Mirrors
/
diffusers
mirror of https://github.com/ShivamShrirao/diffusers.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

[Tests] Fix SD slow tests (#364) 

move to fp16, update ddim
This commit is contained in:
Anton Lozhkov 2022-09-06 17:01:04 +02:00 committed by GitHub
parent f085d2f5c6
commit 7a1229fa29
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 42 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

65
tests/test_pipelines.py
View File

@ -917,7 +917,7 @@ class PipelineTesterMixin(unittest.TestCase):
image_slice = image[0, -3:, -3:, -1] image_slice = image[0, -3:, -3:, -1]
assert image.shape == (1, 512, 512, 3) assert image.shape == (1, 512, 512, 3)
expected_slice = np.array([0.8354, 0.83, 0.866, 0.838, 0.8315, 0.867, 0.836, 0.8584, 0.869]) expected_slice = np.array([0.9326, 0.923, 0.951, 0.9365, 0.9214, 0.951, 0.9365, 0.9414, 0.918])
assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3 assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3
@slow @slow
@ -1054,19 +1054,24 @@ class PipelineTesterMixin(unittest.TestCase):
output_image = ds[0]["image"].resize((768, 512)) output_image = ds[0]["image"].resize((768, 512))
model_id = "CompVis/stable-diffusion-v1-4" model_id = "CompVis/stable-diffusion-v1-4"
pipe = StableDiffusionImg2ImgPipeline.from_pretrained(model_id, use_auth_token=True) pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
model_id,
revision="fp16", # fp16 to infer 768x512 images with 16GB of VRAM
torch_dtype=torch.float16,
use_auth_token=True,
)
pipe.to(torch_device) pipe.to(torch_device)
pipe.set_progress_bar_config(disable=None) pipe.set_progress_bar_config(disable=None)
prompt = "A fantasy landscape, trending on artstation" prompt = "A fantasy landscape, trending on artstation"
generator = torch.Generator(device=torch_device).manual_seed(0) generator = torch.Generator(device=torch_device).manual_seed(0)
image = pipe(prompt=prompt, init_image=init_image, strength=0.75, guidance_scale=7.5, generator=generator)[ with torch.autocast("cuda"):
"sample" output = pipe(prompt=prompt, init_image=init_image, strength=0.75, guidance_scale=7.5, generator=generator)
][0] image = output.images[0]
expected_array = np.array(output_image) expected_array = np.array(output_image) / 255.0
sampled_array = np.array(image) sampled_array = np.array(image) / 255.0
assert sampled_array.shape == (512, 768, 3) assert sampled_array.shape == (512, 768, 3)
assert np.max(np.abs(sampled_array - expected_array)) < 1e-4 assert np.max(np.abs(sampled_array - expected_array)) < 1e-4
@ -1082,25 +1087,32 @@ class PipelineTesterMixin(unittest.TestCase):
lms = LMSDiscreteScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear") lms = LMSDiscreteScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear")
model_id = "CompVis/stable-diffusion-v1-4" model_id = "CompVis/stable-diffusion-v1-4"
pipe = StableDiffusionImg2ImgPipeline.from_pretrained(model_id, scheduler=lms, use_auth_token=True) pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
model_id,
scheduler=lms,
revision="fp16", # fp16 to infer 768x512 images with 16GB of VRAM
torch_dtype=torch.float16,
use_auth_token=True,
)
pipe.to(torch_device) pipe.to(torch_device)
pipe.set_progress_bar_config(disable=None) pipe.set_progress_bar_config(disable=None)
prompt = "A fantasy landscape, trending on artstation" prompt = "A fantasy landscape, trending on artstation"
generator = torch.Generator(device=torch_device).manual_seed(0) generator = torch.Generator(device=torch_device).manual_seed(0)
output = pipe(prompt=prompt, init_image=init_image, strength=0.75, guidance_scale=7.5, generator=generator) with torch.autocast("cuda"):
output = pipe(prompt=prompt, init_image=init_image, strength=0.75, guidance_scale=7.5, generator=generator)
image = output.images[0] image = output.images[0]
expected_array = np.array(output_image) expected_array = np.array(output_image) / 255.0
sampled_array = np.array(image) sampled_array = np.array(image) / 255.0
assert sampled_array.shape == (512, 768, 3) assert sampled_array.shape == (512, 768, 3)
assert np.max(np.abs(sampled_array - expected_array)) < 1e-4 assert np.max(np.abs(sampled_array - expected_array)) < 1e-4
@slow @slow
@unittest.skipIf(torch_device == "cpu", "Stable diffusion is supposed to run on GPU") @unittest.skipIf(torch_device == "cpu", "Stable diffusion is supposed to run on GPU")
def test_stable_diffusion_in_paint_pipeline(self): def test_stable_diffusion_inpaint_pipeline(self):
ds = load_dataset("hf-internal-testing/diffusers-images", split="train") ds = load_dataset("hf-internal-testing/diffusers-images", split="train")
init_image = ds[3]["image"].resize((768, 512)) init_image = ds[3]["image"].resize((768, 512))
@ -1108,24 +1120,31 @@ class PipelineTesterMixin(unittest.TestCase):
output_image = ds[5]["image"].resize((768, 512)) output_image = ds[5]["image"].resize((768, 512))
model_id = "CompVis/stable-diffusion-v1-4" model_id = "CompVis/stable-diffusion-v1-4"
pipe = StableDiffusionInpaintPipeline.from_pretrained(model_id, use_auth_token=True) pipe = StableDiffusionInpaintPipeline.from_pretrained(
model_id,
revision="fp16", # fp16 to infer 768x512 images in 16GB of VRAM
torch_dtype=torch.float16,
use_auth_token=True,
)
pipe.to(torch_device) pipe.to(torch_device)
pipe.set_progress_bar_config(disable=None) pipe.set_progress_bar_config(disable=None)
prompt = "A red cat sitting on a parking bench" prompt = "A red cat sitting on a parking bench"
generator = torch.Generator(device=torch_device).manual_seed(0) generator = torch.Generator(device=torch_device).manual_seed(0)
image = pipe( with torch.autocast("cuda"):
prompt=prompt, output = pipe(
init_image=init_image, prompt=prompt,
mask_image=mask_image, init_image=init_image,
strength=0.75, mask_image=mask_image,
guidance_scale=7.5, strength=0.75,
generator=generator, guidance_scale=7.5,
).images[0] generator=generator,
)
image = output.images[0]
expected_array = np.array(output_image) expected_array = np.array(output_image) / 255.0
sampled_array = np.array(image) sampled_array = np.array(image) / 255.0
assert sampled_array.shape == (512, 768, 3) assert sampled_array.shape == (512, 768, 3)
assert np.max(np.abs(sampled_array - expected_array)) < 1e-3 assert np.max(np.abs(sampled_array - expected_array)) < 1e-3