diff --git a/parlai/core/metrics.py b/parlai/core/metrics.py
index 1e79ec43c17..d2fa8681e47 100644
--- a/parlai/core/metrics.py
+++ b/parlai/core/metrics.py
@@ -535,7 +535,9 @@ def _prec_recall_f1_score(pred_items, gold_items):
         return precision, recall, f1
 
     @staticmethod
-    def compute(guess: str, answers: List[str]) -> F1Metric:
+    def compute(
+        guess: str, answers: List[str], expose_p_and_r: bool = False
+    ) -> Union[F1Metric, Tuple[F1Metric, F1Metric, F1Metric]]:
         if guess is None or answers is None:
             return AverageMetric(0, 0)
         g_tokens = normalize_answer(guess).split()
@@ -546,7 +548,10 @@ def compute(guess: str, answers: List[str]) -> F1Metric:
         max_p, max_r, max_f1 = 0, 0, 0
         for p, r, f1 in scores:
             max_p, max_r, max_f1 = max(max_p, p), max(max_r, r), max(f1, max_f1)
-        return (F1Metric(max_p, 1), F1Metric(max_r, 1), F1Metric(max_f1, 1))
+        if expose_p_and_r:
+            return (F1Metric(max_p, 1), F1Metric(max_r, 1), F1Metric(max_f1, 1))
+        else:
+            return F1Metric(max_f1, 1)
 
 
 class ExactMatchMetric(AverageMetric):
@@ -730,7 +735,9 @@ def compute_many(
         :return: (rouge-1, rouge-2, rouge-L)
         """
         measure = measure.lower()
-        assert measure in ROUGE_METRICS_MEASURES, "Use one of recall 'r' (default), f1 'f', or precision 'p'."
+        assert (
+            measure in ROUGE_METRICS_MEASURES
+        ), "Use one of recall 'r' (default), f1 'f', or precision 'p'."
 
         # possible global initialization
         try:
@@ -1031,7 +1038,9 @@ def evaluate_response(self, observation: Message, labels: List[str]) -> None:
 
         if prediction is not None:
             self.add('accuracy', ExactMatchMetric.compute(prediction, labels))
-            precision, recall, f1 = F1Metric.compute(prediction, labels)
+            precision, recall, f1 = F1Metric.compute(
+                prediction, labels, expose_p_and_r=True
+            )
             self.add('precision', precision)
             self.add('recall', recall)
             self.add('f1', f1)
diff --git a/tests/nightly/gpu/test_bb3.py b/tests/nightly/gpu/test_bb3.py
index f5c68ac980a..f1ff4d70943 100644
--- a/tests/nightly/gpu/test_bb3.py
+++ b/tests/nightly/gpu/test_bb3.py
@@ -19,6 +19,9 @@
 from projects.bb3.tests.opt_presets import INIT_OPT
 
 
+LOCAL = False
+
+
 def _self_memory(text):
     return f'{PROMPT.SELF_MEMORY_PREFIX}: {text}'
 
@@ -115,7 +118,7 @@ class TestOptFtBase(unittest.TestCase):
     def setUp(self):
         self.opt = INIT_OPT
         for k, v in self.opt.items():
-            if 'BB3OPTAgent' in v:
+            if 'BB3OPTAgent' in str(v):
                 self.opt[k] = 'projects.bb3.agents.opt_api_agent:MockOptAgent'
 
         self.opt['search_server'] = 'test'
@@ -340,8 +343,8 @@ def setUp(self):
         opt = copy.deepcopy(self.opt)
         overrides = {
             'knowledge_conditioning': 'combined',
-            'opt_server': 'http://18.117.126.138:3000',
-            'search_server': 'bing_cc',
+            'opt_server': 'http://localhost:6000',
+            'search_server': 'test',
         }
         for k, v in overrides.items():
             opt[k] = v
@@ -351,6 +354,7 @@ def setUp(self):
         self.batch_agent = create_agent(opt)
 
 
+@unittest.skipUnless(LOCAL, "must be local to specify opt server")
 class TestOptMainServerBatching(TestOptMainServerBase):
     def test_batching(self):
         self.batch_agent = create_agent(self.opt)