Skip to content

[Tests] Make sure tests are on GPU #269

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.

Sign up for GitHub

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

Merged
patrickvonplaten merged 3 commits into from
Aug 29, 2022
Merged

[Tests] Make sure tests are on GPU #269

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
3eafc95
[Tests] Make sure tests are on GPU
patrickvonplaten Aug 29, 2022
422b1cc
move more models
patrickvonplaten Aug 29, 2022
4abccc8
speed up tests
patrickvonplaten Aug 29, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
0 tests/test_modeling_utils.py → tests/test_config.py
View file
Open in desktop
File renamed without changes.
9 changes: 7 additions & 2 deletions tests/test_models_unet.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,9 @@
from .test_modeling_common import ModelTesterMixin


torch.backends.cuda.matmul.allow_tf32 = False


class UnetModelTests(ModelTesterMixin, unittest.TestCase):
model_class = UNet2DModel

Expand Down Expand Up @@ -133,18 +136,20 @@ def test_from_pretrained_hub(self):
def test_output_pretrained(self):
model = UNet2DModel.from_pretrained("fusing/unet-ldm-dummy-update")
model.eval()
model.to(torch_device)

torch.manual_seed(0)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(0)

noise = torch.randn(1, model.config.in_channels, model.config.sample_size, model.config.sample_size)
time_step = torch.tensor([10] * noise.shape[0])
noise = noise.to(torch_device)
time_step = torch.tensor([10] * noise.shape[0]).to(torch_device)

with torch.no_grad():
output = model(noise, time_step)["sample"]

output_slice = output[0, -1, -3:, -3:].flatten()
output_slice = output[0, -1, -3:, -3:].flatten().cpu()
# fmt: off
expected_output_slice = torch.tensor([-13.3258, -20.1100, -15.9873, -17.6617, -23.0596, -17.9419, -13.3675, -16.1889, -12.3800])
# fmt: on
Expand Down
7 changes: 6 additions & 1 deletion tests/test_models_vae.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,9 @@
from .test_modeling_common import ModelTesterMixin


torch.backends.cuda.matmul.allow_tf32 = False


class AutoencoderKLTests(ModelTesterMixin, unittest.TestCase):
model_class = AutoencoderKL

Expand Down Expand Up @@ -74,17 +77,19 @@ def test_from_pretrained_hub(self):

def test_output_pretrained(self):
model = AutoencoderKL.from_pretrained("fusing/autoencoder-kl-dummy")
model = model.to(torch_device)
model.eval()

torch.manual_seed(0)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(0)

image = torch.randn(1, model.config.in_channels, model.config.sample_size, model.config.sample_size)
image = image.to(torch_device)
with torch.no_grad():
output = model(image, sample_posterior=True)

output_slice = output[0, -1, -3:, -3:].flatten()
output_slice = output[0, -1, -3:, -3:].flatten().cpu()
# fmt: off
expected_output_slice = torch.tensor([-4.0078e-01, -3.8304e-04, -1.2681e-01, -1.1462e-01, 2.0095e-01, 1.0893e-01, -8.8248e-02, -3.0361e-01, -9.8646e-03])
# fmt: on
Expand Down
8 changes: 6 additions & 2 deletions tests/test_models_vq.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,9 @@
from .test_modeling_common import ModelTesterMixin


torch.backends.cuda.matmul.allow_tf32 = False


class VQModelTests(ModelTesterMixin, unittest.TestCase):
model_class = VQModel

Expand Down Expand Up @@ -73,17 +76,18 @@ def test_from_pretrained_hub(self):

def test_output_pretrained(self):
model = VQModel.from_pretrained("fusing/vqgan-dummy")
model.eval()
model.to(torch_device).eval()

torch.manual_seed(0)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(0)

image = torch.randn(1, model.config.in_channels, model.config.sample_size, model.config.sample_size)
image = image.to(torch_device)
with torch.no_grad():
output = model(image)

