Merge remote-tracking branch 'origin/master' into fazz/refactor-task-coqa

lm_eval/base.py

@@ -2,6 +2,7 @@ import abc
 import random
 import numpy as np
 import sklearn
+import math
 
 
 class LM(abc.ABC):
@@ -58,10 +59,10 @@ class LM(abc.ABC):
         return cls()
 
 
-class Dataset(abc.ABC):
+class Task(abc.ABC):
     def __init__(self):
         self.download()
-        self._traindocs = None
+        self._training_docs = None
 
     def download(self):
         """Downloads the task dataset if necessary"""
@@ -84,23 +85,29 @@ class Dataset(abc.ABC):
 
     def training_docs(self):
         """
-
         :return: Iterable[obj]
             A iterable of any object, that doc_to_text can handle
         """
         return []
 
     def validation_docs(self):
+        """
+        :return: Iterable[obj]
+            A iterable of any object, that doc_to_text can handle
+        """
         return []
 
     def test_docs(self):
+        """
+        :return: Iterable[obj]
+            A iterable of any object, that doc_to_text can handle
+        """
         return []
 
     def fewshot_examples(self, k):
-        if self._traindocs is None:
-            self._traindocs = list(self.training_docs())
-
-        return random.sample(self._traindocs, 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):
@@ -193,7 +200,8 @@ def f1_score(items):
     golds = unzipped_list[0]
     preds = unzipped_list[1]
     fscore = sklearn.metrics.f1_score(golds, preds)
-    return max(fscore)
+
+    return np.max(fscore)
 
 
 def acc_all(items):
@@ -223,6 +231,9 @@ def metric_max_over_ground_truths(metric_fn, prediction, ground_truths):
     return max(scores_for_ground_truths)
 
 
+def perplexity(items):
+    return math.exp(-mean(items))
+
 req_ret_lens = {
     'loglikelihood': 2
 }

lm_eval/evaluator.py

+import collections
+import itertools
+
+
+def evaluate(lm, task_dict, provide_description, num_fewshot, limit):
+    # TODO: completely refactor this entire function to not be a huge mess, ideally breaking it down into smaller pieces
+
+    task_dict_items = [(name, task) for name, task in task_dict.items() if(task.has_validation_docs() or task.has_test_docs())]
+
+    results = collections.defaultdict(dict)
+
+    requests = collections.defaultdict(list)
+    requests_origin = collections.defaultdict(list)
+
+    # if we ever run into issues where the eval tasks don't fit in memory and we can't afford a machine with bigger memory,
+    # we can always modify this plumbing to support that, but i didn't want to include it just yet because overengineering is bad
+    # (or we could make it write the requests to disk and then read them back out again - probably using an sqlite db because of all the moving parts we have
+
+    # TODO: we need unit tests & sanity checks or something to ensure that the return of `validation_docs` is stable
+
+    docs = {}
+
+    # get lists of each type of requeste
+    for task_name, task in task_dict_items:
+        #default to validation doc, fall back to test doc if validation unavailable
+        # TODO: the val-fallback-to-test system isn't final, we should revisit it at some point
+        if task.has_validation_docs():
+            task_doc_func = task.validation_docs
+        elif task.has_test_docs():
+            task_doc_func = task.test_docs
+
+        for doc_id, doc in enumerate(itertools.islice(task_doc_func(), 0, limit)):
+            docs[(task_name, doc_id)] = doc
+
+            ctx = task.fewshot_context(
+                doc=doc,
+                provide_description=provide_description,
+                num_fewshot=num_fewshot,
+            )
+
+            reqs = task.construct_requests(doc, ctx)
+
+            for i, req in enumerate(reqs):
+                requests[req.type].append(req)
+                # i: index in requests for a single task instance
+                # doc_id: unique id that we can get back to a doc using `docs`
+                requests_origin[req.type].append((i, task_name, doc, doc_id))
+
+    # all responses for each (task, doc)
+    process_res_queue = collections.defaultdict(list)
+
+    # execute each type of request
+    for reqtype, reqs in requests.items():
+        # TODO: right now, this code runs multiple seperate LM requests for multiple Requests differing
+        # only in index. We could implement some kind of caching, but that would be more of a bandaid
+        # solution. we could also implement some kind of autogrouping here; they should end up next to each other.
+
+        resps = getattr(lm, reqtype)([req.args for req in reqs])
+
+        resps = [x if req.index is None else x[req.index] for x, req in zip(resps, reqs)]
+
+        for resp, (i, task_name, doc, doc_id) in zip(resps, requests_origin[reqtype]):
+            process_res_queue[(task_name, doc_id)].append((i, resp))
+    
+    vals = collections.defaultdict(list)
+
+    # unpack results and sort back in order and return control to Task
+    for (task_name, doc_id), requests in process_res_queue.items():
+        requests.sort(key=lambda x: x[0])
+        requests = [x[1] for x in requests]
+
+        task = task_dict[task_name]
+        doc = docs[(task_name, doc_id)]
+
+        metrics = task.process_results(doc, requests)
+        for metric, value in metrics.items():
+            vals[(task_name, metric)].append(value)
+    
+    # aggregate results
+    for (task_name, metric), items in vals.items():
+        task = task_dict[task_name]
+        results[task_name][metric] = task.aggregation()[metric](items)
+    
+    return results
\ No newline at end of file

