Update app.py

app.py

@@ -6,9 +6,9 @@ from ctransformers import AutoModelForCausalLM
 llm = AutoModelForCausalLM.from_pretrained("TheBloke/WizardLM-7B-uncensored-GGUF", model_file="WizardLM-7B-uncensored.Q4_K_M.gguf", model_type="llama", stream=True)
 
 def generate_response(message):
-    tokens = llm(message)
-    response_sentence = ' '.join(tokens)
-    yield response_sentence
+    for text in llm(message):
+        toe = "".join(text)
+        yield toe
     
 
 def chatbot(message, history):