diff --git a/lm_eval/tasks/hellaswag/hellaswag_ru.yaml b/lm_eval/tasks/hellaswag/hellaswag_ru.yaml index 4992d61725..add898f576 100644 --- a/lm_eval/tasks/hellaswag/hellaswag_ru.yaml +++ b/lm_eval/tasks/hellaswag/hellaswag_ru.yaml @@ -13,9 +13,6 @@ doc_to_text: "{{query}}" doc_to_target: "{{label}}" doc_to_choice: "choices" metric_list: - - metric: acc - aggregation: mean - higher_is_better: true - metric: acc_norm aggregation: mean higher_is_better: true diff --git a/lm_eval/tasks/winogrande/default_ru.yaml b/lm_eval/tasks/winogrande/default_ru.yaml index 0b6ad4cc13..ceb69f8c71 100644 --- a/lm_eval/tasks/winogrande/default_ru.yaml +++ b/lm_eval/tasks/winogrande/default_ru.yaml @@ -3,6 +3,6 @@ task: winogrande_ru dataset_path: deepvk/winogrande-ru dataset_name: null dataset_kwargs: - revision: vk-edited + revision: second-iteration num_fewshot: 5 fewshot_split: train \ No newline at end of file