diff --git a/elk/evaluation/evaluate.py b/elk/evaluation/evaluate.py
index 4b2c7b87..47d7788c 100644
--- a/elk/evaluation/evaluate.py
+++ b/elk/evaluation/evaluate.py
@@ -72,7 +72,7 @@ def evaluate_reporter(
 
 
 def evaluate_reporters(cfg: EvaluateConfig, out_dir: Optional[Path] = None):
-    if cfg.source == 'zero-shot':
+    if cfg.source == "zero-shot":
         evaluate_zeroshot(cfg, out_dir)
 
         return
@@ -145,12 +145,12 @@ def get_logprobs_tensor(split):
         # returns (num_examples, num_prompts, num_labels)
         logprobs_lst = []
         for ex in tqdm(ds[split]):
-            logprobs_lst.append(ex['logprobs'])
+            logprobs_lst.append(ex["logprobs"])
         logprobs = torch.tensor(logprobs_lst)
         return logprobs
 
-    train_logprobs = get_logprobs_tensor('train')
-    test_logprobs = get_logprobs_tensor('test')
+    train_logprobs = get_logprobs_tensor("train")
+    test_logprobs = get_logprobs_tensor("test")
 
     print(test_logprobs.shape)
 
@@ -174,7 +174,7 @@ def get_logprobs_tensor(split):
     zs_preds_cal = torch.argmax(test_logprobs_cal, dim=-1)
 
     # labels: (num_examples,) -> (num_examples, num_prompts)
-    labels = torch.tensor(ds['test']['label']).unsqueeze(1)
+    labels = torch.tensor(ds["test"]["label"]).unsqueeze(1)
     labels = labels.repeat(1, num_prompts)
 
     # accuracy calculation
@@ -183,16 +183,21 @@ def get_logprobs_tensor(split):
 
     # get aggregate truth probability
     # (1 - e^l_neg) * 0.5 + (e^l_pos) * 0.5
-    truth_prob = ((1 - torch.exp(test_logprobs_cal[:, :, 0])) + torch.exp(train_logprobs[:, :, 1])) * 0.5
+    truth_prob = (
+        (1 - torch.exp(test_logprobs_cal[:, :, 0])) + torch.exp(train_logprobs[:, :, 1])
+    ) * 0.5
 
     # auroc calculation
     # logprobs -> probs
     test_probs = torch.exp(test_logprobs)
     auroc = roc_auc_score(labels.flatten(), truth_prob.flatten())
 
-    print('Raw accuracy: %.4f\nCalibrated accuracy:%.4f\nAUROC: %.4f' % (raw_acc, cal_acc, auroc))    
+    print(
+        "Raw accuracy: %.4f\nCalibrated accuracy:%.4f\nAUROC: %.4f"
+        % (raw_acc, cal_acc, auroc)
+    )
 
-    cols = ['acc', 'cal_acc', 'auroc']
+    cols = ["acc", "cal_acc", "auroc"]
 
     with open(out_dir / "eval.csv", "w") as f:
         writer = csv.writer(f)