Update app.py

app.py

@@ -52,32 +52,34 @@ client = InferenceClient(token=HF_TOKEN)
 
 def respond(message, history, system_message, max_tokens, temperature, top_p):
     if HF_TOKEN is None:
-        yield "⚠️ No HF_TOKEN found. Please add it in your Space secrets."
-        return
+        return "⚠️ No HF_TOKEN found. Please add it in your Space secrets."
 
     try:
         prompt = f"{system_message}\nUser: {message}\nAnswer:"
-        response = ""
 
-        for chunk in client.text_generation(
+        # Non-streaming inference call
+        output = client.text_generation(
             prompt,
-            model="gpt2",   # ✅ causal LM
+            model="gpt2",  # ✅ small model, works with Inference API
             max_new_tokens=max_tokens,
             temperature=temperature,
             top_p=top_p,
-            stream=True,
-        ):
-            if hasattr(chunk, "token") and chunk.token and chunk.token.text:
-                token = chunk.token.text
-                response += token
-                yield response.strip()
+            stream=False,   # 🔑 disable streaming
+        )
 
-        if not response.strip():
-            yield "⚠️ No response from model (check model ID or token)."
+        # Extract text
+        if hasattr(output, "generated_text"):
+            response = output.generated_text
+        elif isinstance(output, dict) and "generated_text" in output:
+            response = output["generated_text"]
+        else:
+            response = str(output)
+
+        return response.strip()
 
     except Exception as e:
         tb = traceback.format_exc()
-        yield f"⚠️ Error: {str(e)}\n\nTraceback:\n{tb}"
+        return f"⚠️ Error: {str(e)}\n\nTraceback:\n{tb}"
 
 
 chatbot = gr.ChatInterface(
@@ -92,7 +94,7 @@ chatbot = gr.ChatInterface(
 )
 
 with gr.Blocks(css=fancy_css) as demo:
-    gr.Markdown("<h1 id='title'>🎶 Chord Bot (API-based with Phi-3-mini) 🎶</h1>")
+    gr.Markdown("<h1 id='title'>🎶 Chord Bot (API-based, GPT-2) 🎶</h1>")
     chatbot.render()
 
 if __name__ == "__main__":