ELYZA-tasks-100

This dataset is a dataset for evaluating models that have undergone instruction-tuning. Features: 100 Japanese data containing complex instructions/tasks. As a useful AI assistant, careful output is required. All data is annotated with evaluation viewpoints, which is expected to suppress fluctuations in evaluation. Specifically, the tasks include the following: Tasks to revise the summary and explain the corrections Tasks that describe abstract lessons from concrete episodes Tasks that act as a helpful AI assistant that understands the user's intentions Complex arithmetic tasks that require case classification A task that requires advanced reasoning to extract patterns from an Unknown language and translate them into Japanese. The task of generating YouTube dialogue based on multiple instructions Tasks that require imagination, such as generation and comprehension of imaginary creatures and idioms.