README.md

metadata

license: cc-by-4.0
task_categories:
  - text-generation
language:
  - en
  - pt
tags:
  - synthetic
  - gemini
  - instruction-following
pretty_name: Synthetic Gemini Multi-Model Dataset
size_categories:
  - 1K<n<10K

Synthetic Gemini Multi-Model Dataset

Dataset Description

This dataset is a collection of high-quality synthetic conversational text data synthesized using various models from the Google Gemini ecosystem. It is formatted explicitly for multi-turn instruction tuning and conversational alignment.

Dataset Summary

• Repository: Synthetic Gemini Dataset
• Languages: English (en) / Portuguese (pt)
• License: Creative Commons Attribution 4.0 International (cc-by-4.0)
• Task Type: Text Generation / Conversational Chat

---

Generation Details & Models Used

The outputs in this dataset were generated across a blend of Gemini model variants to capture diverse reasoning styles, structural formats, and generation characteristics:

• Gemini 3.5 Flash & 3.6 Flash: High-speed reasoning, creative text generation, and fast contextual expansions.
• Gemini 3.5 Flash-Lite: Highly efficient lightweight generation for rapid instruction-response formatting.
• Gemini 3.1 Pro: Multi-step complex instructions, deep technical explanations, and detailed reasoning outputs.
• Other Gemini Variants: Supplemental data generated across various specialized sub-variants.

---

Dataset Structure

Data Fields

The dataset strictly contains a single field formatted for standardized conversational training:

• messages (list of objects): A list of chat turn objects representing the conversation history. Each object contains:
  • role (string): The turn role (system, user, or assistant).
  • content (string): The raw text content of the turn message.

Format Example

{
  "messages": [
    {"role": "system", "content": "You are a helpful AI assistant."},
    {"role": "user", "content": "Explain quantum computing in simple terms."},
    {"role": "assistant", "content": "Quantum computing is a type of computing that..."}
  ]
}

---

Intended Uses & Limitations

Primary Use Cases

• Supervised Fine-Tuning (SFT): Direct compatibility with standard SFT pipelines (such as Unsloth, LLaMA-Factory, or Hugging Face TRL) using standard chat template formatters.
• Multi-Turn Chatbot Training: Aligning base models for multi-turn assistant capabilities.

Limitations & Biases

• Synthetic Artifacts: Outputs may reflect stylistic preferences and formatting quirks native to the Gemini family.
• Fact Verification: Users should validate domain-specific factual correctness before deploying downstream models in critical applications.

---

Licensing & Attribution

This dataset is licensed under CC-BY-4.0. You can freely adapt, share, or build commercial models on top of it, provided you give appropriate attribution to the dataset creator 🧿✨.

---