vietmed/ChemPatentTableQA-Judge

Chem Patent Table QA — Judge数据集包含人类对基于化学专利表格图像的候选问答对的评判，用于训练或评估一个基于视觉语言模型（VLM）的LLM-as-a-judge系统，该系统用于评分来自`vietmed/ChemPatentTableQA`数据集的输出。评判标准包括答案正确性（正确/部分正确/不正确/无法验证）、推理质量（优秀/好/一般/差/错误）、问题质量（好/可接受/差）以及整体评判（包含/修复后包含/排除）。数据集分为训练集（70行）、验证集（15行）和测试集（15行），并包含图像、问题、候选答案、候选推理、上下文等多个字段。

The Chem Patent Table QA — Judge Dataset contains human verdicts over candidate QA pairs grounded in chemical-patent table images, built to train or evaluate an LLM-as-a-judge VLM that scores outputs from `vietmed/ChemPatentTableQA`. The verdicts include structured judgments on answer correctness (correct/partially_correct/incorrect/cannot_verify), reasoning quality (excellent/good/fair/poor/wrong), question quality (good/acceptable/poor), and overall verdict (include/fix_then_include/exclude). The dataset is split into train (70 rows), validation (15 rows), and test (15 rows) sets, and includes fields such as image, question, candidate_answer, candidate_reasoning, context, and more.