4 bit load.

app.py

@@ -220,8 +220,8 @@ if __name__ == '__main__':
     conv_mode = "llama_2"
     model_path = 'DAMO-NLP-SG/VideoLLaMA2-7B'
 
-    handler = Chat(model_path, conv_mode=conv_mode, load_8bit=True, load_4bit=False, device='cuda')
-    handler.model.to(dtype=torch.float16)
+    handler = Chat(model_path, conv_mode=conv_mode, load_8bit=False, load_4bit=True, device='cuda')
+    # handler.model.to(dtype=torch.float16)
 
     if not os.path.exists("temp"):
         os.makedirs("temp")