Ultrainteract Math CoT: 78,349 Math Problems with GPT-5.4-mini Solutions

A knowledge-distillation training set of 78,349 math problems with step-by-step solutions. The dataset was created by regenerating the model predictions from the openbmb/UltraInteract_sft dataset using OpenAI's GPT-5.4-mini model. It was authored by vukien2301 and last updated on Hugging Face in May 2026.

Use Cases

• Fine-tune language models for mathematical problem-solving based on the step-by-step solutions.
• Train models for knowledge distillation using the regenerated GPT-5.4-mini outputs.
• Benchmark model performance on chain-of-thought reasoning tasks based on the 78,349 math prompts.
• Study the characteristics of synthetic reasoning data generated by advanced language models.

Strengths

• Contains 78,349 original math problem prompts, providing a substantial scale for training.
• Features step-by-step solutions generated by a specific, named model (GPT-5.4-mini), offering consistency.
• Derived from a known source dataset (openbmb/UltraInteract_sft), providing traceable provenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect the biases and limitations inherent to the source dataset and the GPT-5.4-mini model used for regeneration.

Provenance

Source

Derived from the openbmb/UltraInteract_sft dataset, filtered for Math_CoT tasks.

Collection Method

Answers were regenerated using OpenAI's GPT-5.4-mini model.

Freshness

Last updated 2026-05-17 04:22:59; freshness should be verified.