Japanese-Mail-Bench

Japanese-Mail-Bench Japanese-Mail-Bench is a Japanese business email generation benchmark. Each item gives a model a Japanese email-writing prompt built from a scenario (sender, receiver, purpose, situation, tone constraints) and pairs it with an LLM-generated reference answer produced by the benchmark construction pipeline. The bundled evaluation script eval.py calls a target model to generate the email, then asks an OpenAI-compatible judge model to score the result on a single… See the full description on the dataset page: huggingface.co/datasets/pfnet/japanese-mail-bench.