output_slice = output[0, -1, -3:, -3:].flatten()
output_slice = output[0, -1, -3:, -3:].flatten().cpu()
# fmt: off
expected_output_slice = torch.tensor([-0.0153, -0.4044, -0.1880, -0.5161, -0.2418, -0.4072, -0.1612, -0.0633, -0.0143])
# fmt: on
Expand Down
37 changes: 28 additions & 9 deletions tests/test_pipelines.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,10 +59,12 @@ def test_from_pretrained_save_pretrained(self):
schedular = DDPMScheduler(num_train_timesteps=10)

ddpm = DDPMPipeline(model, schedular)
ddpm.to(torch_device)

with tempfile.TemporaryDirectory() as tmpdirname:
ddpm.save_pretrained(tmpdirname)
new_ddpm = DDPMPipeline.from_pretrained(tmpdirname)
new_ddpm.to(torch_device)

generator = torch.manual_seed(0)

Expand All @@ -76,11 +78,12 @@ def test_from_pretrained_save_pretrained(self):
def test_from_pretrained_hub(self):
model_path = "google/ddpm-cifar10-32"

ddpm = DDPMPipeline.from_pretrained(model_path)
ddpm_from_hub = DiffusionPipeline.from_pretrained(model_path)
scheduler = DDPMScheduler(num_train_timesteps=10)

ddpm.scheduler.num_timesteps = 10
ddpm_from_hub.scheduler.num_timesteps = 10
ddpm = DDPMPipeline.from_pretrained(model_path, scheduler=scheduler)
ddpm.to(torch_device)
ddpm_from_hub = DiffusionPipeline.from_pretrained(model_path, scheduler=scheduler)
ddpm_from_hub.to(torch_device)

generator = torch.manual_seed(0)

Expand All @@ -94,14 +97,15 @@ def test_from_pretrained_hub(self):
def test_from_pretrained_hub_pass_model(self):
model_path = "google/ddpm-cifar10-32"

scheduler = DDPMScheduler(num_train_timesteps=10)

Comment on lines +100 to +101
Copy link
Member

@anton-l anton-l Aug 29, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice!

# pass unet into DiffusionPipeline
unet = UNet2DModel.from_pretrained(model_path)
ddpm_from_hub_custom_model = DiffusionPipeline.from_pretrained(model_path, unet=unet)

ddpm_from_hub = DiffusionPipeline.from_pretrained(model_path)
ddpm_from_hub_custom_model = DiffusionPipeline.from_pretrained(model_path, unet=unet, scheduler=scheduler)
ddpm_from_hub_custom_model.to(torch_device)

ddpm_from_hub_custom_model.scheduler.num_timesteps = 10
ddpm_from_hub.scheduler.num_timesteps = 10
ddpm_from_hub = DiffusionPipeline.from_pretrained(model_path, scheduler=scheduler)
ddpm_from_hub.to(torch_device)

generator = torch.manual_seed(0)

Expand All @@ -116,6 +120,7 @@ def test_output_format(self):
model_path = "google/ddpm-cifar10-32"

pipe = DDIMPipeline.from_pretrained(model_path)
pipe.to(torch_device)

generator = torch.manual_seed(0)
images = pipe(generator=generator, output_type="numpy")["sample"]
Expand All @@ -141,6 +146,7 @@ def test_ddpm_cifar10(self):
scheduler = scheduler.set_format("pt")

ddpm = DDPMPipeline(unet=unet, scheduler=scheduler)
ddpm.to(torch_device)

generator = torch.manual_seed(0)
image = ddpm(generator=generator, output_type="numpy")["sample"]
Expand All @@ -159,6 +165,7 @@ def test_ddim_lsun(self):
scheduler = DDIMScheduler.from_config(model_id)

ddpm = DDIMPipeline(unet=unet, scheduler=scheduler)
ddpm.to(torch_device)

