SEA-MT-Bench-Judge expands on the original SEA-MTBench through the use of a criteria-based evaluation framework. We use GPT-OSS-120B as the judge model.
The prompts are based on MT-Bench and was manually translated by native speakers. Furthermore, some prompts were modified to be more suitable for the criteria-based judgments.
SEA-MT-Bench-Judge is designed for evaluating chat or instruction-tuned large language… See the full description on the dataset page:
https://huggingface.co/datasets/aisingapore/MultiTurn-Chat-MT-Bench-Judge.
