Refactor GPU warmup logic by directly calling _warmup function and removing redundant demo.load call in request context.

README.md

@@ -4,7 +4,7 @@ emoji: 😻
 colorFrom: indigo
 colorTo: yellow
 sdk: gradio
-sdk_version: 5.23.1
+sdk_version: 5.42.0
 app_file: app.py
 pinned: true
 # hf_oauth: true

app.py

@@ -5,7 +5,6 @@ print(f"Starting up: {time.strftime('%Y-%m-%d %H:%M:%S')}")
 # Standard library imports
 
 import os
-os.environ["MPLCONFIGDIR"] = "/tmp/matplotlib"  # set BEFORE importing matplotlib
 
 #Enforce local cching:
 
@@ -239,10 +238,12 @@ def no_op_decorator(func):
 # decorator_to_use = spaces.GPU() if is_running_in_hf_space() else no_op_decorator
 # #duration=120
 
-@spaces.GPU(duration=1)
-def _warmup():
+@spaces.GPU(duration=1)          # ← forces the detector to see a GPU-aware fn
+def _warmup(): 
     print("Warming up...")
 
+_warmup()
+
 
 # if is_running_in_hf_space():
 @spaces.GPU(duration=30)
@@ -1422,9 +1423,6 @@ with gr.Blocks(theme=theme, css=f"""
         outputs=query_display
     )
 
-    # Run a tiny GPU warmup safely inside a request context
-    demo.load(fn=_warmup, inputs=None, outputs=None, queue=False)
-
 
 # demo.static_dirs = {
 #     "static": str(static_dir)