PhysRL: Physics Reasoning Datasets for Multimodal AI Training

Kun-Xiang created the PhysRL collection to accompany the SeePhys Pro research paper. The dataset includes the full PhysRL-38K corpus and a vision-necessary subset of PhysRL-8K, used for studying multimodal reasoning in physics. It was last updated on HuggingFace on 2026-05-13.

Use Cases

• Training multimodal reinforcement learning agents based on physics reasoning problems.
• Benchmarking model reliance on visual evidence versus text-only reasoning based on the described vision-necessary subset.
• Studying modality transfer and blind-training effects in AI systems based on the paper's described research focus.

Strengths

• Includes a large collection of 38,000 physics reasoning problems.
• Provides a curated subset of 8,000 problems specifically identified as requiring visual evidence.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the main subsets is known, but overall dataset size and file formats are unspecified.

Provenance

Source

HuggingFace dataset repository by author Kun-Xiang.

Collection Method

Created as a training corpus for the associated research paper.

Freshness

Last updated 2026-05-13 06:00:21; freshness should be verified.