Update app.py

app.py

@@ -9,6 +9,11 @@
 import os
 #subprocess.run(['sh', './torch.sh'])
 
+os.putenv('TORCH_LINALG_PREFER_CUSOLVER','1')
+os.putenv('PYTORCH_CUDA_ALLOC_CONF','max_split_size_mb:128')
+os.environ["SAFETENSORS_FAST_GPU"] = "1"
+os.putenv('HF_HUB_ENABLE_HF_TRANSFER','1')
+
 import gradio as gr
 import numpy as np
 from PIL import Image
@@ -58,15 +63,13 @@ styles = {k["name"]: (k["prompt"], k["negative_prompt"]) for k in style_list}
 DEFAULT_STYLE_NAME = "Style Zero"
 STYLE_NAMES = list(styles.keys())
 
-# os.putenv('TORCH_LINALG_PREFER_CUSOLVER','1')
-
 MAX_SEED = np.iinfo(np.int32).max
 
 code = r'''
+import os
 
 import torch
 import paramiko
-import os
 
 torch.backends.cuda.matmul.allow_tf32 = False
 torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction = False
@@ -74,12 +77,11 @@ 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.backends.cuda.preferred_blas_library="cublas"
+torch.backends.cuda.preferred_linalg_library="cusolver"
 torch.set_float32_matmul_precision("highest")
 
 HF_TOKEN = os.getenv("HF_TOKEN")
-
 FTP_HOST = os.getenv("FTP_HOST")
 FTP_USER = os.getenv("FTP_USER")
 FTP_PASS = os.getenv("FTP_PASS")
@@ -173,9 +175,7 @@ import gc
 
 MAX_SEED = np.iinfo(np.int32).max
 
-os.putenv('HF_HUB_ENABLE_HF_TRANSFER','1')
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
-os.environ["SAFETENSORS_FAST_GPU"] = "1"
 
 #accelerator = Accelerator(mixed_precision="bf16") # Example
 
@@ -197,7 +197,7 @@ def load_and_prepare_model():
     )
     pipe.vae = vaeXL #.to(torch.bfloat16)
     #pipe.scheduler = sched
-    pipe.vae.do_resize = False
+    #pipe.vae.do_resize = False
     pipe.vae.do_convert_rgb = True
     pipe.vae.set_default_attn_processor()
     print(f'init noise scale: {pipe.scheduler.init_noise_sigma}')
@@ -253,7 +253,7 @@ def generate_30(
     np.random.seed(seed)
     torch.manual_seed(seed)
     torch.cuda.manual_seed_all(seed)
-    generator = torch.Generator(device='cuda').manual_seed(seed)
+    generator = torch.Generator(device='cpu').manual_seed(seed)
     options = {
         "prompt": [prompt],
         "negative_prompt": [negative_prompt],
@@ -311,7 +311,7 @@ def generate_60(
     np.random.seed(seed)
     torch.manual_seed(seed)
     torch.cuda.manual_seed_all(seed)
-    generator = torch.Generator(device='cuda').manual_seed(seed)
+    generator = torch.Generator(device='cpu').manual_seed(seed)
     options = {
         "prompt": [prompt],
         "negative_prompt": [negative_prompt],
@@ -369,7 +369,7 @@ def generate_90(
     np.random.seed(seed)
     torch.manual_seed(seed)
     torch.cuda.manual_seed_all(seed)
-    generator = torch.Generator(device='cuda').manual_seed(seed)
+    generator = torch.Generator(device='cpu').manual_seed(seed)
     options = {
         "prompt": [prompt],
         "negative_prompt": [negative_prompt],