diff --git a/credit/datasets/era5_multistep_batcher.py b/credit/datasets/era5_multistep_batcher.py
index 83512a5..dc453ff 100644
--- a/credit/datasets/era5_multistep_batcher.py
+++ b/credit/datasets/era5_multistep_batcher.py
@@ -464,16 +464,13 @@ def worker_process(start_idx, end_idx):
                 self.results[k] = sample  # Store the result keyed by its order index
 
         # Split tasks among workers
-        tasks_per_worker = max(1, len(args) // self.num_workers)
         processes = []
-
-        for i in range(self.num_workers):
-            start_idx = i * tasks_per_worker
-            end_idx = min((i + 1) * tasks_per_worker, len(args))
-            if start_idx < end_idx:  # Check if there are tasks for this worker
-                p = multiprocessing.Process(target=worker_process, args=(start_idx, end_idx))
-                processes.append(p)
-                p.start()
+        splits = np.array_split(range(len(args)), self.num_workers)
+        start_ends = [(split[0], split[-1] + 1) for split in splits]
+        for start_idx, end_idx in start_ends:
+            p = multiprocessing.Process(target=worker_process, args=(start_idx, end_idx))
+            processes.append(p)
+            p.start()
 
         # Wait for all processes to finish
         for p in processes:
@@ -769,14 +766,14 @@ def __exit__(self, exc_type, exc_val, exc_tb):
     data_config = setup_data_loading(conf)
 
     epoch = 0
-    batch_size = 2
+    batch_size = 5
     data_config["forecast_len"] = 6
-    data_config["history_len"] = 3
+    data_config["history_len"] = 1
     shuffle = True
 
     rank = 0
     world_size = 2
-    num_workers = 2
+    num_workers = 4
 
     set_globals(data_config, namespace=globals())