med-synth-questions-gemma-3-27b-it

# openmed-community/med-synth-questions-gemma-3-27b-it ## What is this? **Med Synth Questions — Gemma 3 27B IT** is a compact, instruction-only dataset of **33,325 English medical questions** generated from the texts in **`gamino/wiki_medical_terms`**. Each row contains a single question (`input`), the structured `generation_settings` used to produce it, and an ISO-8601 `timestamp`. The source corpus consists of Wikipedia-based medical term pages assembled in `wiki_medical_terms` (licensed **GPL-3.0**), from which this dataset inherits its license. Questions were produced with **`google/gemma-3-27b-it`** (instruction-tuned) and include the exact model+parameters in `generation_settings`, enabling reproducibility and filtering by model version. See the model card and Gemma docs for capability and context-length details, and **ensure you follow Google’s Gemma Terms** when using model-generated content. --- ## Dataset structure ```json DatasetDict({ train: Dataset({ features: ['input', 'generation_settings', 'timestamp'], num_rows: 33325 }) }) ``` **Features** - `input` *(string)* — a single, self-contained medical question. - `generation_settings` *(dict)* — structured metadata, typically including: - `model` (e.g., `"google/gemma-3-27b-it"`), - `provider` (e.g., `"openrouter"`), - `num_questions_requested`, `num_questions_generated`, - decoding params such as `max_tokens`. - `timestamp` *(string)* — ISO-8601 creation time. **Example** ```json { "input": "Hi there! Considering miscarriage rates vary with age, what is the approximate risk for a woman under 35 compared to a woman over 40?", "generation_settings": { "max_tokens": 4096, "model": "google/gemma-3-27b-it", "num_questions_generated": 5, "num_questions_requested": 5, "provider": "openrouter" }, "timestamp": "2025-08-17T01:24:57.106856" } ``` --- ## Source & provenance * **Source texts**: [gamino/wiki_medical_terms](https://huggingface.co/datasets/gamino/wiki_medical_terms) (≈6.8k entries of medical terms with Wikipedia text). License: **GPL-3.0**. * **Generator model**: `google/gemma-3-27b-it` (Gemma 3, instruction-tuned). See the model card and Gemma 3 overview. ([Hugging Face][1], [Google AI for Developers][3]) * **Terms**: Use of Gemma outputs is governed by Google’s **Gemma Terms of Use** and **Prohibited Use Policy** in addition to this dataset’s license. ([Google AI for Developers][4]) --- ## Intended uses * **Instruction-only fine-tuning scaffolds** (e.g., generate answers/feedback with your own pipeline). * **Retrieval & evaluation**: as a bank of domain-specific queries for RAG/QA systems. * **Question generation research**: study prompt styles, difficulty, and topic coverage. ### Out-of-scope / caveats * The dataset contains **questions only**; no ground-truth answers are provided. * Content originates from web-derived Wikipedia text; questions may reflect **biases, gaps, or outdated info** - perform task-appropriate validation before use. * **Not medical advice**. Do not use for clinical decision-making. --- ## How to load ```python from datasets import load_dataset ds = load_dataset("openmed-community/med-synth-questions-gemma-3-27b-it", split="train") # Peek row = ds[0] print(row["input"]) print(row["generation_settings"]) print(row["timestamp"]) ``` --- ## Licensing & responsible use * **Dataset license**: **GPL-3.0**, inherited from `gamino/wiki_medical_terms`. Downstream distributions and derivatives must comply with GPL-3.0. ([Hugging Face][1]) * **Model terms**: Outputs created with **Gemma 3** are subject to **Gemma Terms of Use** (and the **Prohibited Use Policy**). Using this dataset **does not grant** additional rights to use Gemma or its outputs beyond those terms; review and comply when training, evaluating, or redistributing derivatives. ([Google AI for Developers][4]) --- ## Acknowledgements * **Source dataset**: `gamino/wiki_medical_terms`. ([Hugging Face][1]) * **Model**: `google/gemma-3-27b-it` (Gemma 3). ([Hugging Face][2]) --- ## Citation If you use this dataset, please cite: * `gamino/wiki_medical_terms` (Hugging Face dataset card). ([Hugging Face][1]) * `google/gemma-3-27b-it` (model card / docs). ([Hugging Face][2], [Google AI for Developers][3]) --- ## Disclaimer This resource is provided **for research and educational use**. It is **not** a source of medical advice. Always consult qualified professionals and follow relevant laws, ethics, and platform/model terms. Use responsibly. - [1]: [gamino wiki_medical_terms](https://huggingface.co/datasets/gamino/wiki_medical_terms) - [2]: [google gemma-3-27b-it](https://huggingface.co/google/gemma-3-27b-it) - [3]: [Gemma 3 model overview | Google AI for Developers](https://ai.google.dev/gemma/docs/core) - [4]: [Gemma Terms of Use | Google AI for Developers - Gemini API](https://ai.google.dev/gemma/terms) --- ## Reproduce To reproduce or adapt the pipeline, see our open-source [Synthetic Questions Generation tool](https://github.com/mkurman/synthetic-questions-generation)