MetaQuery Instruct 2.4M: Multimodal Instruction-Tuning Dataset

The MetaQuery dataset, created by author xcpan, was last updated on June 30, 2025. It is sourced from the MMC4 dataset and contains 2.4 million instruction-response pairs, likely for training or fine-tuning AI models. The data is released under multiple licenses, including CC-BY-NC, ODC-BY, and Common Crawl terms of use.

Use Cases

• Fine-tuning language models for instruction-following based on the 2.4M instruction-response pairs.
• Training multimodal AI systems based on the dataset's likely combination of text and other modalities.
• Benchmarking model performance on instruction-based tasks using the structured query-response format.

Strengths

• Contains 2.4 million instruction-response pairs, providing a substantial scale for model training.
• Sourced from the established MMC4 dataset, suggesting a known origin.
• Explicitly lists multiple applicable licenses (CC-BY-NC, ODC-BY, Common Crawl), clarifying usage terms.

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

Sourced from the MMC4 dataset.

Collection Method

Likely involves processing and structuring data from MMC4 into instruction-response pairs.

Freshness

Last updated 2025-06-30 19:22:09; freshness should be verified.