garrykuwanto/wc-vqa-continual-pretraining

garrykuwanto/wc-vqa-continual-pretraining English subset of worldcuisines/vqa-v1.1 Task 1 (dish name prediction), packaged for continual pretraining / SFT of small vision-language models (specifically SmolVLM2-256M). Splits split rows source train 27000 task1 / train / lang=en validation 300 task1 / test_small / lang=en test 1500 task1 / test_large / lang=en Schema field type notes image Image bytes from upstream… See the full description on the dataset page: https://huggingface.co/datasets/garrykuwanto/wc-vqa-continual-pretraining.