change from auto model to vision encoder decoder model

inference.py

@@ -1,5 +1,5 @@
 import torch
-from transformers import AutoTokenizer, AutoModel
+from transformers import AutoTokenizer, VisionEncoderDecoderModel
 
 import utils
 
@@ -8,7 +8,7 @@ class Inference:
     self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 
     self.tokenizer = AutoTokenizer.from_pretrained(decoder_model_name)
-    self.encoder_decoder_model = AutoModel.from_pretrained('armgabrielyan/video-summarization')
+    self.encoder_decoder_model = VisionEncoderDecoderModel.from_pretrained('armgabrielyan/video-summarization')
     self.encoder_decoder_model.to(self.device)
 
     self.max_length = max_length

requirements.txt

@@ -1,4 +1,2 @@
-nltk==3.7
-tqdm==4.64.0
 scikit-learn==1.1.1
 sentence-transformers==2.2.0

utils.py

@@ -2,8 +2,6 @@ from transformers import ViTFeatureExtractor
 import torchvision
 import torchvision.transforms.functional as fn
 import torch as th
-import os
-import pickle
 
 
 def video2image_from_path(video_path, feature_extractor_name):