Skip to content

GitLab

Unverified Commit dc85b578 authored by Dhruv Nair's avatar Dhruv Nair Committed by GitHub
Browse files

Move tests for SD inference variant pipelines into their own modules (#6707) 

* update

* update

* update
parent 0d927c75
Expand all Show whitespace changes
Inline Side-by-side
Showing with 0 additions and 1986 deletions
tests/pipelines/stable_diffusion/test_cycle_diffusion.py deleted 100644 → 0
View file @ 0d927c75
# coding=utf-8
# Copyright 2023 HuggingFace Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import gc
import random
import unittest
import numpy as np
import torch
from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer
from diffusers import AutoencoderKL, CycleDiffusionPipeline, DDIMScheduler, UNet2DConditionModel
from diffusers.utils.testing_utils import (
enable_full_determinism,
floats_tensor,
load_image,
load_numpy,
nightly,
require_torch_gpu,
skip_mps,
torch_device,
)
from ..pipeline_params import (
IMAGE_TO_IMAGE_IMAGE_PARAMS,
TEXT_GUIDED_IMAGE_VARIATION_BATCH_PARAMS,
TEXT_GUIDED_IMAGE_VARIATION_PARAMS,
)
from ..test_pipelines_common import PipelineLatentTesterMixin, PipelineTesterMixin
enable_full_determinism()
class CycleDiffusionPipelineFastTests(PipelineLatentTesterMixin, PipelineTesterMixin, unittest.TestCase):
pipeline_class = CycleDiffusionPipeline
params = TEXT_GUIDED_IMAGE_VARIATION_PARAMS - {
"negative_prompt",
"height",
"width",
"negative_prompt_embeds",
}
required_optional_params = PipelineTesterMixin.required_optional_params - {"latents"}
batch_params = TEXT_GUIDED_IMAGE_VARIATION_BATCH_PARAMS.union({"source_prompt"})
image_params = IMAGE_TO_IMAGE_IMAGE_PARAMS
image_latents_params = IMAGE_TO_IMAGE_IMAGE_PARAMS
def get_dummy_components(self):
torch.manual_seed(0)
unet = UNet2DConditionModel(
block_out_channels=(32, 64),
layers_per_block=2,
sample_size=32,
in_channels=4,
out_channels=4,
down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
cross_attention_dim=32,
)
scheduler = DDIMScheduler(
beta_start=0.00085,
beta_end=0.012,
beta_schedule="scaled_linear",
num_train_timesteps=1000,
clip_sample=False,
set_alpha_to_one=False,
)
torch.manual_seed(0)
vae = AutoencoderKL(
block_out_channels=[32, 64],
in_channels=3,
out_channels=3,
down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"],
up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"],
latent_channels=4,
)
torch.manual_seed(0)
text_encoder_config = CLIPTextConfig(
bos_token_id=0,
eos_token_id=2,
hidden_size=32,
intermediate_size=37,
layer_norm_eps=1e-05,
num_attention_heads=4,
num_hidden_layers=5,
pad_token_id=1,
vocab_size=1000,
)
text_encoder = CLIPTextModel(text_encoder_config)
tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")
components = {
"unet": unet,
"scheduler": scheduler,
"vae": vae,
"text_encoder": text_encoder,
"tokenizer": tokenizer,
"safety_checker": None,
"feature_extractor": None,
}
return components
def get_dummy_inputs(self, device, seed=0):
image = floats_tensor((1, 3, 32, 32), rng=random.Random(seed)).to(device)
image = image / 2 + 0.5
if str(device).startswith("mps"):
generator = torch.manual_seed(seed)
else:
generator = torch.Generator(device=device).manual_seed(seed)
inputs = {
"prompt": "An astronaut riding an elephant",
"source_prompt": "An astronaut riding a horse",
"image": image,
"generator": generator,
"num_inference_steps": 2,
"eta": 0.1,
