Update app.py

app.py

@@ -106,7 +106,6 @@ def apply_style(style_name: str, positive: str, negative: str = "") -> Tuple[str
         negative = ""
     return p.replace("{prompt}", positive), n + negative
     
-
 def load_and_prepare_model():
     unetX = UNet2DConditionModel.from_pretrained('ford442/RealVisXL_V5.0_BF16', subfolder='unet', low_cpu_mem_usage=False, token=True) #.to(device).to(torch.bfloat16) #.to(device=device, dtype=torch.bfloat16)
     vaeX = 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)
@@ -121,7 +120,7 @@ def load_and_prepare_model():
         text_encoder=None,
         text_encoder_2=None,
         vae=None,
-        unet=unetX,
+        unet=None,
     )
     
     '''
@@ -138,8 +137,9 @@ def load_and_prepare_model():
             `force_upcast` can be set to `False` - see: https://huggingface.co/madebyollin/sdxl-vae-fp16-fix
         
     '''
-    pipe.vae=vaeX
+    #pipe.vae=vaeX
     pipe.to(device=device, dtype=torch.bfloat16)
+    pipe.vae=vaeX
     #pipe.vae.to(device=device, dtype=torch.bfloat16)
     #pipe.vae.do_resize=False
     #pipe.vae.do_rescale=False
@@ -174,7 +174,7 @@ txt_tokenizer.tokenizer_legacy=False
 model = Phi3ForCausalLM.from_pretrained(checkpoint).to('cuda:0')
 #model = AutoModelForCausalLM.from_pretrained(checkpoint, device_map='cuda') #.to('cuda')
 
-ip_model = IPAdapterXL(pipe, local_folder, ip_ckpt, device)
+#ip_model = IPAdapterXL(pipe, local_folder, ip_ckpt, device)
 text_encoder_1=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)
    
@@ -448,7 +448,8 @@ def generate_30(
         global text_encoder_2
         pipe.text_encoder=text_encoder_1.to(device=device, dtype=torch.bfloat16)
         pipe.text_encoder_2=text_encoder_2.to(device=device, dtype=torch.bfloat16)
-        #pipe.unet=unetX.to(device=device, dtype=torch.bfloat16)
+        pipe.unet=unetX.to(device=device, dtype=torch.bfloat16)
+        ip_model = IPAdapterXL(pipe, local_folder, ip_ckpt, device)
         print('-- generating image --')
         sd_image = ip_model.generate(
                 pil_image_1=sd_image_a,
@@ -599,7 +600,8 @@ def generate_60(
         global text_encoder_2
         pipe.text_encoder=text_encoder_1.to(device=device, dtype=torch.bfloat16)
         pipe.text_encoder_2=text_encoder_2.to(device=device, dtype=torch.bfloat16)
-        #pipe.unet=unetX.to(device=device, dtype=torch.bfloat16)
+        pipe.unet=unetX.to(device=device, dtype=torch.bfloat16)
+        ip_model = IPAdapterXL(pipe, local_folder, ip_ckpt, device)
         print('-- generating image --')
         sd_image = ip_model.generate(
                 pil_image_1=sd_image_a,
@@ -750,7 +752,8 @@ def generate_90(
         global text_encoder_2
         pipe.text_encoder=text_encoder_1.to(device=device, dtype=torch.bfloat16)
         pipe.text_encoder_2=text_encoder_2.to(device=device, dtype=torch.bfloat16)
-        #pipe.unet=unetX.to(device=device, dtype=torch.bfloat16)
+        pipe.unet=unetX.to(device=device, dtype=torch.bfloat16)
+        ip_model = IPAdapterXL(pipe, local_folder, ip_ckpt, device)
         print('-- generating image --')
         sd_image = ip_model.generate(
                 pil_image_1=sd_image_a,