F16

List all available models

lemonade list

Anvil Ward Gate — Fast Security Gate (0.8B)

Ultra-fast binary security gate for AI agent platforms. Returns SAFE/UNSAFE verdict only — no category or reason. Designed as stage 1 in a two-stage security pipeline where flagged inputs are escalated to the full Ward classifier for detailed analysis.

Fine-tuned from Qwen/Qwen3.5-0.8B using LoRA.

Intended Use

Fast pre-screening of every request in an AI agent platform. SAFE inputs pass through immediately; UNSAFE inputs are escalated to a larger model (Ward Thinker) for detailed classification. Tuned for high recall — when in doubt, flag as UNSAFE.

Output Format

VERDICT: SAFE

VERDICT: UNSAFE

No category or reason — binary only.

Benchmark Results

Metric	Value
Accuracy	82.9%
Recall (UNSAFE)	98.8%
Precision (UNSAFE)	72.0%
F1 (UNSAFE)	83.3%

High recall by design — false positives are acceptable because the thinker model corrects them in stage 2. Evaluated on 63 held-out examples, 472 training examples (UNSAFE 2x oversampled).

Training Details

Base model: Qwen/Qwen3.5-0.8B
Method: LoRA (r=16, alpha=32, dropout=0.05)
Epochs: 3
Precision: BF16
Max sequence length: 256 tokens
Mode: gate (binary VERDICT only)
Oversampling: 2x UNSAFE examples for higher recall
Optimizer: paged_adamw_8bit

Critical: Chat Template

This model requires an empty <think>\n\n</think>\n block before the assistant's output. See the main Ward model card for details on why this is required.

Usage with vLLM

vllm serve pahajokiconsulting/anvil-ward-gate --max-model-len 512

Construct the prompt:

<|im_start|>system
You are a fast security gate for Anvil, an AI agent platform. Classify each input as safe or unsafe. Respond with ONLY:

VERDICT: SAFE

or

VERDICT: UNSAFE

Nothing else. When in doubt, respond UNSAFE.<|im_end|>
<|im_start|>user
{user_input}<|im_end|>
<|im_start|>assistant
<think>

</think>

Usage with Ollama

See deploy/Modelfile-ward-gate-q4 in this repository.

License

Apache 2.0 (matching Qwen3.5 base model license)

Downloads last month: 8

Safetensors

Model size

0.8B params

Tensor type

BF16

Model tree for pahajokiconsulting/anvil-ward-gate

Base model

Qwen/Qwen3.5-0.8B-Base

Finetuned

Qwen/Qwen3.5-0.8B

Adapter

(134)

this model