"strength": 0.8,
"guidance_scale": 3,
"source_guidance_scale": 1,
"output_type": "numpy",
}
return inputs
def test_stable_diffusion_cycle(self):
device = "cpu" # ensure determinism for the device-dependent torch.Generator
components = self.get_dummy_components()
pipe = CycleDiffusionPipeline(**components)
pipe = pipe.to(device)
pipe.set_progress_bar_config(disable=None)
inputs = self.get_dummy_inputs(device)
output = pipe(**inputs)
images = output.images
image_slice = images[0, -3:, -3:, -1]
assert images.shape == (1, 32, 32, 3)
expected_slice = np.array([0.4459, 0.4943, 0.4544, 0.6643, 0.5474, 0.4327, 0.5701, 0.5959, 0.5179])
assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
@unittest.skipIf(torch_device != "cuda", "This test requires a GPU")
def test_stable_diffusion_cycle_fp16(self):
components = self.get_dummy_components()
for name, module in components.items():
if hasattr(module, "half"):
components[name] = module.half()
pipe = CycleDiffusionPipeline(**components)
pipe = pipe.to(torch_device)
pipe.set_progress_bar_config(disable=None)
inputs = self.get_dummy_inputs(torch_device)
output = pipe(**inputs)
images = output.images
image_slice = images[0, -3:, -3:, -1]
assert images.shape == (1, 32, 32, 3)
expected_slice = np.array([0.3506, 0.4543, 0.446, 0.4575, 0.5195, 0.4155, 0.5273, 0.518, 0.4116])
assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
@skip_mps
def test_save_load_local(self):
return super().test_save_load_local()
@unittest.skip("non-deterministic pipeline")
def test_inference_batch_single_identical(self):
return super().test_inference_batch_single_identical()
@skip_mps
def test_dict_tuple_outputs_equivalent(self):
return super().test_dict_tuple_outputs_equivalent()
@skip_mps
def test_save_load_optional_components(self):
return super().test_save_load_optional_components()
@skip_mps
def test_attention_slicing_forward_pass(self):
return super().test_attention_slicing_forward_pass()
@nightly
@require_torch_gpu
class CycleDiffusionPipelineIntegrationTests(unittest.TestCase):
def tearDown(self):
# clean up the VRAM after each test
super().tearDown()
gc.collect()
torch.cuda.empty_cache()
def test_cycle_diffusion_pipeline_fp16(self):
init_image = load_image(
"https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"
"/cycle-diffusion/black_colored_car.png"
)
expected_image = load_numpy(
"https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/cycle-diffusion/blue_colored_car_fp16.npy"
)
init_image = init_image.resize((512, 512))
model_id = "CompVis/stable-diffusion-v1-4"
scheduler = DDIMScheduler.from_pretrained(model_id, subfolder="scheduler")
pipe = CycleDiffusionPipeline.from_pretrained(
model_id, scheduler=scheduler, safety_checker=None, torch_dtype=torch.float16, revision="fp16"
)
pipe.to(torch_device)
pipe.set_progress_bar_config(disable=None)
pipe.enable_attention_slicing()
source_prompt = "A black colored car"
prompt = "A blue colored car"
generator = torch.manual_seed(0)
output = pipe(
prompt=prompt,
source_prompt=source_prompt,
image=init_image,
num_inference_steps=100,
eta=0.1,
strength=0.85,
guidance_scale=3,
source_guidance_scale=1,
generator=generator,
output_type="np",
)
image = output.images
# the values aren't exactly equal, but the images look the same visually
assert np.abs(image - expected_image).max() < 5e-1
def test_cycle_diffusion_pipeline(self):
