Sign in to view source links and access this dataset
Description
1 million high-quality synthetic examples covering math, science, code, and puzzles. The dataset builds upon the previous OpenThoughts-114k and augments it with data from sources like OpenR1. It was created by open-thoughts and last updated on June 5, 2025.
Use Cases
Training reasoning models based on the described math and science problems.
Fine-tuning code generation models based on the included programming examples.
Benchmarking model performance on puzzle-solving tasks mentioned in the description.
Augmenting existing instruction-tuning datasets with high-quality synthetic reasoning data.
Strengths
Contains 1 million examples, a significant scale for training.
Builds upon and augments established datasets like OpenThoughts-114k and OpenR1.
Was used to train specific models (OpenThinker2-7B and OpenThinker2-32B), indicating practical utility.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count is known, but specific file formats, size, and license details are unknown.
Data may reflect source bias inherent to the aggregated synthetic and existing datasets.
Provenance
Source
open-thoughts on Hugging Face.
Collection Method
Synthetic generation and augmentation of existing datasets like OpenR1.
Freshness
Last updated 2025-06-05 16:27:27; freshness should be verified.
License is unknown; users must verify terms before use.