PinkPixel/Reasoning-Mix-100k

🧠 Reasoning Mix 100k ✨

A high-quality, balanced reasoning dataset consisting of 99,999 samples extracted from three reasoning datasets. This dataset is specifically formatted for models to utilize a /think block for step-by-step reasoning.

📊 Dataset Summary

The Reasoning Mix 100k is a curated collection of reasoning tasks, primarily focused on mathematics, logic, and general problem-solving. It combines the strengths of three high-performing datasets into a unified format optimized for Chain-of-Thought (CoT) training.

🛠️ Formatting

The dataset uses the ChatML format with a specialized system prompt and assistant response structure.

📝 System Prompt

"You are a highly capable AI assistant. For every response, you must first show your step-by-step reasoning process within a /think block, followed by the final answer."

🤖 Assistant Responses

Every response is wrapped in a /think block structure:

/think
[Step-by-step reasoning process...]
/think

[Final answer or solution]

📂 Dataset Structure

Each sample in the reasoning_mix_100k.jsonl file follows this structure:

{
  "messages": [
    {
      "role": "system",
      "content": "You are a highly capable AI assistant..."
    },
    {
      "role": "user",
      "content": "Evaluate the integral..."
    },
    {
      "role": "assistant",
      "content": "/think\n[Reasoning...]\n/think\n\n[Final Answer]"
    }
  ]
}

🚀 Usage

You can load this dataset using the Hugging Face datasets library:

from datasets import load_dataset

dataset = load_dataset("pinkpixel/reasoning-mix-100k")
print(dataset["train"][0])

📜 License

This dataset is released under the Apache License 2.0.

Made with ❤️ by Pink Pixel ✨