Update app.py

app.py

@@ -133,7 +133,18 @@ def clear_history(state, state_):
         True, state, state_, state.to_gradio_chatbot(), [[], []])
 
 
+conv_mode = "llava_v1"
+model_path = 'LanguageBind/Video-LLaVA-7B'
+device = 'cpu'
+load_8bit = False
+load_4bit = True
+dtype = torch.get_default_dtype()
+handler = Chat(model_path, conv_mode=conv_mode, load_8bit=load_8bit, load_4bit=load_8bit, device=device)
+# handler.model.to(dtype=dtype)
+if not os.path.exists("temp"):
+    os.makedirs("temp")
 
+'''
 conv_mode = "llava_v1"
 model_path = 'LanguageBind/Video-LLaVA-7B'
 device = 'cuda'
@@ -144,7 +155,7 @@ handler = Chat(model_path, conv_mode=conv_mode, load_8bit=load_8bit, load_4bit=l
 # handler.model.to(dtype=dtype)
 if not os.path.exists("temp"):
     os.makedirs("temp")
-
+'''
 print(torch.cuda.memory_allocated())
 print(torch.cuda.max_memory_allocated())