Upload app.py with huggingface_hub

app.py

@@ -1,5 +1,4 @@
 import spaces  # noqa: E402  (must be imported before torch/transformers)
-# v2: added kernels dependency for FP8 inference
 
 import json
 import re
@@ -44,7 +43,7 @@ def _parse_tool_calls(text: str):
     return leftover, tool_calls
 
 
-@spaces.GPU(duration=45)
+@spaces.GPU(duration=120)
 def chat_completions(messages_json: str, tools_json: str = "", max_new_tokens: int = 2048) -> str:
     """Run one chat-completion turn against Qwen3-Coder-30B-A3B-Instruct-FP8.
 
@@ -89,7 +88,7 @@ demo = gr.Interface(
         gr.Number(label="max_new_tokens", value=2048, precision=0),
     ],
     outputs=gr.Text(label="completion_json"),
-    title="Qwen3-Coder-30B-A3B-Instruct-FP8 — chat completions",
+    title="Qwen3-Coder-30B-A3B-Instruct-FP8 \u2014 chat completions",
     description=(
         "Internal inference endpoint for the Coding Model project. "
         "Called via the local OpenAI-compatible shim, not directly by end users."