Update app.py

app.py

@@ -6,6 +6,12 @@
 # copies of the Software, and to permit persons to whom the Software is
 import spaces
 import os
+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 random
 import uuid
 import gradio as gr
@@ -29,12 +35,10 @@ 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")
-os.putenv("HF_HUB_ENABLE_HF_TRANSFER","1")
-os.environ["SAFETENSORS_FAST_GPU"] = "1"
 
 FTP_HOST = "1ink.us"
 FTP_USER = "ford442"
@@ -249,8 +253,8 @@ def generate_30(
     #torch.set_default_device('cuda')
     #pipe.set_adapters(["skin"], adapter_weights=[lora_scale])
     seed = int(randomize_seed_fn())
-    generator = torch.Generator(device='cuda').manual_seed(seed)
-
+    generator = torch.Generator(device='cpu').manual_seed(seed)
+    torch.set_float32_matmul_precision("highest")
     pipe.text_encoder=text_encoder.to(device=device, dtype=torch.bfloat16)
     pipe.text_encoder_2=text_encoder_2.to(device=device, dtype=torch.bfloat16)
     #prompt, negative_prompt = apply_style(style_selection, prompt, negative_prompt)
@@ -307,8 +311,8 @@ def generate_60(
     #torch.set_default_device('cuda')
     #pipe.set_adapters(["skin"], adapter_weights=[lora_scale])
     seed = int(randomize_seed_fn())
-    generator = torch.Generator(device='cuda').manual_seed(seed)
-
+    generator = torch.Generator(device='cpu').manual_seed(seed)
+    torch.set_float32_matmul_precision("highest")
     pipe.text_encoder=text_encoder.to(device=device, dtype=torch.bfloat16)
     pipe.text_encoder_2=text_encoder_2.to(device=device, dtype=torch.bfloat16)
     options = {
@@ -362,8 +366,8 @@ def generate_90(
     #torch.set_default_device('cuda')
     #pipe.set_adapters(["skin"], adapter_weights=[lora_scale])
     seed = int(randomize_seed_fn())
-    generator = torch.Generator(device='cuda').manual_seed(seed)
-
+    generator = torch.Generator(device='cpu').manual_seed(seed)
+    torch.set_float32_matmul_precision("highest")
     pipe.text_encoder=text_encoder.to(device=device, dtype=torch.bfloat16)
     pipe.text_encoder_2=text_encoder_2.to(device=device, dtype=torch.bfloat16)
     options = {