marcoorasch/llama-3.2-3B-instruct-hatespeech-offensive-classification

Disclaimer:

This model is provided as-is and may be overtrained, leading to potential biases or inaccuracies. Use it at your own discretion and only for personal or non-critical applications. The author is not responsible for any outcomes, damages, or misuse. Always verify results before relying on them.

Model Details

Developed by: Marco Orasch, Andre Salzer

Model Type: Transformer-based Language Model

Language(s): English

Finetuned from model: llama-3.2-3B

Training Epochs: 1

Batch Size: 1

Learning Rate: 2e-4

Optimizer: AdamW

Model Description

This model is fine-tuned for the task of classifying hate speech and offensive language in English text. It leverages the llama-3.2-3B model architecture, refined through a supervised learning approach using a dataset specifically curated for identifying harmful and offensive content. The model can distinguish between hate speech, offensive language, and neutral content, making it suitable for content moderation and safety applications.

Training Procedure

Run Details: https://api.wandb.ai/links/marcoor-universit-t-klagenfurt/oxoxik67

Base Model: llama-3.2-3B

Fine-tuning Framework: PyTorch with Hugging Face Transformers

Tutorial Used: https://www.datacamp.com/tutorial/fine-tuning-llama-3-1

Jupyter Notebook: https://huggingface.co/marcoorasch/llama-3.2-3B-instruct-hatespeech-offensive-classification/blob/main/Hatespeech_Offensive_Classification_llama3.2-3B-instruct.ipynb

Intended Use

Content moderation on social media platforms

Automated filtering of offensive content in forums and chat applications

Assisting research in online safety and digital well-being