zerogpu updates

app.py

@@ -5,10 +5,12 @@ import gradio as gr
 from PIL import Image
 from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
 from typing import List
+from functools import lru_cache
 
 MODEL_ID = "remyxai/SpaceThinker-Qwen2.5VL-3B"
 
 @spaces.GPU
+@lru_cache(maxsize=1)
 def load_model():
     print("Loading model and processor...")
     device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -19,8 +21,6 @@ def load_model():
     processor = AutoProcessor.from_pretrained(MODEL_ID)
     return model, processor
 
-model, processor = load_model()
-
 def process_image(image_path_or_obj):
     """Loads, resizes, and preprocesses an image path or Pillow Image."""
     if isinstance(image_path_or_obj, str):
@@ -73,6 +73,7 @@ def only_assistant_text(full_text: str) -> str:
     return full_text.strip()
 
 def run_inference(image, prompt):
+    model, processor = load_model()
     """Runs Qwen2.5-VL inference on a single image and text prompt."""
     system_msg = (
         "You are VL-Thinking 🤔, a helpful assistant with excellent reasoning ability. "