Update app.py

app.py

@@ -1,8 +1,14 @@
 import spaces
 import os
 
+os.putenv('PYTORCH_NVML_BASED_CUDA_CHECK','1')
 os.putenv('TORCH_LINALG_PREFER_CUSOLVER','1')
-os.putenv('PYTORCH_CUDA_ALLOC_CONF','max_split_size_mb:128')
+os.putenv('PYTORCH_CUDA_ALLOC_CONF', 'max_split_size_mb:128')
+os.putenv('PYTORCH_CUDA_ALLOC_CONF', 'backend:cudaMallocAsync')
+os.putenv('PYTORCH_CUDA_ALLOC_CONF', 'expandable_segments:True')
+os.putenv('PYTORCH_CUDA_ALLOC_CONF', 'pinned_use_cuda_host_register:True')
+os.putenv('PYTORCH_CUDA_ALLOC_CONF', 'pinned_num_register_threads:8')
+os.putenv('PYTORCH_CUDA_ALLOC_CONF', 'pinned_use_background_threads:True')
 os.environ["SAFETENSORS_FAST_GPU"] = "1"
 os.putenv('HF_HUB_ENABLE_HF_TRANSFER','1')