Remove model preloading at startup to prevent Zero GPU initialization issues; models will now load and cache on first use.

app.py

@@ -529,24 +529,12 @@ with tranception_design:
     gr.Markdown("<p><b>Tranception: Protein Fitness Prediction with Autoregressive Transformers and Inference-time Retrieval</b><br>Pascal Notin, Mafalda Dias, Jonathan Frazer, Javier Marchena-Hurtado, Aidan N. Gomez, Debora S. Marks<sup>*</sup>, Yarin Gal<sup>*</sup><br><sup>* equal senior authorship</sup></p>")
     gr.Markdown("Links: <a href='https://proceedings.mlr.press/v162/notin22a.html' target='_blank'>Paper</a>  <a href='https://github.com/OATML-Markslab/Tranception' target='_blank'>Code</a>  <a href='https://sites.google.com/view/proteingym/substitutions' target='_blank'>ProteinGym</a>  <a href='https://igem.org/teams/5247' target='_blank'>BASIS-China iGEM Team</a>")
 
-# Preload models function
-def preload_models():
-    """Preload models at startup to avoid downloading during inference"""
-    print("Preloading models at startup...")
-    try:
-        # Try to load Small model (fastest)
-        load_model_cached("Small")
-        print("Small model preloaded successfully")
-    except Exception as e:
-        print(f"Could not preload Small model: {e}")
-    
-    # Optionally preload other models
-    # load_model_cached("Medium")
-    # load_model_cached("Large")
-
 if __name__ == "__main__":
-    # Preload models before launching
-    preload_models()
+    # Don't preload models at startup - this can cause Zero GPU initialization issues
+    # Models will be loaded and cached on first use
+    print("Starting Tranception app...")
+    print("Note: Models will be downloaded on first use")
+    print("Zero GPU spaces may sleep after ~15 minutes of inactivity")
     
     # Simple launch without queue to avoid Zero GPU conflicts
     tranception_design.launch(