Merge branch 'main' into weight_by_size

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_interior_architecture_and_design.yaml

+dataset_name: interior_architecture_and_design
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_interior_architecture_and_design

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_korean_history.yaml

+dataset_name: korean_history
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_korean_history

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_law.yaml

+dataset_name: law
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_law

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_machine_design_and_manufacturing.yaml

+dataset_name: machine_design_and_manufacturing
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_machine_design_and_manufacturing

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_management.yaml

+dataset_name: management
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_management

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_maritime_engineering.yaml

+dataset_name: maritime_engineering
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_maritime_engineering

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_marketing.yaml

+dataset_name: marketing
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_marketing

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_materials_engineering.yaml

+dataset_name: materials_engineering
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_materials_engineering

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_math.yaml

+dataset_name: math
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_math

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_mechanical_engineering.yaml

+dataset_name: mechanical_engineering
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_mechanical_engineering

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_nondestructive_testing.yaml

+dataset_name: nondestructive_testing
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_nondestructive_testing

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_patent.yaml

+dataset_name: patent
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_patent

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_political_science_and_sociology.yaml

+dataset_name: political_science_and_sociology
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_political_science_and_sociology

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_psychology.yaml

+dataset_name: psychology
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_psychology

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_public_safety.yaml

+dataset_name: public_safety
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_public_safety

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_railway_and_automotive_engineering.yaml

+dataset_name: railway_and_automotive_engineering
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_railway_and_automotive_engineering

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_real_estate.yaml

+dataset_name: real_estate
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_real_estate

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_refrigerating_machinery.yaml

+dataset_name: refrigerating_machinery
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_refrigerating_machinery

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_social_welfare.yaml

+dataset_name: social_welfare
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_social_welfare

lm_eval/tasks/kmmlu/direct_hard/kmmlu_direct_hard_taxation.yaml

+dataset_name: taxation
+include: _direct_hard_kmmlu_yaml
+task: kmmlu_hard_direct_taxation