lm_eval/models/__init__.py

 from . import gpt2
 from . import gpt3
+from . import dummy
 
 MODEL_REGISTRY = {
     "gpt2": gpt2.GPT2LM,
     "gpt3": gpt3.GPT3LM,
+    "dummy": dummy.DummyLM,
 }
 
 

lm_eval/models/gpt3.py

-# REMINDER: this code needs to be rewritten for the new framework. Remove this comment when the code is fully converted.
-
 import os
 import transformers
 from lm_eval.base import LM
 from lm_eval import utils
+from tqdm import tqdm
+
+
+def get_result(response, ctxlen):
+    is_greedy = True
+    logprobs = response["logprobs"]["token_logprobs"]
+    continuation_logprobs = sum(logprobs[ctxlen:])
+
+    for i in range(ctxlen, len(response["logprobs"]["tokens"])):
+        token = response["logprobs"]["tokens"][i]
+        top_tokens = response["logprobs"]["top_logprobs"][i]
+        top_token = max(top_tokens.keys(), key=lambda x: top_tokens[x])
+        if top_token != token:
+            is_greedy = False
+            break
+    
+    return continuation_logprobs, is_greedy
 
 
 class GPT3LM(LM):
 
     MAX_LENGTH = 2048
+    REQ_CHUNK_SIZE = 64
+    MAX_GEN_TOKS = 256
 
     def __init__(self, engine, truncate=False):
         """
@@ -31,23 +48,52 @@ class GPT3LM(LM):
         args = utils.simple_parse_args_string(arg_string)
         return cls(engine=args.get("engine", "davinci"))
 
-    def loglikelihood(self, context, continuation):
-        # TODO: implement new framework
-        
+    def loglikelihood(self, requests):
         import openai
-        
+        for chunk in tqdm(utils.chunks(requests, self.REQ_CHUNK_SIZE)):
+            inps = []
+            ctxlens = []
+            for context, continuation in chunk:
+                print(context)
                 context_enc = self.tokenizer.encode(context)
                 continuation_enc = self.tokenizer.encode(continuation)
-        inp = (context_enc + continuation_enc)[-1024:]
-        ctxlen = len(context_enc) - max(0, len(context_enc) + len(continuation_enc) - 1024)
+                inp = (context_enc + continuation_enc)[-self.MAX_LENGTH:]
+                ctxlen = len(context_enc) - max(0, len(context_enc) + len(continuation_enc) - self.MAX_LENGTH)
+
+                inps.append(inp)
+                ctxlens.append(ctxlen)
 