init_image = load_image(
"https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"
"/cycle-diffusion/black_colored_car.png"
)
expected_image = load_numpy(
"https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/cycle-diffusion/blue_colored_car.npy"
)
init_image = init_image.resize((512, 512))
model_id = "CompVis/stable-diffusion-v1-4"
scheduler = DDIMScheduler.from_pretrained(model_id, subfolder="scheduler")
pipe = CycleDiffusionPipeline.from_pretrained(model_id, scheduler=scheduler, safety_checker=None)
pipe.to(torch_device)
pipe.set_progress_bar_config(disable=None)
pipe.enable_attention_slicing()
source_prompt = "A black colored car"
prompt = "A blue colored car"
generator = torch.manual_seed(0)
output = pipe(
prompt=prompt,
source_prompt=source_prompt,
image=init_image,
num_inference_steps=100,
eta=0.1,
strength=0.85,
guidance_scale=3,
source_guidance_scale=1,
generator=generator,
output_type="np",
)
image = output.images
assert np.abs(image - expected_image).max() < 2e-2
tests/pipelines/stable_diffusion/test_stable_diffusion_model_editing.py deleted 100644 → 0
View file @ 0d927c75
# coding=utf-8
# Copyright 2023 HuggingFace Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import gc
import unittest
import numpy as np
import torch
from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer
from diffusers import (
AutoencoderKL,
DDIMScheduler,
EulerAncestralDiscreteScheduler,
PNDMScheduler,
StableDiffusionModelEditingPipeline,
UNet2DConditionModel,
)
from diffusers.utils.testing_utils import enable_full_determinism, nightly, require_torch_gpu, skip_mps, torch_device
from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_IMAGE_PARAMS, TEXT_TO_IMAGE_PARAMS
from ..test_pipelines_common import PipelineKarrasSchedulerTesterMixin, PipelineLatentTesterMixin, PipelineTesterMixin
enable_full_determinism()
@skip_mps
class StableDiffusionModelEditingPipelineFastTests(
PipelineLatentTesterMixin, PipelineKarrasSchedulerTesterMixin, PipelineTesterMixin, unittest.TestCase
):
pipeline_class = StableDiffusionModelEditingPipeline
params = TEXT_TO_IMAGE_PARAMS
batch_params = TEXT_TO_IMAGE_BATCH_PARAMS
image_params = TEXT_TO_IMAGE_IMAGE_PARAMS
image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS
def get_dummy_components(self):
torch.manual_seed(0)
unet = UNet2DConditionModel(
block_out_channels=(32, 64),
layers_per_block=2,
sample_size=32,
in_channels=4,
out_channels=4,
down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
cross_attention_dim=32,
)
scheduler = DDIMScheduler()
torch.manual_seed(0)
vae = AutoencoderKL(
block_out_channels=[32, 64],
in_channels=3,
out_channels=3,
down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"],
up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"],
latent_channels=4,
)
torch.manual_seed(0)
text_encoder_config = CLIPTextConfig(
bos_token_id=0,
eos_token_id=2,
hidden_size=32,
intermediate_size=37,
layer_norm_eps=1e-05,
num_attention_heads=4,
num_hidden_layers=5,
pad_token_id=1,
vocab_size=1000,
)
text_encoder = CLIPTextModel(text_encoder_config)
tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")
components = {
"unet": unet,
"scheduler": scheduler,
"vae": vae,
"text_encoder": text_encoder,
"tokenizer": tokenizer,
"safety_checker": None,
"feature_extractor": None,
}
return components
def get_dummy_inputs(self, device, seed=0):
generator = torch.manual_seed(seed)
