resolved merge conflict

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_fra.yaml

 dataset_name: fra
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_fra

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_hau.yaml

 dataset_name: hau
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_hau

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_ibo.yaml

 dataset_name: ibo
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_ibo

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_kin.yaml

 dataset_name: kin
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_kin

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_lin.yaml

 dataset_name: lin
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_lin

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_lug.yaml

 dataset_name: lug
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_lug

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_orm.yaml

 dataset_name: orm
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_orm

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_sna.yaml

 dataset_name: sna
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_sna

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_sot.yaml

 dataset_name: sot
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_sot

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_swa.yaml

 dataset_name: swa
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_swa

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_twi.yaml

 dataset_name: twi
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_twi

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_wol.yaml

 dataset_name: wol
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_wol

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_xho.yaml

 dataset_name: xho
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_xho

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_yor.yaml

 dataset_name: yor
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_yor

lm_eval/tasks/afrimmlu/translate/afrimmlu_translate_zul.yaml

 dataset_name: zul
+<<<<<<< HEAD
 include: _translate_yaml
+=======
+include: afrimmlu_common_translate_yaml
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8
 task: afrimmlu_translate_zul

lm_eval/tasks/afrimmlu/translate/utils.py

-from sklearn.metrics import f1_score
+from lm_eval.utils import weighted_f1_score
 
 
 def doc_to_choice(doc):
@@ -30,6 +30,7 @@ def doc_to_text(doc):
         choice4=choices[3],
     )
     return text
+<<<<<<< HEAD
 
 
 def weighted_f1_score(items):
@@ -38,3 +39,5 @@ def weighted_f1_score(items):
     preds = unzipped_list[1]
     fscore = f1_score(golds, preds, average="weighted")
     return fscore
+=======
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8

lm_eval/tasks/afrimmlu/utils.py

-from sklearn.metrics import f1_score
+from lm_eval.utils import weighted_f1_score
 
 
 def doc_to_choice(doc):
@@ -30,6 +30,7 @@ def doc_to_text(doc):
         choice4=choices[3],
     )
     return text
+<<<<<<< HEAD
 
 
 def weighted_f1_score(items):
@@ -38,3 +39,5 @@ def weighted_f1_score(items):
     preds = unzipped_list[1]
     fscore = f1_score(golds, preds, average="weighted")
     return fscore
+=======
+>>>>>>> 0571eeb14d4e48aac51956a726c62cd8b382b3d8

lm_eval/tasks/afrixnli/_afrixnli.yaml

+group: afrixnli
+group_alias: AfriXNLI
+task:
+  - group: afrixnli_direct
+    group_alias: Prompt in African Languages
+    task:
+      - afrixnli_native_direct
+  - afrixnli_native_direct_eng_only
+  - group: afrixnli_lai_prompt_translate
+    group_alias: Translate-Test
+    task:
+      - afrixnli_lai_prompt_translate
\ No newline at end of file

lm_eval/tasks/afrixnli/_afrixnli_ablations.yaml

+group: afrixnli
+group_alias: AfriXNLI
+task:
+  - group: afrixnli_direct
+    group_alias: Prompt in African Languages
+    task:
+      - afrixnli_native_direct
+  - afrixnli_native_direct_eng_only
+  - group: afrixnli_translate
+    group_alias: Translate-Test
+    task:
+      - afrixnli_translate
\ No newline at end of file

lm_eval/tasks/afrixnli/anli_prompt/en-direct/utils.py

-from sklearn.metrics import f1_score
+from lm_eval.utils import weighted_f1_score
 
 
 def doc_to_target(doc):
     replacements = {0: "True", 1: "Neither", 2: "False"}
     return replacements[doc["label"]]
-
-
-def weighted_f1_score(items):
-    unzipped_list = list(zip(*items))
-    golds = unzipped_list[0]
-    preds = unzipped_list[1]
-    fscore = f1_score(golds, preds, average="weighted")
-    return fscore