This dataset contains prompt-only benchmark instances for Emoji-Bench.
example_id: unique row id
base_id: shared id across clean/error variants of the same underlying problem
split: train / validation / test
difficulty: easy / medium / hard / expert
condition: clean or error_injected
error_type: null or an injected error label such as E-RES, E-INV, E-CASC, or E-RECONV
has_error: whether the prompt contains an injected error… See the full description on the dataset page:
https://huggingface.co/datasets/huyxdang/emoji-bench-e-reconv-1000.
