0

app.py

@@ -11,7 +11,7 @@ import gradio as gr
 import tempfile
 
 # Set environment variable to reduce memory fragmentation
-os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True"
+#os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True"
 
 # Initialize pipeline as None - will be loaded in GPU function
 pipe = None
@@ -30,7 +30,7 @@ def load_pipeline():
         )
         
         # Enable memory optimizations
-        pipe.enable_model_cpu_offload()
+        #pipe.enable_model_cpu_offload()
         
         # Try to enable memory efficient attention
         try: