format

docs/new_task_guide.md

@@ -50,7 +50,7 @@ dataset_kwargs: null # any extra keyword arguments that should be passed to the
 ```
 dataset_path: json
 dataset_name: null
-dataset_kwargs: 
+dataset_kwargs:
   data_files: /path/to/my/json
 ```
 -------------------------------

lm_eval/api/metrics.py

@@ -111,9 +111,9 @@ def ter(items):
 def brier_score(items):  # This is a passthrough function
     gold, predictions = list(zip(*items))
     gold = list(gold)
-    gold_one_hot = np.eye(np.max(gold)+1)[gold]
+    gold_one_hot = np.eye(np.max(gold) + 1)[gold]
     predictions = list(zip(*items))[1]
-    return np.mean(np.sum((predictions - gold_one_hot)**2, axis=1))
+    return np.mean(np.sum((predictions - gold_one_hot) ** 2, axis=1))
 
 
 @register_metric(

lm_eval/api/task.py

@@ -1066,14 +1066,18 @@ class ConfigurableTask(Task):
             prob_norm = utils.softmax(lls)
 
             # TODO use keyword arguments to the metric?
-            # gold, pred, norm stuff, the original lls, 
+            # gold, pred, norm stuff, the original lls,
             result_dict = {
                 **({"acc": acc} if "acc" in use_metric else {}),
                 **({"f1": (gold, pred)} if "f1" in use_metric else {}),
                 **({"mcc": (gold, pred)} if "mcc" in use_metric else {}),
                 **({"acc_norm": acc_norm} if "acc_norm" in use_metric else {}),
                 **({"exact_match": exact_match} if "exact_match" in use_metric else {}),
-                **({"brier_score": (gold, prob_norm)} if "brier_score" in use_metric else {}),
+                **(
+                    {"brier_score": (gold, prob_norm)}
+                    if "brier_score" in use_metric
+                    else {}
+                ),
             }
 
             if "acc_mutual_info" in use_metric:

lm_eval/tasks/bigbench/aux_metric.py

@@ -2,9 +2,13 @@ from textdistance import levenshtein
 from transformers import AutoTokenizer
 
 # Change this tokenizer to fit with the model you are using.
-tokenizer = AutoTokenizer.from_pretrained("EleutherAI/pythia-2.8b")
+tokenizer = AutoTokenizer.from_pretrained("EleutherAI/pythia-2.8b", max_new_tokens=128)
+
 
 def token_edit_distance(references, predictions, **kwargs):
+    print(references)
+    print(predictions)
+    print("###")
     ref_tokens = tokenizer.encode(references[0])
     pred_tokens = tokenizer.encode(predictions[0])
     return levenshtein.distance(ref_tokens, pred_tokens)

lm_eval/tasks/bigbench/generate_until_template_yaml

@@ -16,4 +16,4 @@ metric_list:
     ignore_punctuation: true
   - metric: !function aux_metric.token_edit_distance # pip install textdistance
     aggregation: mean
-    higher_is_better: false
+    higher_is_better: false
\ No newline at end of file

lm_eval/tasks/mmlu/default/_default_template_yaml

@@ -17,4 +17,4 @@ metric_list:
     higher_is_better: true
   - metric: brier_score
     aggregation: mean
-    higher_is_better: false
+    higher_is_better: false
\ No newline at end of file