photomaker

.gitignore

@@ -6,5 +6,7 @@ __pycache__/
 .idea/
 *.bin
 *.safetensors
-SG16122/
-*vscode*
+SG161222/
+*vscode*
\ No newline at end of file
+prof_analyze/
+profiler*
\ No newline at end of file

Dockerfile

-FROM image.sourcefind.cn:5000/dcu/admin/base/pytorch:2.1.0-centos7.6-dtk23.10.1-py38
+FROM image.sourcefind.cn:5000/dcu/admin/base/pytorch:1.13.1-centos7.6-dtk-23.04.1-py39-latest
\ No newline at end of file

README.md

@@ -28,7 +28,7 @@
 
 ### Docker（方法一）
 
-    docker pull image.sourcefind.cn:5000/dcu/admin/base/pytorch:2.1.0-centos7.6-dtk23.10.1-py38
+    docker pull image.sourcefind.cn:5000/dcu/admin/base/pytorch:1.13.1-centos7.6-dtk-23.04.1-py39-latest
     docker run --shm-size 10g --network=host --name=photomaker --privileged --device=/dev/kfd --device=/dev/dri --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -v /opt/hyhal:/opt/hyhal-v 项目地址(绝对路径):/home/ -it <your IMAGE ID> bash
 
     pip install -r requirements.txt

gradio_demo/app.py

@@ -38,6 +38,7 @@ if device == "mps":
     torch_dtype = torch.float16
 else:
     torch_dtype = torch.float16
+    
 pipe = PhotoMakerStableDiffusionXLPipeline.from_pretrained(
     base_model_path, 
     torch_dtype=torch_dtype,
@@ -58,10 +59,10 @@ pipe.id_encoder.to(device)
 pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config)
 
 print("Loading lora...")
-lora_path = "./lora/ral-opal-sdxl.safetensors"
-pipe.load_lora_weights(os.path.dirname(lora_path), weight_name=os.path.basename(lora_path), adapter_name="xl_more_art-full")
-pipe.set_adapters(["photomaker", "xl_more_art-full"], adapter_weights=[1.0, 0.5])
-# pipe.set_adapters(["photomaker"], adapter_weights=[1.0])
+# lora_path = "./lora/ral-opal-sdxl.safetensors"
+# pipe.load_lora_weights(os.path.dirname(lora_path), weight_name=os.path.basename(lora_path), adapter_name="xl_more_art-full")
+# pipe.set_adapters(["photomaker", "xl_more_art-full"], adapter_weights=[1.0, 0.5])
+pipe.set_adapters(["photomaker"], adapter_weights=[1.0])
 pipe.fuse_lora()
 
 @spaces.GPU(enable_queue=True)
@@ -93,10 +94,7 @@ def generate_image(upload_images, prompt, negative_prompt, style_name, num_steps
     if start_merge_step > 30:
         start_merge_step = 30
     print(start_merge_step)
-    # with torch.profiler.profile(activities=[
-    #     torch.profiler.ProfilerActivity.CPU,
-    #     torch.profiler.ProfilerActivity.CUDA,
-    # ]) as p:
+
     images = pipe(
         prompt=prompt,
         input_id_images=input_id_images,
@@ -107,8 +105,7 @@ def generate_image(upload_images, prompt, negative_prompt, style_name, num_steps
         generator=generator,
         guidance_scale=guidance_scale,
     ).images
-    # print(p.key_averages().table(sort_by="self_cuda_time_total", row_limit=-1))
-    # p.export_chrome_trace("profiler_results_dcu61.json")
+
     return images, gr.update(visible=True)
 
 def swap_to_gallery(images):

photomaker/pipeline.py

@@ -4,6 +4,8 @@ import os
 import PIL
 import numpy as np 
 
+import time
+
 import torch
 from torchvision import transforms as T
 
@@ -469,12 +471,13 @@ class PhotoMakerStableDiffusionXLPipeline(StableDiffusionXLPipeline):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
                         callback(i, t, latents)
-
+                        
         # make sure the VAE is in float32 mode, as it overflows in float16
-        if self.vae.dtype == torch.float16 and self.vae.config.force_upcast:
-            self.upcast_vae()
-            latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+        # if self.vae.dtype == torch.float16 and self.vae.config.force_upcast:
+        #     self.upcast_vae()
+        #     latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
 
+        latents = latents.to(memory_format=torch.channels_last)
         if not output_type == "latent":
             image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
         else: