MultiTasks-v2: Twelve Multimodal Benchmarks in Image-QA Format

MultiTasks-v2 is a collection of twelve multimodal benchmark subsets normalized into a shared image-question-answer format. The dataset, created by LoserLi, provides a train split and a test split for each subset. It was last updated on 2026-05-28.

Use Cases

• Benchmarking multimodal model performance based on the twelve normalized subsets
• Training vision-language models on a standardized image-question-answer format
• Evaluating model generalization across different multimodal tasks

Strengths

• Contains twelve distinct multimodal benchmark subsets
• Examples are normalized into a consistent image-question-answer structure
• Each subset provides dedicated train and test splits

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

LoserLi on Hugging Face

Collection Method

Collection and normalization of twelve existing multimodal benchmark subsets

Freshness

2026-05-28 14:40:22