Files changed (1) hide show
  1. README.md +6 -6
README.md CHANGED
@@ -45,8 +45,8 @@ controlnet = ControlNetModel.from_pretrained(
45
  variant="fp16",
46
  use_safetensors=True,
47
  torch_dtype=torch.float16,
48
- ).to("cuda")
49
- vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16).to("cuda")
50
  pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
51
  "stabilityai/stable-diffusion-xl-base-1.0",
52
  controlnet=controlnet,
@@ -54,7 +54,7 @@ pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
54
  variant="fp16",
55
  use_safetensors=True,
56
  torch_dtype=torch.float16,
57
- ).to("cuda")
58
  pipe.enable_model_cpu_offload()
59
 
60
  def get_depth_map(image):
@@ -92,7 +92,7 @@ images[0]
92
  images[0].save(f"stormtrooper.png")
93
  ```
94
 
95
- To more details, check out the official documentation of [`StableDiffusionXLControlNetPipeline`](https://huggingface.co/docs/diffusers/main/en/api/pipelines/controlnet_sdxl).
96
 
97
  ### Training
98
 
@@ -102,10 +102,10 @@ Our training script was built on top of the official training script that we pro
102
  The model is trained on 3M image-text pairs from LAION-Aesthetics V2. The model is trained for 700 GPU hours on 80GB A100 GPUs.
103
 
104
  #### Batch size
105
- Data parallel with a single gpu batch size of 8 for a total batch size of 256.
106
 
107
  #### Hyper Parameters
108
- Constant learning rate of 1e-5.
109
 
110
  #### Mixed precision
111
  fp16
 
45
  variant="fp16",
46
  use_safetensors=True,
47
  torch_dtype=torch.float16,
48
+ )
49
+ vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
50
  pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
51
  "stabilityai/stable-diffusion-xl-base-1.0",
52
  controlnet=controlnet,
 
54
  variant="fp16",
55
  use_safetensors=True,
56
  torch_dtype=torch.float16,
57
+ )
58
  pipe.enable_model_cpu_offload()
59
 
60
  def get_depth_map(image):
 
92
  images[0].save(f"stormtrooper.png")
93
  ```
94
 
95
+ For more details, check out the official documentation of [`StableDiffusionXLControlNetPipeline`](https://huggingface.co/docs/diffusers/main/en/api/pipelines/controlnet_sdxl).
96
 
97
  ### Training
98
 
 
102
  The model is trained on 3M image-text pairs from LAION-Aesthetics V2. The model is trained for 700 GPU hours on 80GB A100 GPUs.
103
 
104
  #### Batch size
105
+ Data parallel with a single GPU batch size of 8 for a total batch size of 256.
106
 
107
  #### Hyper Parameters
108
+ The constant learning rate of 1e-5.
109
 
110
  #### Mixed precision
111
  fp16