update

app.py

@@ -62,10 +62,13 @@ with open("defaults_data.json", "r") as file:
 #device = "cuda" 
 if torch.cuda.is_available():
     device = "cuda"
+    dtype = torch.float16
 elif torch.backends.mps.is_available():
     device = "mps"
+    dtype = torch.float32
 else:
     device = "cpu"
+    dtype = torch.float32
 
 state_dicts = {}
 
@@ -118,13 +121,13 @@ controlnet_path = f'data/checkpoints/ControlNetModel'
 
 # load IdentityNet
 st = time.time()
-identitynet = ControlNetModel.from_pretrained(controlnet_path, torch_dtype=torch.float16)
-zoedepthnet = ControlNetModel.from_pretrained("diffusers/controlnet-zoe-depth-sdxl-1.0",torch_dtype=torch.float16)
+identitynet = ControlNetModel.from_pretrained(controlnet_path, torch_dtype=dtype)
+zoedepthnet = ControlNetModel.from_pretrained("diffusers/controlnet-zoe-depth-sdxl-1.0",torch_dtype=dtype)
 et = time.time()
 elapsed_time = et - st
 print('Loading ControlNet took: ', elapsed_time, 'seconds')
 st = time.time()
-vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
+vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=dtype)
 et = time.time()
 elapsed_time = et - st
 print('Loading VAE took: ', elapsed_time, 'seconds')
@@ -132,7 +135,7 @@ st = time.time()
 pipe = StableDiffusionXLInstantIDImg2ImgPipeline.from_pretrained("rubbrband/albedobaseXL_v21",
                                                                  vae=vae,
                                                                  controlnet=[identitynet, zoedepthnet],
-                                                                 torch_dtype=torch.float16)
+                                                                 torch_dtype=dtype)
 pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config, use_karras_sigmas=True)
 pipe.load_ip_adapter_instantid(face_adapter)
 pipe.set_ip_adapter_scale(0.8)
@@ -225,7 +228,7 @@ def merge_incompatible_lora(full_path_lora, lora_scale):
                     for_inference=True,
                 )
                 lora_model.merge_to(
-                    pipe.text_encoder, pipe.unet, weights_sd, torch.float16, "cuda"
+                    pipe.text_encoder, pipe.unet, weights_sd, dtype, device
                 )
                 del weights_sd
                 del lora_model