             response = openai.Completion.create(
                 engine=self.engine,
-            prompt=inp,
+                prompt=inps,
                 echo=True,
-            max_tokens=0, temperature=0.0,
-            logprobs=0,
+                max_tokens=0, temperature=0.,
+                logprobs=10,
             )
-        logprobs = response.choices[0]["logprobs"]["token_logprobs"]
-        continuation_logprobs = logprobs[ctxlen:]
-        return sum(continuation_logprobs)
+
+            for resp, ctxlen in zip(response.choices, ctxlens):
+                res.append(get_result(resp, ctxlen))
+            
+        return res
+
+    def greedy_until(self, requests):
+        import openai
+        res = []
+
+        for context, until in tqdm(requests):
+            context_enc = self.tokenizer.encode(context)
+            inp = context_enc[-(self.MAX_LENGTH - self.MAX_GEN_TOKS):]
+            ctxlen = len(context_enc) - max(0, len(context_enc) - (self.MAX_LENGTH - self.MAX_GEN_TOKS))
+
+            response = openai.Completion.create(
+                engine=self.engine,
+                prompt=[inp],
+                max_tokens=self.MAX_GEN_TOKS, 
+                temperature=0.,
+                logprobs=10,
+            )
+
+            res.append(response.choices[0]['text'])
+        
+        return res
+

lm_eval/tasks/__init__.py

@@ -18,6 +18,7 @@ from . import lambada
 from . import race 
 from . import piqa
 from . import triviaqa
+from . import webqs
 
 
 TASK_REGISTRY = {
@@ -37,7 +38,7 @@ TASK_REGISTRY = {
     "cb": superglue.CommitmentBank,
     "copa": superglue.Copa,
     "multirc": superglue.MultiRC,
-    "record": superglue.ReCoRD,
+    #"record": superglue.ReCoRD,
     "wic": superglue.WordsInContext,
     "wsc": superglue.SGWinogradSchemaChallenge,
     
@@ -56,9 +57,9 @@ TASK_REGISTRY = {
     # "squad": squad.SQuAD, # not implemented yet
     "race": race.RACE,
     # "naturalqs": naturalqs.NaturalQs, # not implemented yet
-    # "webqs": webqs.WebQs, # not implemented yet
-    # "wsc273": wsc273.WinogradSchemaChallenge273, # not implemented yet
-    # "winogrande": winogrande.Winogrande, # not implemented yet
+    "webqs": webqs.WebQs,
+    "wsc273": wsc273.WinogradSchemaChallenge273,
+    "winogrande": winogrande.Winogrande,
     "anli_r1": anli.ANLIRound1,
     "anli_r2": anli.ANLIRound2,
     "anli_r3": anli.ANLIRound3,

lm_eval/tasks/arithmetic.py

@@ -2,12 +2,12 @@ import abc
 import json
 import os
 from collections import namedtuple
-from lm_eval.base import Dataset, mean, rf
+from lm_eval.base import Task, mean, rf
 from best_download import download_file
 
 ArithmeticDoc = namedtuple('ArithmeticDoc', ['context', 'completion'])
 
-class Arithmetic(Dataset):
+class Arithmetic(Task):
     directory = 'data/arithmetic/'
 
     def __init__(self):

lm_eval/tasks/common.py

 import datasets
 import numpy as np
-import random
-from ..base import Dataset
+from ..base import Task
 
 
-class HFTask(Dataset):
+class HFTask(Task):
     DATASET_PATH = None
     DATASET_NAME = None
 
     def __init__(self):
+        self.data = None
         super().__init__()
-        self._training_docs = None
 
     def download(self):
         self.data = datasets.load_dataset(path=self.DATASET_PATH, name=self.DATASET_NAME)

lm_eval/tasks/drop.py

@@ -5,9 +5,9 @@ from sklearn.metrics import f1_score, matthews_corrcoef
 from tqdm import auto as tqdm_lib
 from . common import HFTask, simple_accuracy_metric, yesno
 from pathlib import Path
-from ..base import Dataset
+from ..base import Task
 
-class DROP(Dataset):
+class DROP(Task):
     DATAFOLDER = Path(__file__).parent / "../../data/drop"
     
     def __init__(self):

lm_eval/tasks/lambada.py

-from lm_eval.base import Dataset, rf, mean
+from lm_eval.base import Task, rf, mean, perplexity
 from lm_eval.utils import sh
 import json
 import math
 from best_download import download_file
 
