work with f16

app.py

@@ -32,18 +32,24 @@ device = "cuda" if torch.cuda.is_available() else "cpu"
 
 if torch.cuda.is_available():
     power_device = "GPU"
+    torch.cuda.max_memory_allocated(device=device)
 else:
     power_device = "CPU"
 
 scheduler_class = MyEulerAncestralDiscreteScheduler
 
-pipe_inversion = SDXLDDIMPipeline.from_pretrained("stabilityai/sdxl-turbo", use_safetensors=True)  # .to('cpu')
-pipe_inference = AutoPipelineForImage2Image.from_pretrained("stabilityai/sdxl-turbo",
+pipe_inversion = SDXLDDIMPipeline.from_pretrained("stabilityai/sdxl-turbo", torch_dtype=torch.float16, variant="fp16", use_safetensors=True)  # .to('cpu')
+pipe_inference = AutoPipelineForImage2Image.from_pretrained("stabilityai/sdxl-turbo", torch_dtype=torch.float16, variant="fp16",
                                                             use_safetensors=True)  # .to('cpu')
 pipe_inference.scheduler = scheduler_class.from_config(pipe_inference.scheduler.config)
 pipe_inversion.scheduler = scheduler_class.from_config(pipe_inversion.scheduler.config)
 pipe_inversion.scheduler_inference = scheduler_class.from_config(pipe_inference.scheduler.config)
 
+pipe_inference.enable_xformers_memory_efficient_attention()
+pipe_inversion.enable_xformers_memory_efficient_attention()
+
+
+
 # with gr.Blocks(css=css) as demo:
 with gr.Blocks(css="style.css") as demo:
     gr.Markdown(f""" # Real Time Editing with RNRI Inversion 🍎⚡️