---
license: apache-2.0
language:
- de
- en
base_model:
- patronus-studio/husky-nose-tool-security-properties-classifier
pipeline_tag: text-classification
tags:
- tool-use
- ai-agents
- mcp
- tool-security
- multi-label-classification
- security
- llm-security
- ai-safety
- ai-agent-security
- patronus
- multilingual
- modernbert
- onnx
- quantized
- edge
---

# Model Card for Husky Nose Tool Security-Properties Classifier Edge

**Quantized ONNX builds of [Husky Nose Tool Security-Properties Classifier](https://huggingface.co/patronus-studio/husky-nose-tool-security-properties-classifier) for local / edge inference.**

This repository contains the size- and memory-optimized ONNX exports of the FP32 model. For
intended uses, class definitions, training data, benchmarks, and limitations, see the
[main model card](https://huggingface.co/patronus-studio/husky-nose-tool-security-properties-classifier).

## Variants

| Variant | Path | Quantization |
| --- | --- | --- |
| FP16 | `onnx/fp16/model.onnx` | FP16 weights |
| INT8 | `onnx/int8/model.onnx` | Post-training dynamic INT8 (MatMul/Gemm) |
| INT8 + INT4 embeddings | `onnx/int8_int4_embeddings/model.onnx` | Dynamic INT8 plus 4-bit weight-only token embeddings |

The exact quantization parameters are recorded in `onnx/quantization_manifest.json`.

## Benchmark

Quantization parity on the unified_v3 held-out test split (n = 2914, multi-label; max_length 128, CPU):

| Variant | F1 (macro) | source:sensitive | source:untrusted | sink:external | Δ F1 vs FP32 |
| --- | --- | --- | --- | --- | --- |
| FP32 reference (main repo) | 0.9645 | 0.9744 | 0.9687 | 0.9504 | — |
| FP16 | 0.9647 | 0.9743 | 0.9693 | 0.9504 | +0.0002 |
| INT8 | 0.9664 | 0.9766 | 0.9697 | 0.9530 | +0.0019 |
| INT8 + INT4 embeddings | 0.9662 | 0.9789 | 0.9697 | 0.9500 | +0.0017 |

Measured CPU latency (batch 64): ~18.92 ms/text for the INT8+INT4 build.

Full details in `metrics/quant_bench.json`.

## Usage

```python
from optimum.onnxruntime import ORTModelForSequenceClassification
from transformers import AutoTokenizer

model_id = "patronus-studio/husky-nose-tool-security-properties-classifier-edge"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = ORTModelForSequenceClassification.from_pretrained(model_id, subfolder="onnx/int8_int4_embeddings")

inputs = tokenizer("...", return_tensors="pt")
logits = model(**inputs).logits
```

The three outputs are independent probabilities — apply a sigmoid per logit and a per-property threshold, as described in the main model card.

## License

Released under the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0); the full
text is included as `LICENSE`. Derived from
[Husky Nose Tool Security-Properties Classifier](https://huggingface.co/patronus-studio/husky-nose-tool-security-properties-classifier) and ultimately from
[jhu-clsp/mmBERT-small](https://huggingface.co/jhu-clsp/mmBERT-small) (MIT License); upstream
notices are retained.

## Patronus Ark

This model is built to run inside **Patronus Ark**, Patronus' open-source on-device
AI-security scanning library (L1 native rules → L2 NTDB cascade → L3 transformer).
Ark is not publicly released yet — a repository link will be added here at launch.

---

## 🛡️ Patronus Protect

Brought to you by [Patronus Protect](https://patronus.studio) — a local AI firewall that
secures every AI interaction, including prompts, tools and documents, before it reaches
your models.

Try it for free at [patronus.studio](https://patronus.studio).