Update chatbot.py

chatbot.py

@@ -48,11 +48,12 @@ def load_pipeline():
     pipe = pipeline("text-generation", 
                     model= model, 
                     tokenizer = tokenizer,
-                    max_new_tokens = 252,
+                    max_new_tokens = 20,
                     top_k = 30, 
                     early_stopping=True,
-                    num_beams =2,
+                    num_beams = 2,
                     temperature = 0.1,
+                    top_p = 0.8,
                     repetition_penalty = 1.03)
 
     llm = HuggingFacePipeline(pipeline = pipe)
@@ -77,5 +78,5 @@ def demo_chain(input_text, memory):
         memory=memory
     )
 
-    chat_reply = conversation.invoke(input=input_text, max_new_tokens = 252)
+    chat_reply = conversation.invoke(input=input_text, max_new_tokens = 20)
     return chat_reply["response"]