inputs = {
"prompt": "A field of roses",
"generator": generator,
# Setting height and width to None to prevent OOMs on CPU.
"height": None,
"width": None,
"num_inference_steps": 2,
"guidance_scale": 6.0,
"output_type": "numpy",
}
return inputs
def test_stable_diffusion_model_editing_default_case(self):
device = "cpu" # ensure determinism for the device-dependent torch.Generator
components = self.get_dummy_components()
sd_pipe = StableDiffusionModelEditingPipeline(**components)
sd_pipe = sd_pipe.to(device)
sd_pipe.set_progress_bar_config(disable=None)
inputs = self.get_dummy_inputs(device)
image = sd_pipe(**inputs).images
image_slice = image[0, -3:, -3:, -1]
assert image.shape == (1, 64, 64, 3)
expected_slice = np.array([0.4755, 0.5132, 0.4976, 0.3904, 0.3554, 0.4765, 0.5139, 0.5158, 0.4889])
assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
def test_stable_diffusion_model_editing_negative_prompt(self):
device = "cpu" # ensure determinism for the device-dependent torch.Generator
components = self.get_dummy_components()
sd_pipe = StableDiffusionModelEditingPipeline(**components)
sd_pipe = sd_pipe.to(device)
sd_pipe.set_progress_bar_config(disable=None)
inputs = self.get_dummy_inputs(device)
negative_prompt = "french fries"
output = sd_pipe(**inputs, negative_prompt=negative_prompt)
image = output.images
image_slice = image[0, -3:, -3:, -1]
assert image.shape == (1, 64, 64, 3)
expected_slice = np.array([0.4992, 0.5101, 0.5004, 0.3949, 0.3604, 0.4735, 0.5216, 0.5204, 0.4913])
assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
def test_stable_diffusion_model_editing_euler(self):
device = "cpu" # ensure determinism for the device-dependent torch.Generator
components = self.get_dummy_components()
components["scheduler"] = EulerAncestralDiscreteScheduler(
beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear"
)
sd_pipe = StableDiffusionModelEditingPipeline(**components)
sd_pipe = sd_pipe.to(device)
sd_pipe.set_progress_bar_config(disable=None)
inputs = self.get_dummy_inputs(device)
image = sd_pipe(**inputs).images
image_slice = image[0, -3:, -3:, -1]
assert image.shape == (1, 64, 64, 3)
expected_slice = np.array([0.4747, 0.5372, 0.4779, 0.4982, 0.5543, 0.4816, 0.5238, 0.4904, 0.5027])
assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
def test_stable_diffusion_model_editing_pndm(self):
device = "cpu" # ensure determinism for the device-dependent torch.Generator
components = self.get_dummy_components()
components["scheduler"] = PNDMScheduler()
sd_pipe = StableDiffusionModelEditingPipeline(**components)
sd_pipe = sd_pipe.to(device)
sd_pipe.set_progress_bar_config(disable=None)
inputs = self.get_dummy_inputs(device)
# the pipeline does not expect pndm so test if it raises error.
with self.assertRaises(ValueError):
_ = sd_pipe(**inputs).images
def test_inference_batch_single_identical(self):
super().test_inference_batch_single_identical(expected_max_diff=5e-3)
def test_attention_slicing_forward_pass(self):
super().test_attention_slicing_forward_pass(expected_max_diff=5e-3)
@nightly
@require_torch_gpu
class StableDiffusionModelEditingSlowTests(unittest.TestCase):
def tearDown(self):
super().tearDown()
gc.collect()
torch.cuda.empty_cache()
def get_inputs(self, seed=0):
generator = torch.manual_seed(seed)
inputs = {
"prompt": "A field of roses",
"generator": generator,
"num_inference_steps": 3,
"guidance_scale": 7.5,
"output_type": "numpy",
}
return inputs
def test_stable_diffusion_model_editing_default(self):
model_ckpt = "CompVis/stable-diffusion-v1-4"
pipe = StableDiffusionModelEditingPipeline.from_pretrained(model_ckpt, safety_checker=None)
pipe.to(torch_device)
pipe.set_progress_bar_config(disable=None)
pipe.enable_attention_slicing()
inputs = self.get_inputs()
image = pipe(**inputs).images
image_slice = image[0, -3:, -3:, -1].flatten()
