[Benchmark] Add MLLMJudgeDataset

[Misaki-Wang]

MLLM-AS-A-JUDGE is a benchmark in multimodal domains, with human annotations to assess the judging capability of MLLMs in tasks of Scoring Evaluation, Pair Comparison and Batch Ranking.
Project website: https://mllm-judge.github.io/

[kennymckormick]

Hi, @Misaki-Wang ,
Just checked the implementation and find the evaluation doesn't run properly.
For example, the batch_data.tsv does not have the index field, which is required by our evaluation script.
Would you please first make sure that the three newly added benchmarks can properly run on at least one VLM. You can ping me after you accomplished this.

[kennymckormick]

Hi, @Misaki-Wang , any updates?