ChemPile-Education

A comprehensive chemistry education dataset containing 129.26M+ tokens from diverse educational sources

📋 Dataset Summary

ChemPile-Education is a large-scale chemistry natural language dataset extracted from diverse educational resources, including open-source textbooks, lecture notes, transcripts, and other educational materials. This dataset captures fundamental chemistry knowledge and concepts as students would encounter them in textbooks and classroom settings.

📊 Dataset Statistics

Subset	Tokens	Documents	Description
LibreText Chemistry	114M	58,9K	Open-source chemistry textbooks
MIT OCW Lecture Transcripts	10M	58K	University-level lecture transcripts
US Olympiad Problems	260K	1,22K	Competition chemistry problems with solutions
YouTube Transcripts	5M	5,09K	Educational chemistry video transcripts
Total	~129.26M	~123K	Chemical educational content

🗂️ Dataset Configurations

The dataset includes four distinct subsets available as Hugging Face configurations:

• LibreText_Chemistry-default
• mit-ocw-lecture-transcripts-default
• us-olympiad-problems-default
• youtube-transcripts-as-lectures-default

📜 License

All content is released under the CC BY-NC-SA 4.0 license, which allows for:

• ✅ Non-commercial use
• ✅ Sharing and redistribution
• ✅ Adaptation and modification
• ⚠️ Attribution required
• ⚠️ Share-alike (derivatives must use same license)

📖 Dataset Details

🧪 LibreText Chemistry

Source: LibreTexts Chemistry - A comprehensive open-source chemistry education platform

Coverage: General chemistry, organic chemistry, inorganic chemistry, physical chemistry, and biochemistry

Extraction Method: Custom web scraper for HTML content extraction

Fields:

• url: Source URL for transparency and verification
• text: Educational content about chemistry concepts

Statistics: 114M tokens across 58,946 documents covering undergraduate to graduate-level chemistry topics

🎓 MIT OCW Lecture Transcripts

Source: MIT OpenCourseWare - Free online course materials from MIT

Coverage: Chemistry, biology, chemical engineering, and physics courses

Extraction Method: Automated download using keyword-based course identification

Fields:

• course: Course name and identifier
• url: Original source URL for reference
• topic: Specific lecture topic
• text: Lecture transcript content
• index: Document identifier

Statistics: 10M tokens from approximately 500 university-level lectures

🏆 US Olympiad Problems

Source: American Chemical Society (ACS) US Chemistry Olympiad materials

Coverage: Competition-level chemistry problems with detailed solutions

Extraction Method: PDF processing with LLM-generated Q&A pairs using Gemini 2.0 Flash Thinking Experimental 01-21

Quality Control:

• Minimum 250 characters per answer
• Expert manual review by chemistry professionals
• Filtered for educational value

Fields:

• metadata: Problem details (year, number, topic)
• problem_statement: Original olympiad problem
• options: Multiple choice options
• solution: Detailed solution explanation
• correct_answer: Correct option identifier
• text: Combined problem and solution content
• filter: Quality control flag
• index: Document identifier

Statistics: 260K+ tokens from ~400 competition problems with comprehensive solutions

📺 YouTube Transcripts as Lectures

Source: Educational chemistry videos from YouTube with permissive CC licenses

Coverage: Diverse chemistry topics from educational content creators

Extraction Method:

• LLM-generated search keywords for educational chemistry content
• Automated transcript extraction from CC-licensed videos
• Content cleaning and structuring using GPT-4.1

Fields:

• id: Unique YouTube video identifier
• title: Video title
• text: Cleaned and structured transcript content
• index: Document identifier

Statistics: 5M+ tokens from ~1,000 educational chemistry videos

🚀 Quick Start

from datasets import load_dataset, get_dataset_config_names

# List all available configurations
configs = get_dataset_config_names("jablonkagroup/chempile-education")
print(f"Available configs: {configs}")
# ['LibreText_Chemistry-default', 'mit-ocw-lecture-transcripts-default', 
#  'us-olympiad-problems-default', 'youtube-transcripts-as-lectures-default']

# Load a specific subset
dataset = load_dataset("jablonkagroup/chempile-education", name="LibreText_Chemistry-default")

print(dataset)
# DatasetDict({
#     train: Dataset({
#         features: ['url', 'text'],
#         num_rows: 53051
#     })
#     test: Dataset({
#         features: ['url', 'text'], 
#         num_rows: 2948
#     })
#     val: Dataset({
#         features: ['url', 'text'],
#         num_rows: 2947
#     })
# })

# Access a sample
sample = dataset['train'][0]
print(f"Sample URL: {sample['url'][:50]}...")
print(f"Sample text: {sample['text'][:200]}...")

🎯 Use Cases

• 🤖 Language Model Training: Pre-training or fine-tuning models for chemistry domain
• 📚 Educational AI: Building chemistry tutoring and Q&A systems
• 🔍 Information Retrieval: Chemistry knowledge base construction
• 📝 Content Generation: Automated chemistry educational content creation
• 🧠 Domain Adaptation: Adapting general models to chemistry domain

⚠️ Limitations & Considerations

• Language: English only (monolingual dataset)
• Scope: Focused on educational content; may not cover cutting-edge research and some knowledge might be missing
• Quality: Variable quality across sources; some automated transcripts may contain errors
• Bias: Reflects biases present in educational materials and sources
• License: Non-commercial use only due to CC BY-NC-SA 4.0 license

🛠️ Data Processing Pipeline

1. Collection: Automated scraping and API-based collection from sources
2. Extraction: Text extraction from various formats (HTML, PDF, video transcripts)
3. Cleaning: LLM-assisted content cleaning and structuring
4. Quality Control: Expert review and filtering
5. Standardization: Consistent formatting and metadata addition
6. Validation: Split creation and final quality checks

🏗️ ChemPile Collection

This dataset is part of the ChemPile collection, a comprehensive open dataset containing over 75 billion tokens of curated chemical data for training and evaluating general-purpose models in the chemical sciences.

Collection Overview

• 📊 Scale: 75+ billion tokens across multiple modalities
• 🧬 Modalities: Structured representations (SMILES, SELFIES, IUPAC, InChI), scientific text, executable code, and molecular images
• 🎯 Design: Integrates foundational educational knowledge with specialized scientific literature
• 🔬 Curation: Extensive expert curation and validation
• 📈 Benchmarking: Standardized train/validation/test splits for robust evaluation
• 🌐 Availability: Openly released via Hugging Face

📄 Citation

If you use this dataset in your research, please cite:

@article{mirza2025chempile0,
  title   = {ChemPile: A 250GB Diverse and Curated Dataset for Chemical Foundation Models},
  author  = {Adrian Mirza and Nawaf Alampara and Martiño Ríos-García and others},
  year    = {2025},
  journal = {arXiv preprint arXiv:2505.12534}
}

👥 Contact & Support

• Paper: arXiv:2505.12534
• Website: ChemPile Project
• Dataset: Hugging Face
• Issues: Please report data issues or questions via the Hugging Face dataset page

---

Part of the ChemPile project - Advancing AI for Chemical Sciences