Update app.py

app.py

@@ -504,7 +504,7 @@ def _llm_json(system, user):
     # Bounded timeout: a hung provider must fail over, not freeze the UI at the composing step.
     client = OpenAI(base_url="https://router.huggingface.co/v1", api_key=os.environ["HF_TOKEN"], timeout=60, max_retries=0)
     last_error = None
-    for model in ("MiniMaxAI/MiniMax-M3:fireworks-ai", "MiniMaxAI/MiniMax-M3:novita", "MiniMaxAI/MiniMax-M3:fastest"):
+    for model in ("deepseek-ai/DeepSeek-V4-Flash-0731:baseten", "deepseek-ai/DeepSeek-V4-Flash-0731", "deepseek-ai/DeepSeek-V4-Flash-0731:fastest"):
         try:
             completion = client.chat.completions.create(
                 model=model,