New arabicmmlu (#2541)

* point to the original ArabicMMLU dataset

* create the new subtasks files

* fix bug when the context filed is empty

lm_eval/tasks/arabicmmlu/arabicmmlu_univ_accounting.yaml

-"dataset_name": "Univ Accounting"
-"tag": "arabicmmlu_social_science_tasks"
-"include": "_default_arabicmmlu_template_yaml"
-"task": "arabicmmlu_univ_accounting"
-"task_alias": "Univ Accounting"

lm_eval/tasks/arabicmmlu/arabicmmlu_univ_computer_science.yaml

-"dataset_name": "Univ Computer Science"
-"tag": "arabicmmlu_stem_tasks"
-"include": "_default_arabicmmlu_template_yaml"
-"task": "arabicmmlu_univ_computer_science"
-"task_alias": "Univ Computer Science"

lm_eval/tasks/arabicmmlu/arabicmmlu_univ_economics.yaml

-"dataset_name": "Univ Economics"
-"tag": "arabicmmlu_social_science_tasks"
-"include": "_default_arabicmmlu_template_yaml"
-"task": "arabicmmlu_univ_economics"
-"task_alias": "Univ Economics"

lm_eval/tasks/arabicmmlu/arabicmmlu_univ_political_science.yaml

-"dataset_name": "Univ Political Science"
-"tag": "arabicmmlu_social_science_tasks"
-"include": "_default_arabicmmlu_template_yaml"
-"task": "arabicmmlu_univ_political_science"
-"task_alias": "Univ Political Science"

lm_eval/tasks/arabicmmlu/utils.py

@@ -23,7 +23,7 @@ def doc_to_text(doc):
 
     question = (
         doc["Question"]
-        if doc["Context"] == ""
+        if not doc["Context"]
         else f"{doc['Context']}\n\n{doc['Question']}"
     )