AI/diffusers
1
0
Fork 0
mirror of https://github.com/huggingface/diffusers.git synced 2026-01-27 17:22:53 +03:00
Code Activity

org test case

Browse Source
This commit is contained in:
Chong
2023-08-23 20:25:06 +08:00
parent 1ad175c064
commit 616c3ff2d7
2 changed files with 16 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/diffusers/models/unet_2d_condition.py
View File

@@ -966,7 +966,7 @@ class UNet2DConditionModel(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin)
encoder_attention_mask=encoder_attention_mask,
)
# To support T2I-Adapter-XL
if is_adapter and len(down_block_additional_residuals) > 0:
if is_adapter and len(down_block_additional_residuals) > 0 and sample.shape == down_block_additional_residuals[0].shape:
sample += down_block_additional_residuals.pop(0)
if is_controlnet:

34
tests/pipelines/stable_diffusion/test_stable_diffusion_adapter.py
View File

@@ -46,34 +46,24 @@ class AdapterTests:
def get_dummy_components(self, adapter_type):
torch.manual_seed(0)
if adapter_type == 'light_adapter':
channels = [32, 32, 32]
else:
channels = [32, 32, 32, 32]
torch.manual_seed(0)
unet = UNet2DConditionModel(
block_out_channels=[32, 32, 32, 32],
block_out_channels=(32, 64),
layers_per_block=2,
sample_size=32,
in_channels=4,
out_channels=4,
down_block_types=(
"CrossAttnDownBlock2D",
"CrossAttnDownBlock2D",
"CrossAttnDownBlock2D",
"DownBlock2D",
),
up_block_types= ("UpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D"),
down_block_types=("CrossAttnDownBlock2D", "DownBlock2D"),
up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
cross_attention_dim=32,
)
scheduler = PNDMScheduler(skip_prk_steps=True)
torch.manual_seed(0)
vae = AutoencoderKL(
block_out_channels=[32, 32, 32, 32],
block_out_channels=[32, 64],
in_channels=3,
out_channels=3,
down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D", "DownEncoderBlock2D", "DownEncoderBlock2D"],
up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D", "UpDecoderBlock2D", "UpDecoderBlock2D"],
down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"],
up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"],
latent_channels=4,
)
torch.manual_seed(0)
@@ -94,9 +84,9 @@ class AdapterTests:
torch.manual_seed(0)
adapter = T2IAdapter(
in_channels=3,
channels=channels,
channels=[32, 64],
num_res_blocks=2,
downscale_factor=8,
downscale_factor=2,
adapter_type=adapter_type,
)
@@ -155,8 +145,11 @@ class StableDiffusionFullAdapterPipelineFastTests(AdapterTests, PipelineTesterMi
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([0.4858, 0.5500, 0.4278, 0.4669, 0.6184, 0.4322, 0.5010, 0.5033, 0.4746])
assert np.abs(image_slice.flatten() - expected_slice).max() < 5e-3
class StableDiffusionLightAdapterPipelineFastTests(AdapterTests, PipelineTesterMixin, unittest.TestCase):
@@ -172,8 +165,11 @@ class StableDiffusionLightAdapterPipelineFastTests(AdapterTests, PipelineTesterM
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([0.4965, 0.5548, 0.4330, 0.4771, 0.6226, 0.4382, 0.5037, 0.5071, 0.4782])
assert np.abs(image_slice.flatten() - expected_slice).max() < 5e-3
@slow
@@ -317,4 +313,4 @@ class StableDiffusionAdapterPipelineSlowTests(unittest.TestCase):
pipe(prompt="foo", image=image, num_inference_steps=2)
mem_bytes = torch.cuda.max_memory_allocated()
assert mem_bytes < 5 * 10**9
assert mem_bytes < 5 * 10**9