Update app.py

app.py

@@ -1,8 +1,9 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
 import spaces
+import os
 
-@spaces.GPU  # مهم لـ ZeroGPU!
+@spaces.GPU
 def respond(
     message,
     history: list[dict[str, str]],
@@ -11,7 +12,9 @@ def respond(
     temperature,
     top_p,
 ):
-    client = InferenceClient(model="anaspro/iraqi-kashif-2b")
+    # استخدم Token من Secrets
+    token = os.environ.get("HF_TOKEN")
+    client = InferenceClient(model="anaspro/iraqi-kashif-2b", token=token)
     
     messages = [{"role": "system", "content": system_message}]
     messages.extend(history)
@@ -26,8 +29,7 @@ def respond(
         top_p=top_p,
     ):
         if msg.choices and msg.choices[0].delta.content:
-            token = msg.choices[0].delta.content
-            response += token
+            response += msg.choices[0].delta.content
             yield response
 
 chatbot = gr.ChatInterface(