MetaQuery Instruct 2.4M: Multimodal Instruction-Tuning Data

MetaQuery Instruct 2.4M is a dataset sourced from the MMC4 corpus, indicating a large-scale collection of multimodal content. The dataset was created by author xcpan and was last updated on the Hugging Face platform in June 2025. Its licensing involves a mix of CC-BY-NC, ODC-BY, and Common Crawl terms of use.

Use Cases

• Training models to follow multimodal instructions based on the dataset's name and source.
• Fine-tuning vision-language models for complex tasks based on the 'instruct' label.
• Benchmarking model performance on cross-modal understanding based on the implied data structure.

Strengths

• Dataset name suggests a scale of 2.4 million entries.
• Sourced from the established MMC4 corpus, implying a known provenance.

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

xcpan (author), sourced from MMC4 corpus.

Collection Method

Likely aggregated and processed from the MMC4 web crawl.

Time Range

null

Freshness

Last updated 2025-06-30 19:21:43; freshness should be verified.

Geography

null