Update app.py

app.py

@@ -15,7 +15,7 @@ import numpy as np
 from PIL import Image
 import torch
 #import diffusers
-from diffusers import AutoencoderKL, StableDiffusionXLPipeline
+from diffusers import AutoencoderKL, StableDiffusionXLPipeline, UNet2DConditionModel
 from diffusers import EulerAncestralDiscreteScheduler
 from typing import Tuple
 import paramiko
@@ -86,6 +86,7 @@ os.putenv("HF_HUB_ENABLE_HF_TRANSFER","1")
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
     
 def load_and_prepare_model():
+    unet = UNet2DConditionModel.from_pretrained("ford442/RealVisXL_V5.0_BF16", low_cpu_mem_usage=False, subfolder='unet', upcast_attention=True, attention_type='gated-text-image', token=True)
     #vaeRV = AutoencoderKL.from_pretrained("SG161222/RealVisXL_V5.0", subfolder='vae', safety_checker=None, use_safetensors=True, token=True)
     #vaeXL = AutoencoderKL.from_pretrained("stabilityai/sdxl-vae", safety_checker=None, use_safetensors=False, low_cpu_mem_usage=False, torch_dtype=torch.float32, token=True) #.to(device).to(torch.bfloat16) #.to(device=device, dtype=torch.bfloat16)
     vaeXL = AutoencoderKL.from_pretrained("stabilityai/sdxl-vae", low_cpu_mem_usage=False, safety_checker=None, use_safetensors=False, torch_dtype=torch.float32, token=True) #.to(device).to(torch.bfloat16) #.to(device=device, dtype=torch.bfloat16)
@@ -100,6 +101,7 @@ def load_and_prepare_model():
         add_watermarker=False,
         #text_encoder=None,
         #text_encoder_2=None,
+        unet=unet,
         vae=None,
     )
     #pipe.vae = vaeXL #.to(torch.bfloat16)
@@ -122,8 +124,8 @@ def load_and_prepare_model():
     
 pipe = load_and_prepare_model()
 
-text_encoder=CLIPTextModel.from_pretrained('ford442/RealVisXL_V5.0_BF16', subfolder='text_encoder',token=True)#.to(device=device, dtype=torch.bfloat16)
-text_encoder_2=CLIPTextModelWithProjection.from_pretrained('ford442/RealVisXL_V5.0_BF16', subfolder='text_encoder_2',token=True)#.to(device=device, dtype=torch.bfloat16)
+text_encoder=CLIPTextModel.from_pretrained('ford442/RealVisXL_V5.0_BF16', low_cpu_mem_usage=False, subfolder='text_encoder',token=True)#.to(device=device, dtype=torch.bfloat16)
+text_encoder_2=CLIPTextModelWithProjection.from_pretrained('ford442/RealVisXL_V5.0_BF16', low_cpu_mem_usage=False, subfolder='text_encoder_2',token=True)#.to(device=device, dtype=torch.bfloat16)
    
 MAX_SEED = np.iinfo(np.int32).max