kollessisopod/vqa-mixed

VQA Mixed (GQA + VizWiz + VQAv2 subset) This dataset is a curated mixture of three major VQA benchmarks, subsampled and reformatted into Parquet shards for efficient model training (e.g., for VLMs like LLaVA or Qwen-VL). Dataset Details The dataset consists of image-question-answer triplets stored in Parquet format. Data Splits Train: train/data-*.parquet Validation: validation/data-*.parquet Columns image: The visual input (Image… See the full description on the dataset page: https://huggingface.co/datasets/kollessisopod/vqa-mixed.