add lambada

lm_eval/tasks/lambada.py

+from lm_eval.base import Dataset
+from ..utils import sh
+import json
+import requests
+import ftfy
+
+
+class Lambada(Dataset):
+
+    def download(self):
+        sh("mkdir -p data/lambada")
+        with open("data/lambada/lambada_test.json", 'w') as f:
+            req = requests.get("https://storage.googleapis.com/gpt-2/data/lambada_test.jsonl")
+            req.raise_for_status()
+            jsons = [json.loads(l) for l in req.iter_lines()]
+            texts = [ftfy.fix_text(j['text'], normalization='NFKC') for j in jsons]
+            json.dump(texts, f)
+
+    def has_training_docs(self):
+        return False
+
+    def has_validation_docs(self):
+        return False
+
+    def has_test_docs(self):
+        return True
+
+    def training_docs(self):
+        pass
+
+    def validation_docs(self):
+        pass
+
+    def load_doc(self, myjson):
+        return [doc['text'] for doc in myjson]
+
+    def test_docs(self):
+        myjson = json.load(open("data/lambada/lambada_test.json"))
+        return self.load_doc(myjson)
+
+    def doc_to_text(self, doc, include_target=True):
+        pass
+
+    def evaluate(self, docs, lm, provide_description, num_fewshot):
+        pass
\ No newline at end of file