Perhaps that's better

pmd.py

@@ -95,7 +95,7 @@ class DatasetsLoader(BaseLoader):
         self.num_proc = num_proc
         # In order to not have memory explode we define a batch size per proc
         self.batch_size_per_proc = batch_size // self.num_proc
-        assert self.batch_size_per_proc > 1
+        assert self.batch_size_per_proc >= 1
 
     @abstractmethod
     def cast_to_pmd_features(self, row: Dict) -> List[Dict[str, Any]]:
@@ -111,7 +111,7 @@ class DatasetsLoader(BaseLoader):
             for i in range(batch_size)
         ]
 
-    def _generate_rows_casted_pmd_features(self, dset: Dataset, batch_start: int):
+    def _generate_rows_casted_pmd_features(self, batch_start: int, dset: Dataset):
         dataset_size = len(dset)
         batch_end = min(batch_start + self.batch_size_per_proc, dataset_size)
         batch = dset[batch_start:batch_end]
@@ -138,7 +138,7 @@ class DatasetsLoader(BaseLoader):
             assert self.num_proc > 1
             with Pool(self.num_proc) as pool:
                 rows_casted_pmd_features = pool.imap(
-                    partial(self._generate_rows_casted_pmd_features, dataset),
+                    partial(self._generate_rows_casted_pmd_features, dset=dataset),
                     range(0, dataset_size, self.batch_size_per_proc),
                 )
                 for row_casted_pmd_features in rows_casted_pmd_features: