refactor '.predict' add SystemMessage

app.py

@@ -18,8 +18,8 @@ with gr.Blocks() as demo:
         with gr.Column():
             url = gr.Textbox(value="https://www.gradio.app/docs/gradio/chatinterface", label="Docs URL", render=True)
             chat = gr.ChatInterface(
-                model.respond,
-                # model.predict,
+                # model.respond,
+                model.predict,
                 # model.rag,
                 additional_inputs=[
                     url,

model.py

@@ -1,7 +1,8 @@
 import os
 
 from huggingface_hub import InferenceClient
-from langchain.schema import AIMessage, HumanMessage
+from langchain.schema import SystemMessage, AIMessage, HumanMessage
+
 from langchain.chains import RetrievalQA
 from langchain.prompts import PromptTemplate
 
@@ -56,16 +57,20 @@ class Model:
         )
 
     def predict(self, message, history, url, max_tokens, temperature, top_p):
-        history_langchain_format = []
+        history_langchain_format = [SystemMessage(content="You're a helpful python developer assistant")]
         for msg in history:
             if msg['role'] == "user":
                 history_langchain_format.append(HumanMessage(content=msg['content']))
             elif msg['role'] == "assistant":
                 history_langchain_format.append(AIMessage(content=msg['content']))
         history_langchain_format.append(HumanMessage(content=message))
-        # gpt_response = self.chat_model(history_langchain_format)
-        # return gpt_response.content
-        return self._retrieval_qa(url).invoke({"query": message})['result']
+
+        # ai_msg = self.chat_model.invoke(history_langchain_format)
+        # return ai_msg.content
+
+        ret = self._retrieval_qa(url)
+        return ret.invoke({"query": message})['result']
+
 
     def respond(
         self,