diff --git a/finetune_demo/finetune.py b/finetune_demo/finetune.py
index 321836d..2552fa0 100644
--- a/finetune_demo/finetune.py
+++ b/finetune_demo/finetune.py
@@ -348,7 +348,7 @@ def load_tokenizer_and_model(
         model_dir: str,
         peft_config: Optional[PeftConfig] = None,
 ):
-    tokenizer = AutoTokenizer.from_pretrained(model_dir, trust_remote_code=True)
+    tokenizer = AutoTokenizer.from_pretrained(model_dir, padding_side='left', trust_remote_code=True)
     if peft_config is not None:
         model = AutoModelForCausalLM.from_pretrained(
             model_dir,
@@ -370,6 +370,8 @@ def load_tokenizer_and_model(
 
 def compute_metrics(eval_preds: EvalPrediction, tokenizer):
     batched_pred_ids, batched_label_ids = eval_preds
+    batched_pred_ids[batched_pred_ids==-100] = tokenizer.pad_token_id
+    batched_label_ids[batched_label_ids==-100] = tokenizer.pad_token_id
     metrics_dct = {'rouge-1': [], 'rouge-2': [], 'rouge-l': [], 'bleu-4': []}
     for pred_ids, label_ids in zip(batched_pred_ids, batched_label_ids):
         pred_txt = tokenizer.decode(pred_ids).strip()
diff --git a/finetune_demo/finetune_vision.py b/finetune_demo/finetune_vision.py
index c64cb05..cafe7e6 100644
--- a/finetune_demo/finetune_vision.py
+++ b/finetune_demo/finetune_vision.py
@@ -364,7 +364,7 @@ def process_batch_eval(
         if not dialogue_parts or dialogue_parts[-1] != len(input_ids):
             dialogue_parts.append(len(input_ids))
 
-            # Split the conversation into multiple dialogue segments
+        # Split the conversation into multiple dialogue segments
         for end_idx in range(1, len(dialogue_parts)):
             input_segment = input_ids[:dialogue_parts[end_idx]]
             attention_segment = attention_mask[:dialogue_parts[end_idx]]
@@ -394,7 +394,7 @@ def load_tokenizer_and_model(
         model_dir: str,
         peft_config: Optional[PeftConfig] = None,
 ):
-    tokenizer = AutoTokenizer.from_pretrained(model_dir, trust_remote_code=True)
+    tokenizer = AutoTokenizer.from_pretrained(model_dir, padding_side='left', trust_remote_code=True)
     if peft_config is not None:
         model = AutoModelForCausalLM.from_pretrained(
             model_dir,
@@ -416,6 +416,8 @@ def load_tokenizer_and_model(
 
 def compute_metrics(eval_preds: EvalPrediction, tokenizer):
     batched_pred_ids, batched_label_ids = eval_preds
+    batched_pred_ids[batched_pred_ids==-100] = tokenizer.pad_token_id
+    batched_label_ids[batched_label_ids==-100] = tokenizer.pad_token_id 
     metrics_dct = {'rouge-1': [], 'rouge-2': [], 'rouge-l': [], 'bleu-4': []}
     for pred_ids, label_ids in zip(batched_pred_ids, batched_label_ids):
         pred_txt = tokenizer.decode(pred_ids).strip()