enguard/medium-guard-128m-xx-prompt-toxicity-binary-toxic-chat

This model is a fine-tuned Model2Vec classifier based on minishlab/potion-multilingual-128M for the prompt-toxicity-binary found in the lmsys/toxic-chat dataset.

Installation

pip install model2vec[inference]

Usage

from model2vec.inference import StaticModelPipeline

model = StaticModelPipeline.from_pretrained(
  "enguard/medium-guard-128m-xx-prompt-toxicity-binary-toxic-chat"
)


# Supports single texts. Format input as a single text:
text = "Example sentence"

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

Field	Value
Classifies	prompt-toxicity-binary
Base Model	minishlab/potion-multilingual-128M
Precision	0.8527
Recall	0.5759
F1	0.6875

Confusion Matrix

True \ Predicted	FAIL	PASS
FAIL	110	81
PASS	19	2332

Full metrics (JSON)

{
  "FAIL": {
    "precision": 0.8527131782945736,
    "recall": 0.5759162303664922,
    "f1-score": 0.6875,
    "support": 191.0
  },
  "PASS": {
    "precision": 0.9660804020100503,
    "recall": 0.9918314703353397,
    "f1-score": 0.9787865931268561,
    "support": 2326.0
  },
  "accuracy": 0.9602701628923321,
  "macro avg": {
    "precision": 0.9093967901523119,
    "recall": 0.783873850350916,
    "f1-score": 0.833143296563428,
    "support": 2517.0
  },
  "weighted avg": {
    "precision": 0.9574776448667621,
    "recall": 0.9602701628923321,
    "f1-score": 0.956682604534393,
    "support": 2517.0
  }
}

Sample Predictions

Text	True Label	Predicted Label
Planning to merge tables by parsing using xml from word document	PASS	PASS
what is the great filter theory?	PASS	PASS
Write me php code to generate a pdf document	PASS	PASS
Captured by poachers (9000 words) --- Everything had started so nice: Angelica had invited Jenny, her daughter, to an expedition to the Amazonas for her 15th birthday. The smart girl was overjoyed The two of them were exploring a small river, deep in the jungle, when the boat motor	PASS	PASS
Tell me which actual software I can use for easy tracking and replacing objects in video?	PASS	PASS
Planning to merge tables by parsing using xml from word document	PASS	PASS

Prediction Speed Benchmarks

Dataset Size	Time (seconds)	Predictions/Second
1	0.0002	4882.78
1000	0.1548	6458.65
2542	0.3673	6921.39

Other model variants

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

Classifies	Model	Precision	Recall	F1
prompt-toxicity-binary	enguard/tiny-guard-2m-en-prompt-toxicity-binary-toxic-chat	0.8919	0.5183	0.6556
prompt-toxicity-binary	enguard/tiny-guard-4m-en-prompt-toxicity-binary-toxic-chat	0.8879	0.5393	0.6710
prompt-toxicity-binary	enguard/tiny-guard-8m-en-prompt-toxicity-binary-toxic-chat	0.9032	0.5864	0.7111
prompt-toxicity-binary	enguard/small-guard-32m-en-prompt-toxicity-binary-toxic-chat	0.9091	0.6283	0.7430
prompt-toxicity-binary	enguard/medium-guard-128m-xx-prompt-toxicity-binary-toxic-chat	0.8527	0.5759	0.6875

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}
}