use 2 steps unet

app.py

@@ -18,7 +18,7 @@ from sfast.compilers.diffusion_pipeline_compiler import compile, CompilationConf
 BASE = "stabilityai/stable-diffusion-xl-base-1.0"
 REPO = "ByteDance/SDXL-Lightning"
 # 1-step
-CHECKPOINT = "sdxl_lightning_2step_lora.safetensors"
+CHECKPOINT = "sdxl_lightning_2step_unet.safetensors"
 
 # {
 #     "1-Step": ["sdxl_lightning_1step_unet_x0.safetensors", 1],
@@ -39,11 +39,13 @@ print(f"TORCH_COMPILE: {TORCH_COMPILE}")
 print(f"device: {device}")
 
 
+unet = UNet2DConditionModel.from_config(BASE, subfolder="unet").to(
+    "cuda", torch.float16
+)
+unet.load_state_dict(load_file(hf_hub_download(REPO, CHECKPOINT), device="cuda"))
 pipe = StableDiffusionXLPipeline.from_pretrained(
-    BASE, torch_dtype=torch.float16, variant="fp16"
+    BASE, unet=unet, torch_dtype=torch.float16, variant="fp16"
 ).to("cuda")
-pipe.load_lora_weights(hf_hub_download(REPO, CHECKPOINT))
-pipe.fuse_lora()
 
 # Ensure sampler uses "trailing" timesteps.
 pipe.scheduler = EulerDiscreteScheduler.from_config(
@@ -82,7 +84,6 @@ def predict(prompt, seed=1231231):
         guidance_scale=0.0,
         # width=768,
         # height=768,
-        # original_inference_steps=params.lcm_steps,
         output_type="pil",
     )
     print(f"Pipe took {time.time() - last_time} seconds")
@@ -133,23 +134,24 @@ with gr.Blocks(css=css) as demo:
                 """## Running SDXL-Lightning with `diffusers`
 ```py
 import torch
-from diffusers import StableDiffusionXLPipeline, EulerDiscreteScheduler
+from diffusers import StableDiffusionXLPipeline, UNet2DConditionModel, EulerDiscreteScheduler
 from huggingface_hub import hf_hub_download
+from safetensors.torch import load_file
 
 base = "stabilityai/stable-diffusion-xl-base-1.0"
 repo = "ByteDance/SDXL-Lightning"
-ckpt = "sdxl_lightning_4step_lora.safetensors" # Use the correct ckpt for your step setting!
+ckpt = "sdxl_lightning_2step_unet.safetensors" # Use the correct ckpt for your step setting!
 
 # Load model.
-pipe = StableDiffusionXLPipeline.from_pretrained(base, torch_dtype=torch.float16, variant="fp16").to("cuda")
-pipe.load_lora_weights(hf_hub_download(repo, ckpt))
-pipe.fuse_lora()
+unet = UNet2DConditionModel.from_config(base, subfolder="unet").to("cuda", torch.float16)
+unet.load_state_dict(load_file(hf_hub_download(repo, ckpt), device="cuda"))
+pipe = StableDiffusionXLPipeline.from_pretrained(base, unet=unet, torch_dtype=torch.float16, variant="fp16").to("cuda")
 
 # Ensure sampler uses "trailing" timesteps.
 pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing")
 
 # Ensure using the same inference steps as the loaded model and CFG set to 0.
-pipe("A girl smiling", num_inference_steps=4, guidance_scale=0).images[0].save("output.png")
+pipe("A girl smiling", num_inference_steps=2, guidance_scale=0).images[0].save("output.png")
 ```
             """
             )