Merge branch 'master' into hellaswag

lm_eval/tasks/__init__.py

@@ -5,6 +5,8 @@ from . import race
 from . import webqs
 from . import anli
 from . import hellaswag
+from . import openbookqa
+from . import squad
 
 TASK_REGISTRY = {
     # GLUE
@@ -28,6 +30,8 @@ TASK_REGISTRY = {
     "arc_easy": arc.ARCEasy,
     "arc_challenge": arc.ARCChallenge,
     "hellaswag": hellaswag.HellaSwag,
+    "openbookqa": openbookqa.OpenBookQA,
+    "squad": squad.SQuAD,
     "race": race.RACE,
     "webqs": webqs.WebQs,
     "anli_r1": anli.ANLIRound1,

lm_eval/tasks/openbookqa.py

+import numpy as np
+from scipy.stats import pearsonr, spearmanr
+from sklearn.metrics import f1_score, matthews_corrcoef
+from tqdm import auto as tqdm_lib
+from . common import HFTask, simple_accuracy_metric, yesno
+
+class OpenBookQA(HFTask):
+    DATASET_PATH = "openbookqa"
+    DATASET_NAME = "main"
+
+    def has_training_docs(self):
+        return True
+
+    def has_validation_docs(self):
+        return True
+
+    def has_test_docs(self):
+        return True
+
+    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
+
+    def validation_docs(self):
+        if self.has_validation_docs():
+            return self.data["validation"]
+
+    def test_docs(self):
+        if self.has_test_docs():
+            return self.data["test"]
+
+    def fewshot_description(self):
+        return "Text of the question prompt\nText of the answer completion"
+
+    def doc_to_text(self, doc, include_target=True):
+        text = doc['question_stem'] + '\n'
+        if include_target:
+            letter_answer = doc['answerKey']
+            if letter_answer == 'A':
+                index = 0
+            elif letter_answer == 'B':
+                index = 1
+            elif letter_answer == 'C':
+                index = 2
+            elif letter_answer == 'D':
+                index = 3
+            else:
+                raise ValueError("OpenBookQA from HF datasets contained an invalid answer key")
+            text += doc['choices']['text'][index] + '.'
+        return text
+
+    def evaluate(self, docs, lm, provide_description, num_fewshot):
+        # TODO: Write evaluation function
+        raise NotImplementedError()

lm_eval/tasks/race.py

 from . common import HFTask
 from ..utils_stream import X, each, apply, join, filt, one
 import collections
-import nlp
+import datasets
 
 
 class RACE(HFTask):
@@ -26,7 +26,7 @@ class RACE(HFTask):
         # is shown that one document is made per passage.
 
         r = collections.defaultdict(list)
-        for item in nlp.load_dataset(path=self.DATASET_PATH, name=self.DATASET_NAME)[set]:
+        for item in datasets.load_dataset(path=self.DATASET_PATH, name=self.DATASET_NAME)[set]:
             r[item['article']].append(item)
         
         res = list(r.values() >> each(lambda x: {

lm_eval/tasks/squad.py

+import numpy as np
+from scipy.stats import pearsonr, spearmanr
+from sklearn.metrics import f1_score, matthews_corrcoef
+from tqdm import auto as tqdm_lib
+from . common import HFTask, simple_accuracy_metric, yesno
+
+class SQuAD(HFTask):
+    DATASET_PATH = "squad_v2"
+    DATASET_NAME = None
+
+    def has_training_docs(self):
+        return True
+
+    def has_validation_docs(self):
+        return True
+
+    def has_test_docs(self):
+        return False
+
+    def training_docs(self):
+        if self.has_training_docs():
+            return self.data["train"]
+
+    def validation_docs(self):
+        if self.has_validation_docs():
+            return self.data["validation"]
+
+    def fewshot_description(self):
+        return "Title: The_Title_of_It\n\nBackground: A text passage as background to answer the question with.\n\nQ: Question about the passage.\n\nA: Answer."
+
+    def doc_to_text(self, doc, include_target=True):
+        text = 'Title: ' + doc['title'] + '\n\n' + 'Background: ' + doc['context'] + '\n\n' + 'Q: ' + doc['question'] + '\n\n' + 'A: '
+        if include_target:
+            answer_list = doc['answers']['text']
+            if len(answer_list) > 0:
+                answer = answer_list[0]
+            else:
+                answer = 'unanswerable'
+            text += answer
+        return text
+
+    def evaluate(self, docs, lm, provide_description, num_fewshot):
+        # TODO: Write evaluation function
+        raise NotImplementedError()
\ No newline at end of file

lm_eval/tasks/superglue.py

@@ -94,7 +94,11 @@ class Copa(HFTask):
 
     def doc_to_text(self, doc, include_target=True):
         # Drop the period
-        text = doc["premise"].strip()[:-1] + " because "
+        connector = {
+            "cause": "because",
+            "effect": "therefore",
+        }[doc["question"]]
+        text = doc["premise"].strip()[:-1] + f" {connector} "
         if include_target:
             correct_choice = doc["choice1"] if doc["label"] == 0 else doc["choice2"]
             # Connect the sentences