ThoughtSource⚡ logo

ThoughtSource⚡

Free

A framework for the science of machine thinking ![GitHub Repo stars](https://img.shields.io/github/stars/OpenBioLink/ThoughtSource?style=social)

FreeFree tier
Inputs: textOutputs: text
Type
Open Source

About ThoughtSource⚡

ThoughtSource is a central, open resource and community centered on data and tools for chain-of-thought (CoT) reasoning in large language models. Developed by the Samwald research group, it provides standardized datasets, dataloaders, and a dataset annotator to support trustworthy and robust reasoning in AI systems, particularly for scientific and medical question answering. The project includes reasoning chains from multiple sources (human-generated, AI few-shot, AI zero-shot) and uses the Hugging Face Datasets format for easy integration.

Key Features

Curated chain-of-thought datasets from multiple sources (human-generated, AI few-shot, AI zero-shot)
Standardized dataloaders in Hugging Face Datasets format
Tutorial notebook and dataset annotator included
Supports general and scientific/medical question answering (CommonsenseQA, StrategyQA, WorldTree, EntailmentBank, etc.)
Open source and community-driven development

Pros & Cons

Pros
  • Open source and completely free to use
  • Provides high-quality, curated reasoning chains from diverse datasets
  • Standardized data format (Hugging Face) simplifies integration with existing ML workflows
  • Actively maintained and community-supported
Cons
  • Focused on research and experimentation, not a production-ready tool
  • Limited to existing question-answering and reasoning datasets
  • Requires familiarity with Python and handling of Python notebooks
  • Some datasets have unknown licensing for AI-generated chains

Best For

Research in chain-of-thought reasoning and large language model evaluationImproving reasoning capabilities of AI systems for scientific researchMedical question answering and clinical decision supportBenchmarking and comparing CoT prompting strategies

FAQ

What is ThoughtSource?
ThoughtSource is an open resource and community providing data and tools for chain-of-thought reasoning in large language models, developed by the Samwald research group.
What datasets are included?
Datasets include CommonsenseQA, StrategyQA, QED, WorldTree, EntailmentBank, and OpenBookQA, each with human and/or AI-generated reasoning chains.
Is ThoughtSource free to use?
Yes, ThoughtSource is open source and hosted on GitHub under an open license.
How are the reasoning chains formatted?
The dataloaders output objects in the Hugging Face Datasets format, providing standardized access to chain-of-thought data.