Multimodal Coding Dataset: 598k Samples for HTML, Chart, and Algorithmic Code Generation

598,000 high-quality samples for training and evaluating multimodal code generation models. The dataset covers HTML generation, chart-to-code, image-augmented QA, and algorithmic problems, supporting research in unifying vision-language understanding with code generation. It was created by author 'lingjie23' and last updated on December 24, —2025.

Use Cases

• Train models for HTML generation based on multimodal inputs.
• Develop chart-to-code translation systems based on visual and textual descriptions.
• Build models for solving algorithmic problems with image-augmented question answering.
• Evaluate the performance of vision-language models on unified code generation benchmarks.

Strengths

• Contains 598,000 high-quality samples, indicating a substantial scale.
• Covers four distinct task types: HTML generation, chart-to-code, image-augmented QA, and algorithmic problems.
• Explicitly designed for research in unifying vision-language understanding with code generation.

Limitations

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

Provenance

Source

lingjie23 on Hugging Face

Collection Method

Method of gathering is not specified in the provided description.

Time Range

Temporal coverage is not specified.

Freshness

Last updated 2025-12-24 02:06:48; freshness should be verified.

Geography

Spatial coverage is not specified.