Fix naming convention to avoid `pytest` name mangling invocation

lm_eval/base.py

@@ -61,7 +61,7 @@ class LM(abc.ABC):
 class Task(abc.ABC):
     def __init__(self):
         self.download()
-        self.__training_docs = None
+        self._training_docs = None
 
     def download(self):
         """Downloads the task dataset if necessary"""
@@ -104,9 +104,9 @@ class Task(abc.ABC):
         return []
 
     def fewshot_examples(self, k):
-        if self.__training_docs is None:
-            self.__training_docs = list(self.training_docs())
-        return random.sample(self.__training_docs, k)
+        if self._training_docs is None:
+            self._training_docs = list(self.training_docs())
+        return random.sample(self._training_docs, k)
 
     @abc.abstractmethod
     def doc_to_text(self, doc):

lm_eval/tasks/anli.py

@@ -18,9 +18,9 @@ class ANLIBase(HFTask):
 
     def training_docs(self):
         if self.has_training_docs():
-            if self.__training_docs is None:
-                self.__training_docs = list(self.data["train_r" + str(self.SPLIT)])
-            return self.__training_docs
+            if self._training_docs is None:
+                self._training_docs = list(self.data["train_r" + str(self.SPLIT)])
+            return self._training_docs
 
     def validation_docs(self):
         if self.has_validation_docs():

lm_eval/tasks/common.py

@@ -30,9 +30,9 @@ class HFTask(Task):
         # Cache training for faster few-shot.
         # If data is too large to fit in memory, override this method.
         if self.has_training_docs():
-            if self.__training_docs is None:
-                self.__training_docs = list(self.data["train"])
-            return self.__training_docs
+            if self._training_docs is None:
+                self._training_docs = list(self.data["train"])
+            return self._training_docs
 
     def validation_docs(self):
         if self.has_validation_docs():

lm_eval/tasks/naturalqs.py

@@ -30,10 +30,10 @@ class NaturalQs(HFTask):
 
     def fewshot_examples(self, k):
         # Data is too large to fit in memory. We just sample from the first bit.
-        if self.__training_docs is None:
-            self.__training_docs = list(islice(self.training_docs(), 0, 100000))
+        if self._training_docs is None:
+            self._training_docs = list(islice(self.training_docs(), 0, 100000))
 
-        return random.sample(self.__training_docs, k)
+        return random.sample(self._training_docs, k)
 
     def doc_to_text(self, doc):
         return 'Q: ' + doc['question']['text'] + '\n\n' + 'A: '

lm_eval/tasks/openbookqa.py

@@ -19,9 +19,9 @@ class OpenBookQA(HFTask):
 
     def training_docs(self):
         if self.has_training_docs():
-            if self.__training_docs is None:
-                self.__training_docs = list(self.data["train"])
-            return self.__training_docs
+            if self._training_docs is None:
+                self._training_docs = list(self.data["train"])
+            return self._training_docs
 
     def validation_docs(self):
         if self.has_validation_docs():

lm_eval/tasks/superglue.py

@@ -273,17 +273,17 @@ class ReCoRD(HFTask):
         # Hence, we one "doc" for each (context + passage, answer) pair.
         # Moreover, we only use the correct answers for context packing
         # (This is not an issue for evaluation, where we can directly score multiple candidates at once).
-        if self.__training_docs is None:
-            self.__training_docs = []
+        if self._training_docs is None:
+            self._training_docs = []
             for doc in self.data["train"]:
                 for entity in list(set(doc["entities"])):
-                    self.__training_docs.append({
+                    self._training_docs.append({
                         "passage": doc["passage"],
                         "query": doc["query"],
                         "entity": entity,
                         "label": entity in doc["answers"],
                     })
-        return self.__training_docs
+        return self._training_docs
 
     def validation_docs(self):
         for doc in self.data["validation"]:
@@ -417,14 +417,14 @@ class SGWinogradSchemaChallenge(HFTask):
 
     def training_docs(self):
         if self.has_training_docs():
-            if self.__training_docs is None:
+            if self._training_docs is None:
                 # GPT-3 Paper's format only uses positive examples for fewshot "training"
-                self.__training_docs = [
+                self._training_docs = [
                     doc for doc in
                     self.data["train"]
                     if doc["label"]
                 ]
-            return self.__training_docs
+            return self._training_docs
 
     def fewshot_description(self):
         return "Final Exam with Answer Key\n" \