 
-class LAMBADA(Dataset):
+class LAMBADA(Task):
     def download(self):
         sh("mkdir -p data/lambada")
         download_file(
@@ -45,7 +45,7 @@ class LAMBADA(Dataset):
         return ""
 
     def construct_requests(self, doc, ctx):
-        ll, is_greedy = rf.loglikelihood(doc, self.doc_to_target(doc))
+        ll, is_greedy = rf.loglikelihood(ctx, self.doc_to_target(doc))
 
         return ll, is_greedy
     
@@ -53,13 +53,13 @@ class LAMBADA(Dataset):
         ll, is_greedy = results
 
         return {
-            'perplexity': math.exp(-ll),
+            'perplexity': ll,
             'accuracy': int(is_greedy)
         }
         
     def aggregation(self):
         return {
-            'perplexity': mean,
+            'perplexity': perplexity,
             'accuracy': mean
         }
 

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._traindocs is None:
-            self._traindocs = 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._traindocs, 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/piqa.py

 import json
 import random
-from lm_eval.base import Dataset, rf, mean
+from lm_eval.base import Task, rf, mean
 from ..utils import sh
 import os
 
-class PiQA(Dataset):
+class PiQA(Task):
     def download(self):
         if not os.path.exists('data/piqa'):
             #TODO: use best_download

lm_eval/tasks/quac.py

 import json
 import random
 import os
-from lm_eval.base import Dataset
+from lm_eval.base import Task
 from ..utils import sh
 
 
-class QuAC(Dataset):    
+class QuAC(Task):    
     def __init__(self):
         super().__init__()
 

lm_eval/tasks/race.py

@@ -3,7 +3,19 @@ import datasets
 import numpy as np
 from lm_eval.base import rf, mean
 from . common import HFTask
-from ..utils_stream import each
+
+import os
+from functools import reduce
+import operator
+from tqdm import tqdm
+import json
+
+class each:
+    def __init__(self, f):
+        self.f = f
+
+    def __rrshift__(self, other):
+        return list(map(self.f, other))
 
 
 class RACE(HFTask):

lm_eval/tasks/sat.py

 import json
 import random
 import os
-from lm_eval.base import Dataset, rf, mean
+from lm_eval.base import Task, rf, mean
 from tqdm import auto as tqdm_lib
 from . common import simple_accuracy_metric
 import numpy as np
 from ..utils import sh
 
 
-class SATAnalogies(Dataset):    
+class SATAnalogies(Task):    
     NEEDS_MANUAL_DL = True
     
     def __init__(self):

lm_eval/tasks/storycloze.py

 import json
 import random
-from lm_eval.base import Dataset
+from lm_eval.base import Task
 from ..utils import sh
 import csv
 
-class StoryCloze(Dataset):
+class StoryCloze(Task):
     NEEDS_MANUAL_DL = True
 
     def download(self):

lm_eval/tasks/superglue.py

@@ -261,7 +261,7 @@ class ReCoRD(HFTask):
         return True
 
     def has_test_docs(self):
-        return True
+        return False
 
     def fewshot_description(self):
         # TODO: figure out actual description
@@ -322,6 +322,7 @@ class ReCoRD(HFTask):
         # - Evaluate the accuracy and token F1 PER EXAMPLE
         # - Average over all examples
         max_idx = np.argmax(np.array(results))
+
         prediction = doc["entities"][max_idx]
         gold_label_set = list(set(doc["answers"]))
         f1 = metric_max_over_ground_truths(squad_metrics.compute_f1, prediction, gold_label_set)

lm_eval/tasks/triviaqa.py

 import os
 import json
 import random
-from lm_eval.base import Dataset, mean, rf
+from lm_eval.base import Task, mean, rf
 from ..utils import sh
 
-class TriviaQA(Dataset):
+class TriviaQA(Task):
     def download(self):
         if not os.path.exists('data/triviaqa'):
             sh("""

lm_eval/tasks/webqs.py

 from . common import HFTask
+from lm_eval.base import mean, rf
 
 class WebQs(HFTask):
     DATASET_PATH = "web_questions"
@@ -18,7 +19,6 @@ class WebQs(HFTask):
         return ""
 
     def doc_to_text(self, doc):
-        print(doc)
         return "Q: " + doc['question'] + '\nA:'
 
     def doc_to_target(self, doc):
@@ -27,47 +27,36 @@ class WebQs(HFTask):
         # TODO: make sure we're actually handling multi-answer correctly
         return " " + doc['answers'][0]
         
