khhuang/

CHOCOLATE

Dataset Description Paper Information Citation

CHOCOLATE is a benchmark for detecting and correcting factual inconsistency in generated chart captions. It consists of captions produced by six most advanced models, which are categorized into three subsets:

LVLM: GPT-4V, Bard (before Gemini) LLM-based Pipeline: DePlot + GPT-4 Fine-tuned Model: ChartT5, MatCha, UniChart

The charts are from two datasets: VisText and the… See the full description on the dataset page: https://huggingface.co/datasets/khhuang/CHOCOLATE.