stable gradio interface but requires inprovement

app.py

@@ -78,8 +78,10 @@ def chat(message, history, hotel_id):
             add_generation_prompt=True
         )
         inputs = tokenizer.encode(input_text, return_tensors="pt").to(device)
-        outputs = model.generate(inputs, max_new_tokens=1024, do_sample=False)
+        with torch.no_grad():
+            outputs = model.generate(inputs, max_new_tokens=1024, do_sample=True)
         decoded = tokenizer.decode(outputs[0], skip_special_tokens=False)
+        print(decoded)
         # Extract assistant response
         response = decoded.split("<|im_start|>assistant")[-1]
         response = response.split("<|im_end|>")[0].strip()
@@ -104,7 +106,8 @@ def chat(message, history, hotel_id):
             add_generation_prompt=True
         )
         inputs = tokenizer.encode(input_text, return_tensors="pt").to(device)
-        outputs = model.generate(inputs, max_new_tokens=1024, do_sample=False)
+        with torch.no_grad():
+            outputs = model.generate(inputs, max_new_tokens=1024, do_sample=True)
         decoded = tokenizer.decode(outputs[0], skip_special_tokens=False)
         response = decoded.split("<|start_of_role|>assistant<|end_of_role|>")[-1]
         response = response.split("<|end_of_text|>")[0].strip()
@@ -118,24 +121,45 @@ def chat(message, history, hotel_id):
 # Available hotels
 hotel_ids = ["cyprus-guesthouse-family", "coastal-villa-family", "village-inn-family"]
 
+# Gradio UI
 # Gradio UI
 with gr.Blocks() as demo:
-    gr.Markdown("### 🏨 Multi-Hotel Chatbot Demo")
-    gr.Markdown(f"**Running:** {model_name}")
-    hotel_selector = gr.Dropdown(hotel_ids, label="Hotel", value=hotel_ids[0])
-    #chatbot = gr.Chatbot(type="messages")
-    with gr.Row():
-        chatbot = gr.Chatbot(type="messages")
-    msg = gr.Textbox(show_label=False, placeholder="Ask about the hotel...")
-    msg.submit(
-        fn=chat,
-        inputs=[msg, chatbot, hotel_selector],
-        outputs=[chatbot, msg]
-    )
+    # ⬇️  NEW panel wrapper
+    with gr.Column(variant="panel"):
+
+        gr.Markdown("### 🏨 Multi‑Hotel Chatbot Demo")
+        gr.Markdown(f"**Running:** {model_name}")
+
+        hotel_selector = gr.Dropdown(
+            hotel_ids,
+            label="Hotel",
+            value=hotel_ids[0]
+        )
+
+        # Chat window in its own row so it stretches
+        with gr.Row():
+            chatbot = gr.Chatbot(type="messages")
+
+        msg = gr.Textbox(
+            show_label=False,
+            placeholder="Ask about the hotel..."
+        )
+
+        # Clear‑history button
+        gr.Button("Clear").click(lambda: ([], ""), None, [chatbot, msg])
+
+        # Wire the textbox to the chat function
+        msg.submit(
+            fn=chat,
+            inputs=[msg, chatbot, hotel_selector],
+            outputs=[chatbot, msg]
+        )
+
+    # Anything outside the column shows below the panel
     gr.Markdown("⚠️ Pause the Space when done to avoid charges.")
 
 # Enable streaming queue for generator-based chat
-demo.queue()
+demo.queue(default_concurrency_limit=2, max_size=32)
 
 if __name__ == "__main__":
     demo.launch()