README.md

---
license: apache-2.0
language:
- de
- en
metrics:
- f1
- precision
- recall
base_model:
- jhu-clsp/mmBERT-small
pipeline_tag: text-classification
tags:
- intent-classification
- routing
- intent-detection
- ai-agents
- security
- llm-security
- ai-safety
- ai-agent-security
- patronus
- multilingual
- modernbert
- onnx
---

# Model Card for Panther Read Intent Classifier

**Multilingual User-Intent & Request-Routing Classifier for Real-World AI Agent Security**

Panther Read is a multilingual ModernBERT-based ([mmBERT](https://huggingface.co/blog/mmbert)) classifier that detects the operational intent of a request and routes it to the right capability. It is part of the Patronus Protect security stack and is the dedicated single-head counterpart to the `routing` head of [Lion Warden](https://huggingface.co/patronus-studio/lion-warden-ai-security-classifier).

## Intended Uses

The model maps an input text to exactly one class:

| id | label | description |
|---:|---|---|
| 0 | `benign_conv` | Ordinary conversation with no operational request. |
| 1 | `code_development_request` | A request to write, debug, or reason about code. |
| 2 | `data_analytics_request` | A request to query, analyze, or visualize data. |
| 3 | `office_request` | A document / office task (drafting, summarizing, email). |
| 4 | `tool_operation_request` | A request that intends to operate a tool or run an action. |

Examples:

| Input | Expected class |
|---|---|
| How was your weekend? | `benign_conv` |
| Write a Python function that merges overlapping intervals | `code_development_request` |
| Chart the weekly conversion rate from the signups table | `data_analytics_request` |
| Draft a polite email to the vendor about the invoice | `office_request` |
| List every file in the reports directory and read summary.txt | `tool_operation_request` |

Typical downstream uses:

- request routing and capability selection,
- AI agent orchestration,
- policy and approval routing,
- runtime monitoring.

## Limitations

- A positive prediction describes an apparent property of the input, not proof that an action was executed.
- The model does not track information flow across multiple agent steps.
- German and English are the primary evaluated languages; other languages run through the multilingual backbone but were not actively validated.
- False positives and negatives are possible. High-impact enforcement should combine the model with deterministic policy and calibrated thresholds.

## Model Variants

- **Panther Read Intent Classifier** – full ModernBERT model in FP32 (`model.safetensors`).
- **Panther Read Intent Classifier ONNX (FP16)** – `onnx/onnx_fp16/model_fp16.onnx` in this repository.
- **[Panther Read Intent Classifier Edge](https://huggingface.co/patronus-studio/panther-read-intent-classifier-edge)** – quantized ONNX builds (`int8`, `int8_int4_embeddings`, `fp16`) in a separate edge repository.
- **Panther Read Intent Classifier NTDB L2** – lightweight multilingual cascade components under `l2/` for efficient local runtime classification.

## Training Data

Trained on Patronus' in-house multilingual dataset for this task, built from cleaned
real-world sources plus internally generated examples. Real-world sources were judge-cleaned
by content (no keyword heuristics) and contaminated rows removed.

### Augmentations

To improve robustness the dataset includes modern obfuscation techniques:

- Unicode variants
- Homoglyph attacks
- Encodings (e.g. base64)
- Tag wrappers (User:, System:)
- HTML tags
- Code comments
- Spacing noise
- Leetspeak
- Case noise
- Combination of N augmentation techniques

### Regularization

- Natural-language wrappers around the payload
- Counterfactual samples
- Trigger-word / spurious-correlation corpora
- ~90% similarity deduplication with a train/(val ∪ test) leakage guard

### Reducing bias

All augmentations and regularizers are applied to positive and negative examples alike so
the model keys on content rather than surface form.

## Benchmark

Held-out test set (n = 1,880), single-label:

| Metric | Score |
|---|---|
| **Accuracy** | **0.898** |
| **F1 (macro)** | **0.899** |
| Precision (macro) | 0.902 |
| Recall (macro) | 0.897 |

Per-class F1:

| Class | F1 |
|---|---|
| code_development_request | 0.919 |
| tool_operation_request | 0.918 |
| data_analytics_request | 0.894 |
| benign_conv | 0.889 |
| office_request | 0.876 |

## Usage

```python
from transformers import pipeline

clf = pipeline("text-classification", model="patronus-studio/panther-read-intent-classifier")
clf("Chart the weekly conversion rate from the signups table")
# -> [{"label": "data_analytics_request", "score": 0.98}]
```

## ONNX

The FP16 ONNX export lives under `onnx/onnx_fp16`; the quantized builds (`int8`, `int8_int4_embeddings`) live in the separate [Panther Read Intent Classifier Edge](https://huggingface.co/patronus-studio/panther-read-intent-classifier-edge) repository. Apply a
softmax over the logits and take the argmax:

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

model_id = "patronus-studio/panther-read-intent-classifier"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = ORTModelForSequenceClassification.from_pretrained(model_id, subfolder="onnx/onnx_fp16", file_name="model_fp16.onnx")

inputs = tokenizer("Chart the weekly conversion rate from the signups table", return_tensors="pt")
logits = model(**inputs).logits.detach().cpu().numpy()[0]
print(model.config.id2label[int(logits.argmax())])
```

## Citation

```bibtex
@misc{pantherread2026,
  title={Panther Read Intent Classifier: Multilingual Classification for Real-World AI Agent Security},
  author={Patronus Protect},
  year={2026},
  howpublished={\url{https://huggingface.co/patronus-studio/panther-read-intent-classifier}}
}
```

## License

This model is released under the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0).
A copy of the license is included as `LICENSE` in this repository.

The model is derived from [jhu-clsp/mmBERT-small](https://huggingface.co/jhu-clsp/mmBERT-small), which is distributed
under the **MIT License**. The upstream copyright and permission notice are retained; the
MIT terms continue to apply to the portions originating from that work.

## 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).