update

app.py

@@ -697,23 +697,23 @@ MODEL_PATH = os.environ.get("MODEL_PATH", "notfound, please set `export MODEL_PA
 
 def launch():
     global demo, llm, DEBUG
+    model_desc = MODEL_DESC
+    model_path = MODEL_PATH
+    assert os.path.exists(model_path), f'{model_path} not found'
+    model_title = MODEL_TITLE
+    tensor_parallel = TENSOR_PARALLEL
+    assert tensor_parallel > 0 , f'{tensor_parallel} invalid'
+    dtype = DTYPE
+    sys_prompt = SYSTEM_PROMPT_1
+    max_tokens = 4096
+
     if DEBUG:
-        model_desc + "<br>This is in debug mode, responses will be copy original"
+        model_desc += "<br>!!!!! This is in debug mode, responses will be copy original"
         response_fn = debug_chat_response_echo
     else:
-        model_desc = MODEL_DESC
-        model_path = MODEL_PATH
-        assert os.path.exists(model_path), f'{model_path} not found'
-        model_title = MODEL_TITLE
-        tensor_parallel = TENSOR_PARALLEL
-        assert tensor_parallel > 0 , f'{tensor_parallel} invalid'
-        dtype = DTYPE
-
         # ! load the model
         llm = LLM(model=model_path, dtype=dtype, tensor_parallel_size=tensor_parallel)
 
-        sys_prompt = SYSTEM_PROMPT_1
-        max_tokens = 4096
         print(f'Use system prompt:\n{sys_prompt}')
 
         # response_fn = chat_response_stream_multiturn if args.multiturn else chat_response_stream