Merge branch 'master' of git@github.com:EleutherAI/lm_evaluation_harness.git

.gitignore

 env
-*.pyc
\ No newline at end of file
+*.pyc
+data/

README.md

@@ -45,3 +45,12 @@ With the data downloader in place, we simply need to (1) expose the val/test exa
 
 ### 3. Adding task training data to LM training set
 This part is the easiest. I guess we just write out some text files containing the training data? We can let the usual LM preprocessing pipeline handle it from there.
+=======
+## Summary (need to convert from google docs at some point):
+https://docs.google.com/document/d/177dwJpH8GHebISXYZSn4NL98sXdCtQMH82b7O5F7jmw/edit?usp=sharing
+
+## Current Datasets:
+
+[] CoQA 
+[] DROP
+

base.py

@@ -12,6 +12,14 @@ class LM(abc.ABC):
 
 
 class Dataset(abc.ABC):
+    @abc.abstractmethod
+    def has_training_docs(self):
+        pass
+    
+    @abc.abstractmethod
+    def has_validation_docs(self):
+        pass
+
     @abc.abstractmethod
     def training_docs(self):
         pass
@@ -42,4 +50,4 @@ class Dataset(abc.ABC):
     @abc.abstractmethod
     def evaluate(self, docs, lm, provide_description, num_fewshot):
         pass
-    
\ No newline at end of file
+    

datasets/coqa/evaluateCoQA.py → coqa.py

 
-from ...base import Dataset
+from base import Dataset
 import os
 import json
+import random
+
 class CoQA(Dataset):
+    def has_training_docs(self):
+        return True
+
+    def has_validation_docs(self):
+        return False
+    
     def training_docs(self):
-        pass
+        myjson = json.load(open('data/coqa/coqa-train-v1.0.json'))['data']
+        return self.load_doc(myjson)
 
     def validation_docs(self):
         pass
-    
+
     def test_docs(self):
-        pass
+        myjson = json.load(open('data/coqa/coqa-dev-v1.0.json'))['data']    
+        return self.load_doc(myjson)
     
     def fewshot_examples(self, k):
         traindocs = list(self.training_docs())
@@ -22,8 +32,22 @@ class CoQA(Dataset):
     def fewshot_description(self):
         pass
 
+    def load_doc(self, myjson):
+        docs = []
+        for item in myjson:
+            new_instance = [item['story']]
+            qa_pairs = zip(item['questions'], item['answers'])
+            for pair in qa_pairs:
+                new_instance.append('\n')
+                new_instance.append(''.join(['Q: ',pair[0]['input_text']]))
+                new_instance.append(''.join(['A: ',pair[1]['input_text']]))
+            docs.append(new_instance)  
+        return docs
+    
     def doc_to_text(self, doc, include_target=True):
-        json.load(open(doc))
+        text = '\n<|endoftext|>\n'.join(['\n'.join(instance) for instance in doc])
+        text = text + '\n<|endoftext|>'
+        return text
 
-        
- 
+    def evaluate(self, docs, lm):
+        pass