SamuelTheophilus/

final_blind_spots

	Qwen3-VL-2B-Instruct Failure Case Dataset




	
	Dataset Summary

This dataset contains 47 curated multimodal failure cases of Qwen3-VL-2B-Instruct, a 2-billion-parameter vision-language model (NB: there is no "Qwen3-VL-2B base" model as the Qwen3-VL family mainly has "Instruct" and "Thinking" versions). Each example captures a scenario where the model produced an incorrect, misleading, incomplete, or poorly grounded response to a multimodal prompt. The goal of this dataset is to… See the full description on the dataset page: https://huggingface.co/datasets/SamuelTheophilus/final_blind_spots.