assert image.shape == (1, 512, 512, 3)
expected_slice = np.array(
[0.6749496, 0.6386453, 0.51443267, 0.66094905, 0.61921215, 0.5491332, 0.5744417, 0.58075106, 0.5174658]
)
assert np.abs(expected_slice - image_slice).max() < 1e-2
# make sure image changes after editing
pipe.edit_model("A pack of roses", "A pack of blue roses")
image = pipe(**inputs).images
image_slice = image[0, -3:, -3:, -1].flatten()
assert image.shape == (1, 512, 512, 3)
assert np.abs(expected_slice - image_slice).max() > 1e-1
def test_stable_diffusion_model_editing_pipeline_with_sequential_cpu_offloading(self):
torch.cuda.empty_cache()
torch.cuda.reset_max_memory_allocated()
torch.cuda.reset_peak_memory_stats()
model_ckpt = "CompVis/stable-diffusion-v1-4"
scheduler = DDIMScheduler.from_pretrained(model_ckpt, subfolder="scheduler")
pipe = StableDiffusionModelEditingPipeline.from_pretrained(
model_ckpt, scheduler=scheduler, safety_checker=None
)
pipe = pipe.to(torch_device)
pipe.set_progress_bar_config(disable=None)
pipe.enable_attention_slicing(1)
pipe.enable_sequential_cpu_offload()
inputs = self.get_inputs()
_ = pipe(**inputs)
mem_bytes = torch.cuda.max_memory_allocated()
# make sure that less than 4.4 GB is allocated
assert mem_bytes < 4.4 * 10**9
tests/pipelines/stable_diffusion/test_stable_diffusion_paradigms.py deleted 100644 → 0
View file @ 0d927c75
# coding=utf-8
# Copyright 2023 HuggingFace Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import gc
import unittest
import numpy as np
import torch
from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer
from diffusers import (
AutoencoderKL,
DDIMParallelScheduler,
DDPMParallelScheduler,
StableDiffusionParadigmsPipeline,
UNet2DConditionModel,
)
from diffusers.utils.testing_utils import (
enable_full_determinism,
nightly,
require_torch_gpu,
torch_device,
)
from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_IMAGE_PARAMS, TEXT_TO_IMAGE_PARAMS
from ..test_pipelines_common import PipelineLatentTesterMixin, PipelineTesterMixin
enable_full_determinism()
class StableDiffusionParadigmsPipelineFastTests(PipelineLatentTesterMixin, PipelineTesterMixin, unittest.TestCase):
pipeline_class = StableDiffusionParadigmsPipeline
params = TEXT_TO_IMAGE_PARAMS
batch_params = TEXT_TO_IMAGE_BATCH_PARAMS
image_params = TEXT_TO_IMAGE_IMAGE_PARAMS
image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS
def get_dummy_components(self):
torch.manual_seed(0)
unet = UNet2DConditionModel(
block_out_channels=(32, 64),
layers_per_block=2,
sample_size=32,
in_channels=4,
out_channels=4,
down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
cross_attention_dim=32,
# SD2-specific config below
attention_head_dim=(2, 4),
use_linear_projection=True,
)
scheduler = DDIMParallelScheduler(
beta_start=0.00085,
beta_end=0.012,
beta_schedule="scaled_linear",
clip_sample=False,
set_alpha_to_one=False,
)
torch.manual_seed(0)
vae = AutoencoderKL(
block_out_channels=[32, 64],
in_channels=3,
out_channels=3,
down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"],
up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"],
latent_channels=4,
sample_size=128,
)
torch.manual_seed(0)
text_encoder_config = CLIPTextConfig(
bos_token_id=0,
eos_token_id=2,
hidden_size=32,
intermediate_size=37,
layer_norm_eps=1e-05,
num_attention_heads=4,
num_hidden_layers=5,
pad_token_id=1,
vocab_size=1000,
# SD2-specific config below
hidden_act="gelu",
projection_dim=512,
)
text_encoder = CLIPTextModel(text_encoder_config)
tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")
components = {
"unet": unet,
"scheduler": scheduler,
"vae": vae,
"text_encoder": text_encoder,
"tokenizer": tokenizer,
"safety_checker": None,
"feature_extractor": None,
}
return components
def get_dummy_inputs(self, device, seed=0):
if str(device).startswith("mps"):
generator = torch.manual_seed(seed)
else:
