diff --git a/wit/train.py b/wit/train.py
index b77f96e..3109c33 100644
--- a/wit/train.py
+++ b/wit/train.py
@@ -17,7 +17,7 @@ pretrain_model_name = None  # "qwen/Qwen-1_8B-Chat"
 learning_rate = 0.0001
 use_tril_attention_mask = None
 precision = "32-true"  # "precision:bf16-mixed,16-mixed,32-true"
-train_batch_size = 32
+train_batch_size = 16
 val_batch_size = 32
 num_proc = 8
 max_epochs = 1000
@@ -39,7 +39,7 @@ if __name__ == "__main__":
     lit_module = LitModule(pretrain_model_name, learning_rate, config, use_tril_attention_mask)
     tokenizer = QWenTokenizer("./wit_b64.tiktoken", "./wit_char.tiktoken")
 
-    level_ratio = 4
+    level_ratio = 6
     start = vocab_size * level_ratio * level_ratio
     end = start * level_ratio
     size = end * level_ratio
@@ -47,15 +47,15 @@ if __name__ == "__main__":
     train_dataset, val_dataset = raw_dataset.Split(0.95)
     train_dataloader = BatchGroupMeaningDataloader(train_dataset, train_batch_size)
     val_dataloader = BatchGroupMeaningDataloader(val_dataset, val_batch_size)
-    it = iter(train_dataloader)
-    print("data samples:")
-    for i in range(10):
-        print(next(it)["input_ids"].numpy().tolist())
+    # it = iter(train_dataloader)
+    # print("data samples:")
+    # for i in range(10):
+    #     print(next(it)["input_ids"].numpy().tolist())
 
     torch.set_float32_matmul_precision("medium")
     lit_trainer = pl.Trainer(
-        accelerator="gpu",
-        # devices=[0],
+        accelerator="cuda",
+        devices=[0, 1],
         precision=precision,
         logger=TBLogger("./", default_hp_metric=False),
         strategy=strategy,