diff --git a/ImageReward/ImageReward.py b/ImageReward/ImageReward.py
index b3dfba3..c183a6e 100644
--- a/ImageReward/ImageReward.py
+++ b/ImageReward/ImageReward.py
@@ -15,6 +15,7 @@
 from PIL import Image
 from .models.BLIP.blip_pretrain import BLIP_Pretrain
 from torchvision.transforms import Compose, Resize, CenterCrop, ToTensor, Normalize
+from transformers import BertTokenizer
 
 try:
     from torchvision.transforms import InterpolationMode
@@ -69,11 +70,11 @@ def forward(self, input):
 
 
 class ImageReward(nn.Module):
-    def __init__(self, med_config, device='cpu'):
+    def __init__(self, med_config, device='cpu', tokenizer=None):
         super().__init__()
         self.device = device
         
-        self.blip = BLIP_Pretrain(image_size=224, vit='large', med_config=med_config)
+        self.blip = BLIP_Pretrain(image_size=224, vit='large', med_config=med_config, tokenizer=tokenizer)
         self.preprocess = _transform(224)
         self.mlp = MLP(768)
         
diff --git a/ImageReward/models/BLIP/blip.py b/ImageReward/models/BLIP/blip.py
index 0dfdb72..b67e840 100644
--- a/ImageReward/models/BLIP/blip.py
+++ b/ImageReward/models/BLIP/blip.py
@@ -13,8 +13,9 @@
 from .vit import VisionTransformer, interpolate_pos_embed
 
 
-def init_tokenizer():
-    tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
+def init_tokenizer(tokenizer: BertTokenizer = None):
+    if tokenizer is None:
+        tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
     tokenizer.add_special_tokens({'bos_token':'[DEC]'})
     tokenizer.add_special_tokens({'additional_special_tokens':['[ENC]']})       
     tokenizer.enc_token_id = tokenizer.additional_special_tokens_ids[0]  
diff --git a/ImageReward/models/BLIP/blip_pretrain.py b/ImageReward/models/BLIP/blip_pretrain.py
index 793cb07..d8e2f0b 100644
--- a/ImageReward/models/BLIP/blip_pretrain.py
+++ b/ImageReward/models/BLIP/blip_pretrain.py
@@ -11,7 +11,8 @@
 from .blip import create_vit, init_tokenizer
 
 class BLIP_Pretrain(nn.Module):
-    def __init__(self,                 
+    def __init__(self,
+                 tokenizer,
                  med_config = "med_config.json",  
                  image_size = 224,
                  vit = 'base',
@@ -31,7 +32,7 @@ def __init__(self,
         
         self.visual_encoder, vision_width = create_vit(vit,image_size, vit_grad_ckpt, vit_ckpt_layer, 0)
         
-        self.tokenizer = init_tokenizer()   
+        self.tokenizer = init_tokenizer(tokenizer)
         encoder_config = BertConfig.from_json_file(med_config)
         encoder_config.encoder_width = vision_width
         self.text_encoder = BertModel(config=encoder_config, add_pooling_layer=False)
diff --git a/setup.py b/setup.py
index cff44dc..4244d27 100644
--- a/setup.py
+++ b/setup.py
@@ -1,6 +1,5 @@
 from setuptools import setup, find_packages
 import os
-import pkg_resources
 from pathlib import Path
 
 long_description = (Path(__file__).parent / "README.md").read_text()