Modify bugs in hellaswag

lm_eval/api/task.py

@@ -26,6 +26,7 @@ import numpy as np
 from tqdm import tqdm
 
 from lm_eval import utils
+from lm_eval.datasets.hellaswag import hellaswag
 from lm_eval.api import samplers
 from lm_eval.api.instance import Instance, OutputType
 from lm_eval.api.metrics import bits_per_byte, mean, weighted_perplexity
@@ -51,6 +52,7 @@ ALL_OUTPUT_TYPES = [
 
 eval_logger = logging.getLogger("lm-eval")
 
+CLASS_DATASETS = {'hellaswag': hellaswag }
 
 @dataclass
 class TaskConfig(dict):
@@ -708,7 +710,7 @@ class ConfigurableTask(Task):
             self.OUTPUT_TYPE = self.config.output_type
 
         if self.config.dataset_path is not None:
-            self.DATASET_PATH = inspect.getfile(self.config.dataset_path)
+            self.DATASET_PATH = inspect.getfile(CLASS_DATASETS[self.config.dataset_name])
 
         if self.config.dataset_name is not None:
             self.DATASET_NAME = self.config.dataset_name

lm_eval/tasks/hellaswag/hellaswag.yaml

 group:
   - multiple_choice
 task: hellaswag
-dataset_path: lm_eval.datasets.hellaswag.hellaswag
+dataset_path: None
 dataset_name: hellaswag
 output_type: multiple_choice
 training_split: train