make style

1

@@ -1,26 +0,0 @@
-#!/usr/bin/env python3
-from diffusers import UNet2DConditionModel
-import torch
-
-unet = UNet2DConditionModel.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", subfolder="unet", variant="fp16", torch_dtype=torch.float16)
-unet.train()
-unet.enable_gradient_checkpointing()
-unet = unet.to("cuda:1")
-
-batch_size = 8
-
-sample = torch.randn((1, 4, 128, 128)).half().to(unet.device).repeat(batch_size, 1, 1, 1)
-time_ids = (torch.arange(6) / 6)[None, :].half().to(unet.device).repeat(batch_size, 1)
-encoder_hidden_states = torch.randn((1, 77, 2048)).half().to(unet.device).repeat(batch_size, 1, 1)
-text_embeds = torch.randn((1, 1280)).half().to(unet.device).repeat(batch_size, 1)
-
-out = unet(sample, 1.0, added_cond_kwargs={"time_ids": time_ids, "text_embeds": text_embeds}, encoder_hidden_states=encoder_hidden_states).sample
-
-loss = ((out - sample) ** 2).mean()
-loss.backward()
-
-print(torch.cuda.max_memory_allocated(device=unet.device))
-
-
-# no gradient checkpointing: 12,276,695,552
-# curr gradient checkpointing: 10,862,276,096

run_local.py

@@ -13,13 +13,13 @@ from io import BytesIO
 
 path = sys.argv[1]
 # path = "ptx0/pseudo-journey-v2"
-# path = "stabilityai/stable-diffusion-2-1"
 
 api = HfApi()
 start_time = time.time()
 pipe = StableDiffusionPipeline.from_pretrained(path, torch_dtype=torch.float16)
+pipe.enable_xformers_memory_efficient_attention()
 
-pipe.unet = torch.compile(pipe.unet)
+# pipe.unet = torch.compile(pipe.unet)
 
 # pipe = StableDiffusionImg2ImgPipeline.from_pretrained(path, torch_dtype=torch.float16, safety_checker=None)
 
@@ -45,7 +45,7 @@ prompt = "A lion in galaxies, spirals, nebulae, stars, smoke, iridescent, intric
 
 generator = torch.Generator(device="cpu").manual_seed(0)
 # images = pipe(prompt=prompt, image=image, generator=generator, num_images_per_prompt=4, num_inference_steps=25).images
-images = pipe(prompt=prompt, generator=generator, num_images_per_prompt=4, num_inference_steps=25).images
+images = pipe(prompt=prompt, generator=generator, num_images_per_prompt=1, num_inference_steps=50).images
 
 for i, image in enumerate(images):
     file_name = f"bb_1_{i}"

run_local_fuse_xl.py

@@ -13,14 +13,14 @@ import torch
 
 pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16)
 pipe.load_lora_weights("stabilityai/stable-diffusion-xl-base-1.0", weight_name="sd_xl_offset_example-lora_1.0.safetensors")
-pipe.unet.fuse_lora()
+# pipe.unet.fuse_lora()
 
 pipe.to(torch_dtype=torch.float16)
 pipe.to("cuda")
 
-torch.manual_seed(0)
+torch.manual_seed(33)
 
-prompt = "beautiful scenery nature glass bottle landscape, , purple galaxy bottle"
+prompt = "beautiful scenery nature glass bottle landscape, purple galaxy bottle"
 negative_prompt = "text, watermark"
 
 image = pipe(prompt, negative_prompt=negative_prompt, num_inference_steps=25).images[0]

run_local_xl.py

@@ -12,12 +12,13 @@ from pathlib import Path
 import requests
 from PIL import Image
 from io import BytesIO
+import xformers
 
 api = HfApi()
 start_time = time.time()
 
 # use_refiner = bool(int(sys.argv[1]))
-use_refiner = True
+use_refiner = False
 use_diffusers = True
 path = "stabilityai/stable-diffusion-xl-base-1.0"
 refiner_path = "stabilityai/stable-diffusion-xl-refiner-1.0"
@@ -27,6 +28,7 @@ vae = AutoencoderKL.from_pretrained(vae_path, torch_dtype=torch.float16, force_u
 if use_diffusers:
     # pipe = StableDiffusionXLPipeline.from_pretrained(path, vae=vae, torch_dtype=torch.float16, variant="fp16", use_safetensors=True, local_files_only=True)
     pipe = StableDiffusionXLPipeline.from_pretrained(path, torch_dtype=torch.float16, vae=vae, variant="fp16", use_safetensors=True, local_files_only=True, add_watermarker=False)
+    # pipe.enable_xformers_memory_efficient_attention()
     print(time.time() - start_time)
     pipe.to("cuda")
 

run_lora.py

@@ -1,5 +1,5 @@
 #!/usr/bin/env python3
-from diffusers import StableDiffusionPipeline, KDPM2DiscreteScheduler, StableDiffusionImg2ImgPipeline, HeunDiscreteScheduler, KDPM2AncestralDiscreteScheduler, DDIMScheduler,  DPMSolverMultistepScheduler
+from diffusers import AutoPipelineForText2Image, StableDiffusionPipeline, KDPM2DiscreteScheduler, StableDiffusionImg2ImgPipeline, HeunDiscreteScheduler, KDPM2AncestralDiscreteScheduler, DDIMScheduler,  DPMSolverMultistepScheduler
 import time
 import os
 from huggingface_hub import HfApi
@@ -11,20 +11,20 @@ import requests
 from PIL import Image
 from io import BytesIO
 
-path = "runwayml/stable-diffusion-v1-5"
-lora_id = "takuma104/lora-test-text-encoder-lora-target"
+path = "stabilityai/stable-diffusion-xl-base-0.9"
 
 api = HfApi()
 start_time = time.time()
-pipe = StableDiffusionPipeline.from_pretrained(path, torch_dtype=torch.float16)
-pipe.load_lora_weights(lora_id)
-pipe = pipe.to("cuda")
+pipe = AutoPipelineForText2Image.from_pretrained(path, torch_dtype=torch.float16)
+pipe.enable_model_cpu_offload()
+lora_model_id = "hf-internal-testing/sdxl-0.9-kamepan-lora"
+lora_filename = "kame_sdxl_v2-000020-16rank.safetensors"
+pipe.load_lora_weights(lora_model_id, weight_name=lora_filename)
 
-prompt = "a red sks dog"
+prompt = "masterpiece, best quality, mountain"
 
 images = pipe(prompt=prompt, 
-    num_inference_steps=15, 
-    cross_attention_kwargs={"scale": 0.5},
+    num_inference_steps=20, 
     generator=torch.manual_seed(0)
 ).images