-    def construct_requests(self, doc, ctx):
-        """ Uses RequestFactory to construct Requests and returns an iterable of 
-        Requests which will be sent to the LM.
+    def _remove_prefixes(self, aliases):
+        # Optimization: Remove any alias that has a strict prefix elsewhere in the list
+        # we can do this because if the prefix is acceptable by isgreedy, we can stop looking
+        aliases.sort()
+        ret = [aliases[0]]
+        for alias in aliases[1:]:
+            if not alias.startswith(ret[-1]):
+                ret.append(alias)
 
-        :param doc:
-            The document as returned from training_docs, validation_docs, or test_docs.
-        :param ctx: str
-            The context string, generated by fewshot_context. This includes the natural 
-            language description, as well as the few shot examples, and the question
-            part of the document for `doc`. 
-        """
-        # TODO: implement evaluation.
-        raise NotImplementedError('Evaluation not implemented')
+        return ret
         
-    def process_results(self, doc, results):
-        """Take a single document and the LM results and evaluates, returning a 
-        dict where keys are the names of submetrics and values are the values of 
-        the metric for that one document
 
-        :param doc:
-            The document as returned from training_docs, validation_docs, or test_docs.
-        :param results:
-            The results of the requests created in construct_requests.
-        """
-        # TODO: implement evaluation.
-        raise NotImplementedError('Evaluation not implemented')
+    def construct_requests(self, doc, ctx):
+        ret = []
+        for alias in self._remove_prefixes(doc['answers']):
+            _, is_prediction = rf.loglikelihood(ctx, " " + alias)
+            ret.append(is_prediction)
+        return ret
+
+    def process_results(self, doc, results):
+        return {
+            "acc": float(any(results))
+        }
 
     def aggregation(self):
-        """
-        :returns: {str: [float] -> float}
-            A dictionary where keys are the names of submetrics and values are 
-            functions that aggregate a list of metrics
-        """
-        # TODO: implement evaluation.
-        raise NotImplementedError('Evaluation not implemented')
+        return {
+            "acc": mean,
+        }
 
     def higher_is_better(self):
-        """
-        :returns: {str: bool}
-            A dictionary where keys are the names of submetrics and values are 
-            whether a higher value of the submetric is better
+        return {
+            "acc": True
+        }
\ No newline at end of file
-        """
-        # TODO: implement evaluation.
-        raise NotImplementedError('Evaluation not implemented')

lm_eval/tasks/winogrande.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
+from . common import HFTask
+from lm_eval.base import rf, mean
+
+"""
+This evaluation of Winogrande uses partial evaluation as described by
+Trinh & Le in Simple Method for Commonsense Reasoning (2018).
+Reference: https://arxiv.org/abs/1806.02847
+"""
+
 
 class Winogrande(HFTask):
     DATASET_PATH = "winogrande"
@@ -17,35 +22,31 @@ class Winogrande(HFTask):
     def has_test_docs(self):
         return True
 
-    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 test_docs(self):
-        if self.has_test_docs():
-            return self.data["test"]
-
     def fewshot_description(self):
         # TODO: redo description
         return "Winograd schema sentence including a either a ___ blank with a missing word, making the pronoun ambiguous, or the same with the word filled in."
 
+    @classmethod
+    def partial_context(cls, doc):
+        # Substitute the pronoun in the sentence with each candidate choice
+        # and ignore everything after.
+        pronoun_loc = doc["sentence"].index("_")
+        context1 = doc["sentence"][:pronoun_loc] + doc["option1"]
+        context2 = doc["sentence"][:pronoun_loc] + doc["option2"]
+        return context1, context2
+
+    @classmethod
+    def partial_target(cls, doc):
+        # The target is everything after the document specified pronoun.
+        pronoun_loc = doc["sentence"].index("_") + 1
+        return doc["sentence"][pronoun_loc:].strip()
+
     def doc_to_text(self, doc):
-        return doc['sentence']
+        context1, context2 = self.partial_context(doc)
+        return context1 + '\n' + context2 + '\n'
 
     def doc_to_target(self, doc):
-        text = doc['sentence']
-        answer_n = doc['answer']
-        if answer_n == '1':
-            answer = doc['option1']
-        elif answer_n == '2':
-            answer = doc['option2']
-        else:
-            raise ValueError("Winogrande from HF datasets contained an invalid answer key")
-        return text.replace("_", answer)
+        return self.partial_target(doc)
 
