Math-VR-train: English Training Subset for Mathematical Visual Reasoning

Math-VR-train-en is a large-scale bilingual training dataset designed for research on mathematical visual reasoning. It accompanies the Math-VR Benchmark and was created by author gogoduan, with its metadata last updated on October 22, 2025. The dataset addresses the need for visual reasoning in mathematical problem-solving, such as drawing auxiliary lines or plotting geometric shapes.

Use Cases

• Training models for mathematical visual reasoning based on the described focus on auxiliary lines and geometric shapes.
• Benchmarking AI performance on bilingual mathematical tasks based on the dataset's stated bilingual nature.
• Researching the intersection of text and visual modalities in education based on the description of combining text solutions with visual reasoning.

Strengths

• Dataset is described as 'large-scale'.
• Specifically designed for mathematical visual reasoning, a niche research area.
• Metadata indicates it is part of a benchmark, suggesting a defined evaluation framework.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Freshness

Last updated 2025-10-22 18:38:55; freshness should be verified.