add many explicit group configs

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

@@ -4,28 +4,28 @@ task:
   - group: stem
     task:
       - mmlu_flan_n_shot_loglikelihood_stem
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: other
     task:
       - mmlu_flan_n_shot_loglikelihood_other
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: social sciences
     task:
       - mmlu_flan_n_shot_loglikelihood_social_sciences
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: humanities
     task:
       - mmlu_flan_n_shot_loglikelihood_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/generative/_mmlu.yaml

@@ -4,28 +4,28 @@ task:
   - group: stem
     task:
       - mmlu_stem_generative
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: other
     task:
       - mmlu_other_generative
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: social sciences
     task:
       - mmlu_social_sciences_generative
-    aggregate_metric:
+    aggregate_metric_list:
       - metric: acc
         weight_by_size: True
   - group: humanities
     task:
       - mmlu_humanities_generative
-    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/okapi/arc_multilingual/_arc_yaml

-group:
+tag:
   - arc_multilingual
 dataset_path: null
 dataset_name: null

lm_eval/tasks/okapi/hellaswag_multilingual/_hellaswag_yaml

-group:
+tag:
   - hellaswag_multilingual
 dataset_path: null
 dataset_name: null

lm_eval/tasks/okapi/mmlu_multilingual/_default_yaml

-group:
+tag:
   - m_mmlu
 dataset_path: alexandrainst/m_mmlu
 test_split: test

lm_eval/tasks/okapi/truthfulqa_multilingual/_truthfulqa_mc1_yaml

-group:
+tag:
   - truthfulqa_multilingual
 dataset_path: null
 dataset_name: null

lm_eval/tasks/paws-x/_pawsx.yaml

+group: pawsx
+task:
+  - paws_en
+  - paws_de
+  - paws_es
+  - paws_fr
+  - paws_ja
+  - paws_ko
+  - paws_zh
+aggregate_metric_list:
+  - metric: acc
+    aggregation: mean
+    weight_by_size: true
+metadata:
+  version: 0.0

lm_eval/tasks/paws-x/pawsx_template_yaml

 # This file will be included in the generated language-specific task configs.
 # It doesn't have a yaml file extension as it is not meant to be imported directly
 # by the harness.
-group: pawsx
 task: null
 dataset_path: paws-x
 dataset_name: null

lm_eval/tasks/qa4mre/qa4mre_2011.yaml

-group:
+tag:
   - qa4mre
 task: qa4mre_2011
 dataset_path: qa4mre