     def construct_requests(self, doc, ctx):
         """ Uses RequestFactory to construct Requests and returns an iterable of 
@@ -58,8 +59,11 @@ class Winogrande(HFTask):
             language description, as well as the few shot examples, and the question
             part of the document for `doc`. 
         """
-        # TODO: implement evaluation.
-        raise NotImplementedError('Evaluation not implemented')
+        target = self.partial_target(doc)
+        context1, context2 = self.partial_context(doc)
+        ll_context1, _ = rf.loglikelihood(context1, " " + target)
+        ll_context2, _ = rf.loglikelihood(context2, " " + target)
+        return ll_context1, ll_context2
 
     def process_results(self, doc, results):
         """Take a single document and the LM results and evaluates, returning a 
@@ -71,8 +75,10 @@ class Winogrande(HFTask):
         :param results:
             The results of the requests created in construct_requests.
         """
-        # TODO: implement evaluation.
-        raise NotImplementedError('Evaluation not implemented')
+        answer = int(doc["answer"]) - 1  # `- 1` b/c doc["answer"] ∈ {'1', '2'}
+        return {
+            "acc": np.argmax(results) == answer
+        }
 
     def aggregation(self):
         """
@@ -80,8 +86,9 @@ class Winogrande(HFTask):
             A dictionary where keys are the names of submetrics and values are 
             functions that aggregate a list of metrics
         """
-        # TODO: implement evaluation.
-        raise NotImplementedError('Evaluation not implemented')
+        return {
+            "acc": mean
+        }
 
     def higher_is_better(self):
         """
@@ -89,5 +96,6 @@ class Winogrande(HFTask):
             A dictionary where keys are the names of submetrics and values are 
             whether a higher value of the submetric is better
         """
-        # TODO: implement evaluation.
-        raise NotImplementedError('Evaluation not implemented')
+        return {
+            "acc": True
+        }

lm_eval/tasks/wsc273.py

-import json
+import numpy as np
 import random
-import os
-from lm_eval.base import Dataset
-from ..utils import sh
+from lm_eval.base import rf, mean
+from . common import HFTask
 
+"""
+NOTE: This evaluation of Winograd Schema Challenge is based on `partial evaluation`
+as described by Trinh & Le in Simple Method for Commonsense Reasoning (2018).
+See: https://arxiv.org/abs/1806.02847
+"""
+
+
+class WinogradSchemaChallenge273(HFTask):
+    DATASET_PATH = "winograd_wsc"
+    DATASET_NAME = "wsc273"
+
+    upper_pronouns = ["A", "An", "The", "She", "He",
+                      "It", "They", "My", "His", "Her", "Their"]
 
-class WinogradSchemaChallenge273(Dataset):    
     def __init__(self):
         super().__init__()
-
-    def download(self):
-        if not os.path.exists('data/wsc273'):
-            sh("""
-                mkdir -p data/wsc273 
-                wget https://git.cse.msu.edu/bakerb15/nlp-final-project/raw/master/Winogard/reproduce/commonsense_test/wsc273.json -O data/wsc273/wsc273.json
-                """)
+        self.data = self.__clean_data()
+
+    def __clean_data(self):
+        # The HF implementation of `wsc273` is not `partial evaluation` friendly.
+        data = []
+        for doc in self.data["test"]:
+            doc["text"] = doc["text"].replace("  ", " ")
+            doc["options"][0] = self.__normalize_option(doc["options"][0], doc)
+            doc["options"][1] = self.__normalize_option(doc["options"][1], doc)
+            data.append(doc)
+        return {"test": data}
+
+    def __normalize_option(self, option, doc):
+        # Append `'s` to possessive determiner based options.
+        if doc["pronoun"].lower() in ["my", "his", "her", "our", "their"]: 
+            option += "'s"
+        # Appropriately lowercase the pronoun in the option.
+        pronoun = option.split()[0]
+        start_of_sentence = doc["text"][doc['pronoun_loc'] - 2] == '.'
+        if not start_of_sentence and pronoun in self.upper_pronouns:
+            return option.replace(pronoun, pronoun.lower())
+        return option
 
     def has_training_docs(self):
         return False
@@ -25,60 +51,35 @@ class WinogradSchemaChallenge273(Dataset):
     def has_test_docs(self):
         return True
 