generator = torch.Generator(device=device).manual_seed(seed)
inputs = {
"prompt": "a photograph of an astronaut riding a horse",
"generator": generator,
"num_inference_steps": 10,
"guidance_scale": 6.0,
"output_type": "numpy",
"parallel": 3,
"debug": True,
}
return inputs
def test_stable_diffusion_paradigms_default_case(self):
device = "cpu" # ensure determinism for the device-dependent torch.Generator
components = self.get_dummy_components()
sd_pipe = StableDiffusionParadigmsPipeline(**components)
sd_pipe = sd_pipe.to(device)
sd_pipe.set_progress_bar_config(disable=None)
inputs = self.get_dummy_inputs(device)
image = sd_pipe(**inputs).images
image_slice = image[0, -3:, -3:, -1]
assert image.shape == (1, 64, 64, 3)
expected_slice = np.array([0.4773, 0.5417, 0.4723, 0.4925, 0.5631, 0.4752, 0.5240, 0.4935, 0.5023])
assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
def test_stable_diffusion_paradigms_default_case_ddpm(self):
device = "cpu" # ensure determinism for the device-dependent torch.Generator
components = self.get_dummy_components()
torch.manual_seed(0)
components["scheduler"] = DDPMParallelScheduler()
torch.manual_seed(0)
sd_pipe = StableDiffusionParadigmsPipeline(**components)
sd_pipe = sd_pipe.to(device)
sd_pipe.set_progress_bar_config(disable=None)
inputs = self.get_dummy_inputs(device)
image = sd_pipe(**inputs).images
image_slice = image[0, -3:, -3:, -1]
assert image.shape == (1, 64, 64, 3)
expected_slice = np.array([0.3573, 0.4420, 0.4960, 0.4799, 0.3796, 0.3879, 0.4819, 0.4365, 0.4468])
assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
# override to speed the overall test timing up.
def test_inference_batch_consistent(self):
super().test_inference_batch_consistent(batch_sizes=[1, 2])
# override to speed the overall test timing up.
def test_inference_batch_single_identical(self):
super().test_inference_batch_single_identical(batch_size=2, expected_max_diff=3e-3)
def test_stable_diffusion_paradigms_negative_prompt(self):
device = "cpu" # ensure determinism for the device-dependent torch.Generator
components = self.get_dummy_components()
sd_pipe = StableDiffusionParadigmsPipeline(**components)
sd_pipe = sd_pipe.to(device)
sd_pipe.set_progress_bar_config(disable=None)
inputs = self.get_dummy_inputs(device)
negative_prompt = "french fries"
output = sd_pipe(**inputs, negative_prompt=negative_prompt)
image = output.images
image_slice = image[0, -3:, -3:, -1]
assert image.shape == (1, 64, 64, 3)
expected_slice = np.array([0.4771, 0.5420, 0.4683, 0.4918, 0.5636, 0.4725, 0.5230, 0.4923, 0.5015])
assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
@nightly
@require_torch_gpu
class StableDiffusionParadigmsPipelineSlowTests(unittest.TestCase):
def tearDown(self):
super().tearDown()
gc.collect()
torch.cuda.empty_cache()
def get_inputs(self, seed=0):
generator = torch.Generator(device=torch_device).manual_seed(seed)
inputs = {
"prompt": "a photograph of an astronaut riding a horse",
"generator": generator,
"num_inference_steps": 10,
"guidance_scale": 7.5,
"output_type": "numpy",
"parallel": 3,
"debug": True,
}
return inputs
def test_stable_diffusion_paradigms_default(self):
model_ckpt = "stabilityai/stable-diffusion-2-base"
scheduler = DDIMParallelScheduler.from_pretrained(model_ckpt, subfolder="scheduler")
pipe = StableDiffusionParadigmsPipeline.from_pretrained(model_ckpt, scheduler=scheduler, safety_checker=None)
pipe.to(torch_device)
pipe.set_progress_bar_config(disable=None)
pipe.enable_attention_slicing()
inputs = self.get_inputs()
image = pipe(**inputs).images
image_slice = image[0, -3:, -3:, -1].flatten()
assert image.shape == (1, 512, 512, 3)
expected_slice = np.array([0.9622, 0.9602, 0.9748, 0.9591, 0.9630, 0.9691, 0.9661, 0.9631, 0.9741])
assert np.abs(expected_slice - image_slice).max() < 1e-2
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment