Update app.py

app.py

@@ -8,22 +8,35 @@
 import spaces
 import os
 
-os.putenv('PYTORCH_NVML_BASED_CUDA_CHECK','1')
-os.putenv('TORCH_LINALG_PREFER_CUSOLVER','1')
-alloc_conf_parts = [
-    'expandable_segments:True',
-    'pinned_use_background_threads:True'  # Specific to pinned memory.
-]
-os.environ['PYTORCH_CUDA_ALLOC_CONF'] = ','.join(alloc_conf_parts)
+os.environ['PYTORCH_NVML_BASED_CUDA_CHECK'] = '1'
+os.environ['TORCH_LINALG_PREFER_CUSOLVER'] = '1'
+os.environ['PYTORCH_ALLOC_CONF'] = 'expandable_segments:True,pinned_use_background_threads:True'
 os.environ["SAFETENSORS_FAST_GPU"] = "1"
-os.putenv('HF_HUB_ENABLE_HF_TRANSFER','1')
+os.environ['HF_HUB_ENABLE_HF_TRANSFER'] = '1'
+
+import torch
+
+torch.backends.cuda.matmul.allow_tf32 = False  #  torch 2.8
+torch.backends.cudnn.allow_tf32 = False        #  torch 2.8
+
+torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction = False
+torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction = False
+#torch.backends.fp32_precision = "ieee"  torch 2.9
+#torch.backends.cuda.matmul.fp32_precision = "ieee"  torch 2.9
+#torch.backends.cudnn.fp32_precision = "ieee"  torch 2.9
+#torch.backends.cudnn.conv.fp32_precision = "ieee"  torch 2.9
+#torch.backends.cudnn.rnn.fp32_precision = "ieee"  torch 2.9
+torch.backends.cudnn.deterministic = False
+torch.backends.cudnn.benchmark = False
+torch.backends.cuda.preferred_blas_library="cublas"
+torch.backends.cuda.preferred_linalg_library="cusolver"
+torch.set_float32_matmul_precision("highest")
 
 import random
 import uuid
 import gradio as gr
 import numpy as np
 from PIL import Image
-import torch
 from typing import Tuple
 import paramiko
 import datetime
@@ -32,16 +45,6 @@ from diffusers import StableDiffusionXLPipeline, StableDiffusionXLImg2ImgPipelin
 from diffusers.models.attention_processor import Attention, AttnProcessor2_0
 from transformers import CLIPTextModelWithProjection, CLIPTextModel, CLIPTokenizer
 
-torch.backends.cuda.matmul.allow_tf32 = False
-torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction = False
-torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction = False
-torch.backends.cudnn.allow_tf32 = False
-torch.backends.cudnn.deterministic = False
-torch.backends.cudnn.benchmark = False
-torch.backends.cuda.preferred_blas_library="cublas"
-torch.backends.cuda.preferred_linalg_library="cusolver"
-torch.set_float32_matmul_precision("highest")
-
 FTP_HOST = 'noahcohn.com'
 FTP_USER = 'ford442'
 FTP_PASS = os.getenv("FTP_PASS")