Update app.py

app.py

@@ -4,10 +4,9 @@ from huggingface_hub import InferenceClient
 
 client = InferenceClient("microsoft/Phi-3.5-mini-instruct")
 
-
 async def respond(
     message,
-    history: list[dict],
+    history: list[tuple[str, str]],
     system_message,
     max_tokens,
     temperature,
@@ -15,7 +14,11 @@ async def respond(
 ):
     # Ensure history is in OpenAI-style 'role' and 'content' format
     messages = [{"role": "system", "content": system_message}]
-    messages.extend(history)  # Add existing history
+    for val in history:
+      if val[0]:
+          messages.append({"role": "user", "content": val[0]})
+      if val[1]:
+          messages.append({"role": "assistant", "content": val[1]})
 
     # Add the user's latest message
     messages.append({"role": "user", "content": message})
@@ -52,3 +55,4 @@ demo = gr.ChatInterface(
 
 if __name__ == "__main__":
     demo.queue().launch()  # Simply call queue without `concurrency_count`
+