VeriEvol-SFT: Multimodal STEM Reasoning Problems with Chain-of-Thought Solutions

VeriEvol-SFT is a supervised fine-tuning dataset for scaling multimodal mathematical reasoning via verifiable evolution instructions. The dataset, created by Ringo1110, contains single-image, single-turn visual STEM reasoning problems paired with long chain-of-thought solutions. Prompts are produced by route-specific evolution operators that rewrite problems, as detailed in the associated paper arXiv:2606.23543.

Use Cases

• Supervised fine-tuning of policy models for mathematical reasoning based on the described SFT stage data.
• Training models on visual STEM problem-solving based on the single-image, single-turn problem structure.
• Research into chain-of-thought reasoning generation based on the long solution format paired with each problem.
• Exploring verifiable instruction evolution based on the route-specific evolution operators mentioned in the description.

Strengths

• Each example is a single-image, single-turn visual STEM reasoning problem paired with a long chain-of-thought solution.
• Prompts are produced by route-specific evolution operators, suggesting a structured generation methodology.
• Dataset is associated with a published paper (arXiv:2606.23543) and code repository, indicating research provenance.

Limitations

• Description metadata is limited; actual data quality, column structure, and file formats require manual inspection after download.
• Row count, file size, and license information are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

huggingface user Ringo1110, associated with the VeriEvol research project.

Collection Method

Data generated for the supervised fine-tuning (SFT) stage to initialize a policy model before GRPO-style reinforcement learning, using route-specific evolution operators.

Freshness

Last updated 2026-07-05 16:37:51; freshness should be verified.