generator = torch.manual_seed(0)
image = ddpm(generator=generator, output_type="numpy")["sample"]
Expand All @@ -177,6 +184,7 @@ def test_ddim_cifar10(self):
scheduler = DDIMScheduler(tensor_format="pt")

ddim = DDIMPipeline(unet=unet, scheduler=scheduler)
ddim.to(torch_device)

generator = torch.manual_seed(0)
image = ddim(generator=generator, eta=0.0, output_type="numpy")["sample"]
Expand All @@ -195,6 +203,7 @@ def test_pndm_cifar10(self):
scheduler = PNDMScheduler(tensor_format="pt")

pndm = PNDMPipeline(unet=unet, scheduler=scheduler)
pndm.to(torch_device)
generator = torch.manual_seed(0)
image = pndm(generator=generator, output_type="numpy")["sample"]

Expand All @@ -207,6 +216,7 @@ def test_pndm_cifar10(self):
@slow
def test_ldm_text2img(self):
ldm = LDMTextToImagePipeline.from_pretrained("CompVis/ldm-text2im-large-256")
ldm.to(torch_device)

prompt = "A painting of a squirrel eating a burger"
generator = torch.manual_seed(0)
Expand All @@ -223,6 +233,7 @@ def test_ldm_text2img(self):
@slow
def test_ldm_text2img_fast(self):
ldm = LDMTextToImagePipeline.from_pretrained("CompVis/ldm-text2im-large-256")
ldm.to(torch_device)

prompt = "A painting of a squirrel eating a burger"
generator = torch.manual_seed(0)
Expand Down Expand Up @@ -290,6 +301,7 @@ def test_score_sde_ve_pipeline(self):
scheduler = ScoreSdeVeScheduler.from_config(model_id)

sde_ve = ScoreSdeVePipeline(unet=model, scheduler=scheduler)
sde_ve.to(torch_device)

torch.manual_seed(0)
image = sde_ve(num_inference_steps=300, output_type="numpy")["sample"]
Expand All @@ -304,6 +316,7 @@ def test_score_sde_ve_pipeline(self):
@slow
def test_ldm_uncond(self):
ldm = LDMPipeline.from_pretrained("CompVis/ldm-celebahq-256")
ldm.to(torch_device)

generator = torch.manual_seed(0)
image = ldm(generator=generator, num_inference_steps=5, output_type="numpy")["sample"]
Expand All @@ -323,7 +336,9 @@ def test_ddpm_ddim_equality(self):
ddim_scheduler = DDIMScheduler(tensor_format="pt")

ddpm = DDPMPipeline(unet=unet, scheduler=ddpm_scheduler)
ddpm.to(torch_device)
ddim = DDIMPipeline(unet=unet, scheduler=ddim_scheduler)
ddim.to(torch_device)

generator = torch.manual_seed(0)
ddpm_image = ddpm(generator=generator, output_type="numpy")["sample"]
Expand All @@ -343,7 +358,10 @@ def test_ddpm_ddim_equality_batched(self):
ddim_scheduler = DDIMScheduler(tensor_format="pt")

ddpm = DDPMPipeline(unet=unet, scheduler=ddpm_scheduler)
ddpm.to(torch_device)

ddim = DDIMPipeline(unet=unet, scheduler=ddim_scheduler)
ddim.to(torch_device)

generator = torch.manual_seed(0)
ddpm_images = ddpm(batch_size=4, generator=generator, output_type="numpy")["sample"]
Expand All @@ -363,6 +381,7 @@ def test_karras_ve_pipeline(self):
scheduler = KarrasVeScheduler(tensor_format="pt")

pipe = KarrasVePipeline(unet=model, scheduler=scheduler)
pipe.to(torch_device)

generator = torch.manual_seed(0)
image = pipe(num_inference_steps=20, generator=generator, output_type="numpy")["sample"]
Expand Down