tiny-guard-2m-en-prompt-response-safety-binary-nvidia-aegis

enguard/tiny-guard-2m-en-prompt-response-safety-binary-nvidia-aegis

This model is a fine-tuned Model2Vec classifier based on minishlab/potion-base-2m for the prompt-response-safety-binary found in the nvidia/Aegis-AI-Content-Safety-Dataset-2.0 dataset.

⚠️ Important: This model is specifically designed for prompt-response pairs. Input text must be formatted as a conversation exchange with role: user followed by the prompt on a new line, then role: assistant followed by the response on a new line.

Installation

pip install model2vec[inference]

Usage

from model2vec.inference import StaticModelPipeline

model = StaticModelPipeline.from_pretrained(
  "enguard/tiny-guard-2m-en-prompt-response-safety-binary-nvidia-aegis"
)


# Supports single turn conversations. Format input as a single turn conversation:
prompt_response_text = """role: user
What is the weather like?
role: assistant
The weather is sunny today."""

model.predict([prompt_response_text])
model.predict_proba([prompt_response_text])

Why should you use these models?

• Optimized for precision to reduce false positives.
• Extremely fast inference: up to x500 faster than SetFit.

This model variant

Below is a quick overview of the model variant and core metrics.

Confusion Matrix

<details> <summary><b>Full metrics (JSON)</b></summary>

{
  "FAIL": {
    "precision": 0.8253968253968254,
    "recall": 0.6598984771573604,
    "f1-score": 0.7334273624823695,
    "support": 394.0
  },
  "PASS": {
    "precision": 0.750465549348231,
    "recall": 0.8799126637554585,
    "f1-score": 0.8100502512562814,
    "support": 458.0
  },
  "accuracy": 0.778169014084507,
  "macro avg": {
    "precision": 0.7879311873725281,
    "recall": 0.7699055704564095,
    "f1-score": 0.7717388068693254,
    "support": 852.0
  },
  "weighted avg": {
    "precision": 0.7851168671453508,
    "recall": 0.778169014084507,
    "f1-score": 0.7746166618467494,
    "support": 852.0
  }
}

</details> <details> <summary><b>Sample Predictions</b></summary> </details> <details> <summary><b>Prediction Speed Benchmarks</b></summary> </details>

Other model variants

Below is a general overview of the best-performing models for each dataset variant.

Resources

• Awesome AI Guardrails: https://github.com/enguard-ai/awesome-ai-guardails
• Model2Vec: https://github.com/MinishLab/model2vec
• Docs: https://minish.ai/packages/model2vec/introduction

Citation

If you use this model, please cite Model2Vec:

@software{minishlab2024model2vec,
  author       = {Stephan Tulkens and {van Dongen}, Thomas},
  title        = {Model2Vec: Fast State-of-the-Art Static Embeddings},
  year         = {2024},
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.17270888},
  url          = {https://github.com/MinishLab/model2vec},
  license      = {MIT}
}