Rabe3/egyptian-arabic-sharegpt

Egyptian Arabic Conversations in ShareGPT Format

This dataset contains conversational examples in Egyptian Arabic dialect, formatted in the ShareGPT format with 'from'/'value' fields that is compatible with Llama 3.1 fine-tuning using Unsloth.

Dataset Structure

Each example contains:

• conversations: A list of messages with from and value fields
• source: Origin of the data ('egyptian_arabic')
• score: Quality score for the conversation (1.0)

Example

{
  "conversations": [
    {
      "from": "human",
      "value": "ممكن نتكلم شوية عن المال؟"
    },
    {
      "from": "gpt",
      "value": "طبعاً، أنا بحب أتكلم عن المال. إيه اللي حابب تعرفه بالظبط؟"
    }
  ],
  "source": "egyptian_arabic",
  "score": 1.0
}

Usage with Unsloth and Llama 3.1

This dataset is specifically formatted to work with Unsloth for fine-tuning Llama 3.1 models:

from datasets import load_dataset
dataset = load_dataset("Rabe3/egyptian-arabic-sharegpt")