add many explicit group configs

lm_eval/tasks/hendrycks_math/hendrycks_math.yaml

@@ -7,3 +7,9 @@ task:
   - hendrycks_math_num_theory
   - hendrycks_math_prealgebra
   - hendrycks_math_precalc
+aggregate_metric_list:
+  - metric: exact_match
+    aggregation: mean
+    weight_by_size: true
+metadata:
+  version: 1.0

lm_eval/tasks/hendrycks_math/hendrycks_math_algebra.yaml

-group:
+tag:
   - math_word_problems
 task: hendrycks_math_algebra
 dataset_path: EleutherAI/hendrycks_math

lm_eval/tasks/kormedmcqa/_kormedmcqa.yaml

+group: kormedmcqa
+task:
+  - kormedmcqa_doctor
+  - kormedmcqa_nurse
+  - kormedmcqa_pharm
+aggregate_metric_list:
+  - metric: exact_match
+    aggregation: mean
+    weight_by_size: true
+metadata:
+  version: 0.0

lm_eval/tasks/kormedmcqa/kormedmcqa_doctor.yaml

-group: kormedmcqa
 task : kormedmcqa_doctor
 dataset_path : sean0042/KorMedMCQA
 dataset_name : doctor

lm_eval/tasks/kormedmcqa/kormedmcqa_nurse.yaml

-group: kormedmcqa
 task : kormedmcqa_nurse
 dataset_path : sean0042/KorMedMCQA
 dataset_name : nurse

lm_eval/tasks/kormedmcqa/kormedmcqa_pharm.yaml

-group: kormedmcqa
 task : kormedmcqa_pharm
 dataset_path : sean0042/KorMedMCQA
 dataset_name : pharm

lm_eval/tasks/lambada/lambada_openai.yaml

-group:
+tag:
   - lambada
 task: lambada_openai
 dataset_path: EleutherAI/lambada_openai

lm_eval/tasks/lambada/lambada_standard.yaml

-group:
+tag:
   - lambada
 task: lambada_standard
 dataset_path: lambada

lm_eval/tasks/lambada_cloze/lambada_openai_cloze.yaml

-group:
+tag:
   - lambada_cloze
 task: lambada_openai_cloze_yaml
 dataset_path: EleutherAI/lambada_openai

lm_eval/tasks/lambada_cloze/lambada_standard_cloze.yaml

-group:
+tag:
   - lambada_cloze
 task: lambada_standard_cloze_yaml
 dataset_path: lambada

lm_eval/tasks/lambada_multilingual/lambada_mt_en.yaml

-group:
+tag:
   - lambada_multilingual
 task: lambada_openai_mt_en
 dataset_path: EleutherAI/lambada_openai

lm_eval/tasks/mmlu/continuation/_mmlu.yaml

@@ -4,28 +4,28 @@ task:
   - group: stem
     task:
       - mmlu_continuation_stem
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: other
     task:
       - mmlu_continuation_other
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: social sciences
     task:
       - mmlu_continuation_social_sciences
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: humanities
     task:
       - mmlu_continuation_humanities
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
-aggregate_metric:
+aggregate_metric_list:
   - metric: acc
     weight_by_size: True
 metadata:

lm_eval/tasks/mmlu/default/_mmlu.yaml

@@ -4,7 +4,7 @@ task:
   - mmlu_other
   - mmlu_social_sciences
   - mmlu_humanities
-aggregate_metric:
+aggregate_metric_list:
   - metric: acc
     weight_by_size: True
 metadata:

lm_eval/tasks/mmlu/default/_mmlu_humanities.yaml

@@ -2,7 +2,7 @@ group: mmlu_humanities
 group_alias: humanities
 task:
   - mmlu_humanities_tasks
-aggregate_metric:
+aggregate_metric_list:
   - metric: acc
     weight_by_size: True
 metadata:

lm_eval/tasks/mmlu/default/_mmlu_other.yaml

@@ -2,7 +2,7 @@ group: mmlu_other
 group_alias: other
 task:
   - mmlu_other_tasks
-aggregate_metric:
+aggregate_metric_list:
   - metric: acc
     weight_by_size: True
 metadata:

lm_eval/tasks/mmlu/default/_mmlu_social_sciences.yaml

@@ -2,7 +2,7 @@ group: mmlu_social_sciences
 group_alias: social sciences
 task:
   - mmlu_social_sciences_tasks
-aggregate_metric:
+aggregate_metric_list:
   - metric: acc
     weight_by_size: True
 metadata:

lm_eval/tasks/mmlu/default/_mmlu_stem.yaml

@@ -2,7 +2,7 @@ group: mmlu_stem
 group_alias: stem
 task:
   - mmlu_stem_tasks
-aggregate_metric:
+aggregate_metric_list:
   - metric: acc
     weight_by_size: True
 metadata:

lm_eval/tasks/mmlu/flan_cot_fewshot/_mmlu.yaml

@@ -4,28 +4,28 @@ task:
   - group: stem
     task:
       - mmlu_flan_cot_fewshot_stem
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: other
     task:
       - mmlu_flan_cot_fewshot_other
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: social sciences
     task:
       - mmlu_flan_cot_fewshot_social_sciences
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: humanities
     task:
       - mmlu_flan_cot_fewshot_humanities
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
-aggregate_metric:
+aggregate_metric_list:
   - metric: acc
     weight_by_size: True
 metadata:

lm_eval/tasks/mmlu/flan_cot_zeroshot/_mmlu.yaml

@@ -4,28 +4,28 @@ task:
   - group: stem
     task:
       - mmlu_flan_cot_zeroshot_stem
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: other
     task:
       - mmlu_flan_cot_zeroshot_other
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: social sciences
     task:
       - mmlu_flan_cot_zeroshot_social_sciences
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: humanities
     task:
       - mmlu_flan_cot_zeroshot_humanities
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
-aggregate_metric:
+aggregate_metric_list:
   - metric: acc
     weight_by_size: True
 metadata:

lm_eval/tasks/mmlu/flan_n_shot/generative/_mmlu.yaml

@@ -4,28 +4,28 @@ task:
   - group: stem
     task:
       - mmlu_flan_n_shot_generative_stem
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: other
     task:
       - mmlu_flan_n_shot_generative_other
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: social sciences
     task:
       - mmlu_flan_n_shot_generative_social_sciences
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: humanities
     task:
       - mmlu_flan_n_shot_generative_humanities
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
-aggregate_metric:
+aggregate_metric_list:
   - metric: acc
     weight_by_size: True
 metadata: