updete client to post

app.py

@@ -10,23 +10,23 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
     """
 
     prompt = f"{system_message}\n\nUser: {message}\n\nAssistant:"
-
+    
     response = ""
 
     try:
-        for token in client.text_generation(
-            prompt,
-            max_new_tokens=max_tokens,
-            temperature=temperature,
-            top_p=top_p,
-            stream=True,  # Enables real-time streaming output
-        ):
-            response += token
-            yield response  # Stream response token by token
+        # 🔥 Use raw API request instead of `text_generation()`
+        result = client.post(
+            json={"inputs": prompt, "parameters": {
+                "max_new_tokens": max_tokens,
+                "temperature": temperature,
+                "top_p": top_p
+            }},
+        )
+        response = result.text
+        yield response
 
     except Exception as e:
-        yield f"Error: {str(e)}"  # Handle potential API errors
-
+        yield f"Error: {str(e)}"
 
 # 🔹 Gradio Chat Interface
 demo = gr.ChatInterface(