Update app.py

app.py

@@ -36,13 +36,8 @@ hf_token = os.getenv("HF_TOKEN")
 if hf_token:
     login(token=hf_token)
 
-# Set environment variables programmatically
-os.environ["TF_ENABLE_ONEDNN_OPTS"] = "0"
-os.environ["TRANSFORMERS_OFFLINE"] = "1"
+os.environ.pop("HF_HUB_OFFLINE", None)
 
-# You can also add these configurations:
-os.environ["CUDA_VISIBLE_DEVICES"] = "0"  # Use first GPU
-os.environ["TORCH_CUDA_ARCH_LIST"] = "7.5"  # Optimize for common GPU architectures
 
 class AdaptiveMedicalBot:
     def __init__(self):
@@ -609,6 +604,7 @@ if __name__ == "__main__":
         login(token=hf_token)
     
     # Launch demo
+    os.environ.pop("HF_HUB_OFFLINE", None)  # Ensure online mode
     demo = create_demo()
     demo.launch(share=True)