diff --git a/configs/train_config.json b/configs/train_config.json
index d3c2392..a833566 100644
--- a/configs/train_config.json
+++ b/configs/train_config.json
@@ -3,10 +3,10 @@
   "data": {
     "data_dict_path": "uniref50_gpt_data.pkl",
     "data_df_path": "uniref50_gzip_subsample.csv",
-    "split_ratios": [0.8, 0.1, 0.1]
+    "split_ratios": [0.88, 0.02, 0.1]
   },
   "datamodule": {
-    "batch_size": 4,
+    "batch_size": 8,
     "max_protein_length": 1500
   },
   "model": {
diff --git a/cprt/data/cprt_datamodule.py b/cprt/data/cprt_datamodule.py
index 9022abc..7e60b99 100644
--- a/cprt/data/cprt_datamodule.py
+++ b/cprt/data/cprt_datamodule.py
@@ -97,23 +97,25 @@ def train_dataloader(self) -> DataLoader:  # type: ignore[type-arg]
             shuffle=True,
             collate_fn=self.collate_fn,
             num_workers=4,
+            drop_last=True,
         )
 
     def val_dataloader(self) -> DataLoader:  # type: ignore[type-arg]
         """Set up val loader."""
         return DataLoader(
             self.val_dataset,
-            batch_size=self.batch_size,
+            batch_size=self.batch_size * 2,
             shuffle=False,
             collate_fn=self.collate_fn,
             num_workers=4,
+            drop_last=True,
         )
 
     def test_dataloader(self) -> DataLoader:  # type: ignore[type-arg]
         """Set up test loader."""
         return DataLoader(
             self.test_dataset,
-            batch_size=self.batch_size,
+            batch_size=self.batch_size * 2,
             shuffle=False,
             collate_fn=self.collate_fn,
             num_workers=4,