Bialy17/Reasoning-Socratic-QandA

Reasoning-Socratic-QandA Dataset

This dataset is a curated mixture of three high-quality data sources, specifically engineered to train Socratic Tutors. It balances the model's ability to think deeply (Reasoning), guide students through questioning (Pedagogy), and provide direct answers when necessary (Support).

Dataset Composition

Source Datasets

1. 1.[Alibaba-Apsara/Superior-Reasoning-SFT-gpt-oss-120b](https://huggingface.co/datasets/Alibaba-Apsara/Superior-Reasoning-SFT-gpt-oss-120b): Used for the "DASD Stage 1" component to provide high-quality Chain-of-Thought (CoT) reasoning.
2. 2.[sanjaypantdsd/socratic-method-conversations](https://huggingface.co/datasets/sanjaypantdsd/socratic-method-conversations): Used for the "Socratic" component to teach the model how to ask leading questions rather than providing immediate solutions.
3. 3.[databricks/databricks-dolly-15k](https://huggingface.co/datasets/databricks/databricks-dolly-15k): Used for the "Q&A" component to ensure the model remains a helpful assistant for general knowledge and direct inquiries.

Intended Use

This dataset is designed for fine-tuning Large Language Models (LLMs) and Small Language Models (SLMs) to act as AI Tutors.

The goal of this specific mixture is to create a model that:

1. 1.Analyzes the problem first (using DASD reasoning).
2. 2.Guides the student via the Socratic method.
3. 3.Explains concepts clearly if the student is unable to proceed.

Formatting

The dataset is typically formatted in a conversational structure:

• —instruction: The prompt or student's question.
• —context: (Optional) Background information.
• —response: The pedagogical response (Reasoning + Socratic Question OR Direct Answer).

Licensing

This dataset is a compilation. Please refer to the licenses of the original source datasets:

• —Dolly-15k: CC BY-SA 3.0
• —Superior-Reasoning-SFT: Apache 2.0 / CC BY 4.0
• —Socratic-Method-Conversations: MIT