-    def training_docs(self):
-        return []
-
-    def validation_docs(self):
-        return []
-
-    def test_docs(self):
-        myjson = json.load(open('data/wsc273/wsc273.json'))
-        return self.load_doc(myjson)
+    def fewshot_examples(self, k):
+        # NOTE: `super().fewshot_examples` samples from training docs which are
+        # not available for this test-set-only dataset.
+        return random.sample(list(self.test_docs()), k)
 
     def fewshot_description(self):
         # TODO: redo description
         return "Winograd schema sentence with correct continuation. True. Winograd schema sentence with incorrect continuation. False."
 
-    def load_doc(self, myjson):
-        docs = []
-        for i in range(0, 273 * 2, 2):
-            item1 = myjson[i]
-            item2 = myjson[i+1]
-
-            if item1['question_id'] != item2['question_id']:
-                raise ValueError("WSC273 has missing completion pair.")
-
-            question_id = item1['question_id']
+    @classmethod
+    def partial_context(cls, doc):
+        # Substitute the pronoun in the original text with each candidate 
+        # choice and ignore everything after. 
+        context1 = doc["text"][:doc["pronoun_loc"]] + doc["options"][0]
+        context2 = doc["text"][:doc["pronoun_loc"]] + doc["options"][1]
+        return context1, context2
 
-            if item1['correctness'] == True:
-                doc = {
-                    'id': question_id,
-                    'completions': {
-                        'T': item1['substitution'],
-                        'F': item2['substitution'],
-                    },
-                }
-                
-            if item2['correctness'] == True:
-                doc = {
-                    'id': question_id,
-                    'completions': {
-                        'F': item1['substitution'],
-                        'T': item2['substitution'],
-                    },
-                }
-
-            docs.append(doc)
- 
-        return docs
+    @classmethod
+    def partial_target(cls, doc):
+        # The target is everything after the document specified pronoun.
+        start_index = doc["pronoun_loc"] + len(doc["pronoun"])
+        return doc["text"][start_index:].strip()
 
     def doc_to_text(self, doc):
-        # TODO: implement
-        pass
+        context1, context2 = self.partial_context(doc)
+        return context1 + '\n' + context2 + '\n'
 
     def doc_to_target(self, doc):
-        # TODO: implement
-        pass
+        return self.partial_target(doc)
 
     def construct_requests(self, doc, ctx):
         """ Uses RequestFactory to construct Requests and returns an iterable of 
@@ -91,8 +92,11 @@ class WinogradSchemaChallenge273(Dataset):
             language description, as well as the few shot examples, and the question
             part of the document for `doc`. 
         """
-        # TODO: implement evaluation.
-        raise NotImplementedError('Evaluation not implemented')
+        target = self.partial_target(doc)
+        context1, context2 = self.partial_context(doc)
+        ll_context1, _ = rf.loglikelihood(context1, " " + target)
+        ll_context2, _ = rf.loglikelihood(context2, " " + target)
+        return ll_context1, ll_context2
 
     def process_results(self, doc, results):
         """Take a single document and the LM results and evaluates, returning a 
@@ -104,8 +108,9 @@ class WinogradSchemaChallenge273(Dataset):
         :param results:
             The results of the requests created in construct_requests.
         """
-        # TODO: implement evaluation.
-        raise NotImplementedError('Evaluation not implemented')
+        return {
+            "acc": np.argmax(results) == doc["label"]
+        }
 
     def aggregation(self):
         """
@@ -113,8 +118,9 @@ class WinogradSchemaChallenge273(Dataset):
             A dictionary where keys are the names of submetrics and values are 
             functions that aggregate a list of metrics
         """
-        # TODO: implement evaluation.
-        raise NotImplementedError('Evaluation not implemented')
+        return {
+            "acc": mean
+        }
 
     def higher_is_better(self):
         """
@@ -122,5 +128,6 @@ class WinogradSchemaChallenge273(Dataset):
             A dictionary where keys are the names of submetrics and values are 
             whether a higher value of the submetric is better
         """
-        # TODO: implement evaluation.
-        raise NotImplementedError('Evaluation not implemented')
+        return {
+            "acc": True
+        }