Scientific Reasoning Dataset from In-the-Wild Literature

WildSci won second prize in the Reasoning Datasets Competition organized by Bespoke Labs, HuggingFace, and Together.AI. This dataset provides diverse, domain-rich science texts from 'in-the-wild' literature to advance scientific reasoning tasks for large language models. It was created by JustinTX and last updated in January 2026.

Use Cases

• Fine-tuning LLMs for scientific text generation using the 'text' modality.
• Benchmarking model performance on domain-rich science reasoning tasks.
• Training models to parse and reason with complex scientific concepts from literature.

Strengths

• Awarded second prize in the Reasoning Datasets Competition.
• Designed to address a noted lack of diverse, domain-rich science datasets.

Limitations

• Specific row count, column names, and sample size are unknown.
• The scope and exact sources of the 'in-the-wild' literature are not detailed.

Provenance

Source

JustinTX on Hugging Face.

Collection Method

Collected from 'in-the-wild' scientific literature.

Freshness

Last updated January 2026.

Geography

Region tag indicates 'us', but specific spatial coverage is unknown.