Update app.py

app.py

@@ -6,21 +6,29 @@ client = InferenceClient(provider="novita", api_key=os.environ["HF_TOKEN"])
 models = ["deepseek-ai/DeepSeek-V3.1","zai-org/GLM-4.5","Qwen/Qwen3-Coder-480B-A35B-Instruct", "Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8","deepseek-ai/DeepSeek-R1","deepseek-ai/DeepSeek-R1-0528","Qwen/Qwen3-Next-80B-A3B-Thinking"]
 
 def respond(message, history, model):
-    completion = client.chat.completions.create(
+    history.append((message, ""))
+    yield "", history
+    full_reply = ""
+    for chunk in client.chat.completions.create(
         model=model,
-        messages=[{"role": "user", "content": message}]
-    )
-    reply = completion.choices[0].message.content
-    history.append((message, reply))
-    return "", history
+        messages=[{"role": "user", "content": message}],
+        stream=True
+    ):
+        if hasattr(chunk.choices[0].delta, 'content') and chunk.choices[0].delta.content:
+            full_reply += chunk.choices[0].delta.content
+            history[-1] = (message, full_reply)
+            yield "", history
 
 with gr.Blocks() as demo:
     gr.Markdown("## AI")
     chatbot = gr.Chatbot(height=400)
     msg = gr.Textbox(label="Ask me smth")
     model_dd = gr.Dropdown(models, label="Model", value=models[0])
-    clear = gr.Button("Clear")
+    with gr.Row():
+        submit_btn = gr.Button("Send")
+        clear = gr.Button("Clear")
     msg.submit(respond, [msg, chatbot, model_dd], [msg, chatbot])
+    submit_btn.click(respond, [msg, chatbot, model_dd], [msg, chatbot])
     clear.click(lambda: None, None, chatbot, queue=False)
 
-demo.launch()
+demo.queue().launch()