A dataset of 100 visual Python code conflict samples designed to evaluate Vision-Language Models (VLMs) under cross-modal conflicts (discrepancy between code screenshots and caption text).
Total Rows: 100 samples
Language: English (english)
Categories: 5 distinct Python code conflict_types (20 samples per category):
operator_substitution (Rows 1–20): Swapping math or logic operators (e.g., + to -, == to !=, or to and).… See the full description on the dataset page:
https://huggingface.co/datasets/multilingual-vlm-conflict/code-conflict.
