Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Does it only support multiple devices and not support acceleration for a single device? I reported an error when using Colab in Controlnet and IP adapter #3

Open
libai-lab opened this issue Jun 17, 2024 · 1 comment

Comments

@libai-lab
Copy link

AssertionError Traceback (most recent call last)
in <cell line: 21>()
19
20 control_image.save("control.png")
---> 21 async_diff = AsyncDiff(pipe, model_n=2, stride=1, time_shift=False)
22
23 async_diff.reset_state(warm_up=1)

/content/AsyncDiff/src/async_sd.py in init(self, pipeline, model_n, stride, warm_up, time_shift)
46 def init(self, pipeline, model_n=2, stride=1, warm_up=1, time_shift=False):
47 dist.init_process_group("nccl")
---> 48 if not dist.get_rank(): assert model_n + stride - 1 == dist.get_world_size(), "[ERROR]: The strategy is not compatible with the number of devices. (model_n + stride - 1) should be equal to world_size."
49 assert stride==1 or stride==2, "[ERROR]: The stride should be set as 1 or 2"
50 self.model_n = model_n

AssertionError: [ERROR]: The strategy is not compatible with the number of devices. (model_n + stride - 1) should be equal to world_size.

from AsyncDiff.src.async_sd import AsyncDiff

start_time=time.time()
image = load_image("https://img0.baidu.com/it/u=1483900031,3964539741&fm=253&fmt=auto&app=138&f=JPEG?w=751&h=500")

ip_image = load_image("https://replicate.delivery/pbxt/Kc2uOvUQ0sVmT6ALE7MIBwOAXlQ6kqBGc2XOz54oKVO2NxvF/style.jpg")

image_orig = image

image = np.array(image)

low_threshold = 100
high_threshold = 200

image = cv2.Canny(image, low_threshold, high_threshold)
image = image[:, :, None]
image = np.concatenate([image, image, image], axis=2)
control_image = Image.fromarray(image)

control_image.save("control.png")
async_diff = AsyncDiff(pipe, model_n=2, stride=1, time_shift=False)

async_diff.reset_state(warm_up=1)
image = pipe(
prompt="a nice room, best quality, highres, wallpaper,",
image=image_orig,
control_image=control_image,
negative_prompt="(low quality, bad quality, worst quality:1.2)",
ip_adapter_image=ip_image,
num_inference_steps=20,
guidance_scale=8,
controlnet_conditioning_scale=0.6,
guess_mode=False,
strength=1.0,
generator=generator,
).images[0]

image.save('./image/dreamshaper8.png')
end_time=time.time()
print("时间",end_time-start_time)

@czg1225
Copy link
Owner

czg1225 commented Jun 18, 2024

Hi @libai-lab ,
As a distributed acceleration strategy, AsyncDiff currently does not support single-device acceleration. We may add support for single-device acceleration in the future.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants