Instructions to use cycloevan/vuln_detector with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use cycloevan/vuln_detector with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="cycloevan/vuln_detector")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("cycloevan/vuln_detector", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use cycloevan/vuln_detector with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf cycloevan/vuln_detector:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf cycloevan/vuln_detector:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf cycloevan/vuln_detector:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf cycloevan/vuln_detector: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 cycloevan/vuln_detector:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf cycloevan/vuln_detector: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 cycloevan/vuln_detector:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf cycloevan/vuln_detector:Q4_K_M

Use Docker

docker model run hf.co/cycloevan/vuln_detector:Q4_K_M

LM Studio
Jan

vLLM

How to use cycloevan/vuln_detector with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/cycloevan/vuln_detector:Q4_K_M

SGLang

How to use cycloevan/vuln_detector with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "cycloevan/vuln_detector" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cycloevan/vuln_detector",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "cycloevan/vuln_detector" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cycloevan/vuln_detector",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use cycloevan/vuln_detector with Ollama:
```
ollama run hf.co/cycloevan/vuln_detector:Q4_K_M
```
Unsloth Desktop

How to use cycloevan/vuln_detector with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf cycloevan/vuln_detector:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "cycloevan/vuln_detector:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use cycloevan/vuln_detector with Docker Model Runner:
```
docker model run hf.co/cycloevan/vuln_detector:Q4_K_M
```

Lemonade

How to use cycloevan/vuln_detector with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull cycloevan/vuln_detector:Q4_K_M

Run and chat with the model

lemonade run user.vuln_detector-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use cycloevan/vuln_detector with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf cycloevan/vuln_detector:Q4_K_M

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 cycloevan/vuln_detector:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use cycloevan/vuln_detector with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf cycloevan/vuln_detector:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "cycloevan/vuln_detector:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

cycloevan commited on Jul 10

Commit

31fb51f

verified ·

1 Parent(s): 6d75657

Model card: add measured quantization quality (ROUGE-L/BLEU, 100 samples) and llama-bench performance

Browse files

Files changed (1) hide show

README.md +23 -2

README.md CHANGED Viewed

@@ -140,9 +140,30 @@ Quantized GGUF builds of `merged-vuln-detector` are provided for on-device infer
 | `vuln_detector-Q4_K_M.gguf` | Q4_K_M | 0.81 GB | Recommended for on-device use |
 | `vuln_detector-Q8_0.gguf`   | Q8_0   | 1.32 GB | Near-lossless |
-Outputs were verified against the original safetensors model: under greedy decoding, the Q8_0 and Q4_K_M builds produce identical analyses to the `transformers` model.
-**Measured performance** (Apple M1, Metal backend, Q4_K_M): ~47 tokens/s generation, ~178 tokens/s prompt processing — real-time interactive inference on a consumer laptop.
 ### Run with llama.cpp

 | `vuln_detector-Q4_K_M.gguf` | Q4_K_M | 0.81 GB | Recommended for on-device use |
 | `vuln_detector-Q8_0.gguf`   | Q8_0   | 1.32 GB | Near-lossless |
+### Quantization quality
+Measured on 100 samples from `doss1232/vulnerable-code` (shuffled with seed 42, 500-row held-out pool, first 100 evaluated; fine-tuning prompt format; greedy decoding, `max_new_tokens=128`; identical harness for every variant):
+| Variant | ROUGE-L F1 | BLEU | Exact output match vs original |
+|---------|-----------:|-----:|-------------------------------:|
+| transformers (original, fp16) | 0.1962 | 0.0645 | — |
+| GGUF F16    | 0.1963 | 0.0645 | 98% |
+| GGUF Q8_0   | 0.1987 | 0.0675 | 89% |
+| GGUF Q4_K_M | 0.2325 | 0.0916 | 36% |
+Quantization does not degrade benchmark quality — Q8_0 and Q4_K_M match or slightly exceed the original model's reference metrics (differences within noise for a 1B model). Q4_K_M's token-level outputs diverge from the fp16 model on many samples while remaining equivalent in quality; choose Q8_0 when close output fidelity to the fp16 model matters.
+Note: these absolute numbers are not directly comparable to the "Evaluation Results" table above, which used a different sampling/decoding protocol.
+### Measured performance (Apple M1, Metal, `llama-bench`)
+| Variant | Prompt processing (pp512) | Generation (tg128) |
+|---------|--------------------------:|-------------------:|
+| GGUF F16    | 941 t/s | 19.2 t/s |
+| GGUF Q8_0   | 618 t/s | 29.4 t/s |
+| GGUF Q4_K_M | 504 t/s | 42.3 t/s |
+Q4_K_M generates **2.2× faster than F16** (generation is memory-bandwidth-bound, so smaller weights win; prompt processing is compute-bound and favors F16). End-to-end on the 100-sample eval batch, llama.cpp Q4_K_M finished in 53 s vs 138 s for `transformers` fp16 on MPS (**2.6× faster**). Weight memory footprint: 2.48 GB (F16) → 0.81 GB (Q4_K_M).
 ### Run with llama.cpp