artificialreply/ai-wit-training-data

AI Wit Training Dataset

This dataset contains witty comeback and humor training data for fine-tuning language models.

Dataset Description

• —Total Samples: 132,156
• —Training Samples: 118,940
• —Validation Samples: 13,216
• —Format: Llama 3.1 chat format
• —Source: Reddit jokes and witty responses

Dataset Structure

Each sample contains:

• —messages: List of user/assistant conversation
• —source: Data source (e.g., "reddit_jokes")
• —style: Response style (e.g., "humorous", "witty")

Usage

This dataset is designed for fine-tuning conversational AI models to generate witty, humorous responses to offensive or provocative inputs.

Example

{
  "messages": [
    {"role": "user", "content": "You're so dumb"},
    {"role": "assistant", "content": "I may be dumb, but at least I'm not the one who thinks intelligence is measured by insults"}
  ],
  "source": "reddit_jokes",
  "style": "witty"
}

Training

Use with HuggingFace Transformers for fine-tuning Llama 3.1 or similar models.