Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Tiny-Toxic-Detector: A compact transformer-based model for toxic content detection

Michiel Kamphuis

TL;DR

The paper tackles the need for efficient toxic content detectors suitable for resource-constrained deployment by introducing Tiny-toxic-detector, a compact transformer with 2.1 million parameters. It relies on a task-specific, no-pretraining training regime using labeled data, achieving strong benchmarking performance (approximately 91% on ToxiGen and ~87% on Jigsaw) despite its small size. The approach emphasizes environmental and computational efficiency, reporting a modest carbon footprint and fast CPU inference comparable to much larger models. The work demonstrates that careful architectural design and data-focused training can deliver scalable, sustainable toxicity moderation, with future work proposing quantization, distillation, multilingual adaptation, and extended context handling.

Abstract

This paper presents Tiny-toxic-detector, a compact transformer-based model designed for toxic content detection. Despite having only 2.1 million parameters, Tiny-toxic-detector achieves competitive performance on benchmark datasets, with 90.97% accuracy on ToxiGen and 86.98% accuracy on the Jigsaw dataset, rivaling models over 50 times its size. This efficiency enables deployment in resource-constrained environments, addressing the need for effective content moderation tools that balance performance with computational efficiency. The model architecture features 4 transformer encoder layers, each with 2 attention heads, an embedding dimension of 64, and a feedforward dimension of 128. Trained on both public and private datasets, Tiny-toxic-detector demonstrates the potential of efficient, task-specific models for addressing online toxicity. The paper covers the model architecture, training process, performance benchmarks, and limitations, underscoring its suitability for applications such as social media monitoring and content moderation. By achieving results comparable to much larger models while significantly reducing computational demands, Tiny-toxic-detector represents progress toward more sustainable and scalable AI-driven content moderation solutions.

Tiny-Toxic-Detector: A compact transformer-based model for toxic content detection

TL;DR

The paper tackles the need for efficient toxic content detectors suitable for resource-constrained deployment by introducing Tiny-toxic-detector, a compact transformer with 2.1 million parameters. It relies on a task-specific, no-pretraining training regime using labeled data, achieving strong benchmarking performance (approximately 91% on ToxiGen and ~87% on Jigsaw) despite its small size. The approach emphasizes environmental and computational efficiency, reporting a modest carbon footprint and fast CPU inference comparable to much larger models. The work demonstrates that careful architectural design and data-focused training can deliver scalable, sustainable toxicity moderation, with future work proposing quantization, distillation, multilingual adaptation, and extended context handling.

Abstract

This paper presents Tiny-toxic-detector, a compact transformer-based model designed for toxic content detection. Despite having only 2.1 million parameters, Tiny-toxic-detector achieves competitive performance on benchmark datasets, with 90.97% accuracy on ToxiGen and 86.98% accuracy on the Jigsaw dataset, rivaling models over 50 times its size. This efficiency enables deployment in resource-constrained environments, addressing the need for effective content moderation tools that balance performance with computational efficiency. The model architecture features 4 transformer encoder layers, each with 2 attention heads, an embedding dimension of 64, and a feedforward dimension of 128. Trained on both public and private datasets, Tiny-toxic-detector demonstrates the potential of efficient, task-specific models for addressing online toxicity. The paper covers the model architecture, training process, performance benchmarks, and limitations, underscoring its suitability for applications such as social media monitoring and content moderation. By achieving results comparable to much larger models while significantly reducing computational demands, Tiny-toxic-detector represents progress toward more sustainable and scalable AI-driven content moderation solutions.
Paper Structure (17 sections, 1 figure, 3 tables)

This paper contains 17 sections, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Model Architecture and Training
  4. Overview
  5. Components
  6. Training Data
  7. Operational Details
  8. CO2 Emission Related to Experiments
  9. Usage and Limitations
  10. Intended Usage
  11. Potential Use-Cases
  12. Limitations
  13. Training Data
  14. Maximum Context Length
  15. Language Ambiguity
  16. ...and 2 more sections

Figures (1)

  • Figure 1: Average Benchmark Score vs. Model Size (Parameters)