Revive make quality (open-mmlab#203)

* Revive Make utils

* Add datasets for training too

Author: anton-l

Makefile

@@ -79,11 +79,6 @@ test:
 test-examples:
 	python -m pytest -n auto --dist=loadfile -s -v ./examples/pytorch/
 
-# Run tests for SageMaker DLC release
-
-test-sagemaker: # install sagemaker dependencies in advance with pip install .[sagemaker]
-	TEST_SAGEMAKER=True python -m pytest -n auto  -s -v ./tests/sagemaker
-
 
 # Release stuff
 

setup.py

@@ -77,19 +77,22 @@
 # 2. once modified, run: `make deps_table_update` to update src/diffusers/dependency_versions_table.py
 _deps = [
     "Pillow",
+    "accelerate>=0.11.0",
     "black~=22.0,>=22.3",
+    "datasets",
     "filelock",
     "flake8>=3.8.3",
+    "hf-doc-builder>=0.3.0",
     "huggingface-hub",
     "importlib_metadata",
     "isort>=5.5.4",
+    "modelcards==0.1.4",
     "numpy",
     "pytest",
     "regex!=2019.12.17",
     "requests",
-    "torch>=1.4",
     "tensorboard",
-    "modelcards==0.1.4"
+    "torch>=1.4",
 ]
 
 # this is a lookup table with items like:
@@ -161,12 +164,10 @@ def run(self):
 
 extras = {}
 extras["quality"] = ["black ~= 22.0", "isort >= 5.5.4", "flake8 >= 3.8.3"]
-extras["docs"] = []
-extras["training"] = ["tensorboard", "modelcards"]
-extras["test"] = [
-    "pytest",
-]
-extras["dev"] = extras["quality"] + extras["test"] + extras["training"]
+extras["docs"] = ["hf-doc-builder"]
+extras["training"] = ["accelerate", "datasets", "tensorboard", "modelcards"]
+extras["test"] = ["pytest"]
+extras["dev"] = extras["quality"] + extras["test"] + extras["training"] + extras["docs"]
 
 install_requires = [
     deps["importlib_metadata"],

src/diffusers/dependency_versions_table.py

@@ -3,17 +3,19 @@
 # 2. run `make deps_table_update``
 deps = {
     "Pillow": "Pillow",
+    "accelerate": "accelerate>=0.11.0",
     "black": "black~=22.0,>=22.3",
     "filelock": "filelock",
     "flake8": "flake8>=3.8.3",
+    "hf-doc-builder": "hf-doc-builder>=0.3.0",
     "huggingface-hub": "huggingface-hub",
     "importlib_metadata": "importlib_metadata",
     "isort": "isort>=5.5.4",
+    "modelcards": "modelcards==0.1.4",
     "numpy": "numpy",
     "pytest": "pytest",
     "regex": "regex!=2019.12.17",
     "requests": "requests",
     "torch": "torch>=1.4",
     "tensorboard": "tensorboard",
-    "modelcards": "modelcards==0.1.4",
 }

src/diffusers/optimization.py

@@ -18,7 +18,6 @@
 from enum import Enum
 from typing import Optional, Union
 
-import torch
 from torch.optim import Optimizer
 from torch.optim.lr_scheduler import LambdaLR
 

src/diffusers/pipelines/__init__.py

@@ -1,4 +1,8 @@
-from ..utils import is_inflect_available, is_transformers_available, is_unidecode_available
+# flake8: noqa
+# There's no way to ignore "F401 '...' imported but unused" warnings in this
+# module, but to preserve other warnings. So, don't check this module at all.
+
+from ..utils import is_transformers_available
 from .ddim import DDIMPipeline
 from .ddpm import DDPMPipeline
 from .latent_diffusion_uncond import LDMPipeline

src/diffusers/pipelines/ddim/__init__.py

@@ -1 +1,2 @@
+# flake8: noqa
 from .pipeline_ddim import DDIMPipeline

src/diffusers/pipelines/ddpm/__init__.py

@@ -1 +1,2 @@
+# flake8: noqa
 from .pipeline_ddpm import DDPMPipeline

src/diffusers/pipelines/latent_diffusion/__init__.py

@@ -1,3 +1,4 @@
+# flake8: noqa
 from ...utils import is_transformers_available
 
 

src/diffusers/pipelines/latent_diffusion_uncond/__init__.py

@@ -1 +1,2 @@
+# flake8: noqa
 from .pipeline_latent_diffusion_uncond import LDMPipeline

src/diffusers/pipelines/pndm/__init__.py

@@ -1 +1,2 @@
+# flake8: noqa
 from .pipeline_pndm import PNDMPipeline

src/diffusers/pipelines/score_sde_ve/__init__.py

@@ -1 +1,2 @@
+# flake8: noqa
 from .pipeline_score_sde_ve import ScoreSdeVePipeline

src/diffusers/pipelines/stable_diffusion/__init__.py

@@ -1,3 +1,4 @@
+# flake8: noqa
 from ...utils import is_transformers_available
 
 

src/diffusers/pipelines/stochatic_karras_ve/__init__.py

@@ -1 +1,2 @@
+# flake8: noqa
 from .pipeline_stochastic_karras_ve import KarrasVePipeline

src/diffusers/schedulers/scheduling_lms_discrete.py

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import List, Union
+from typing import Union
 
 import numpy as np
 import torch

src/diffusers/utils/dummy_scipy_objects.py

@@ -8,17 +8,3 @@ class LMSDiscreteScheduler(metaclass=DummyObject):
 
     def __init__(self, *args, **kwargs):
         requires_backends(self, ["scipy"])
-
-
-class LDMTextToImagePipeline(metaclass=DummyObject):
-    _backends = ["scipy"]
-
-    def __init__(self, *args, **kwargs):
-        requires_backends(self, ["scipy"])
-
-
-class StableDiffusionPipeline(metaclass=DummyObject):
-    _backends = ["scipy"]
-
-    def __init__(self, *args, **kwargs):
-        requires_backends(self, ["scipy"])

tests/test_layers_utils.py

@@ -14,16 +14,13 @@
 # limitations under the License.
 
 
-import inspect
-import tempfile
 import unittest
 
 import numpy as np
 import torch
 
 from diffusers.models.embeddings import get_timestep_embedding
-from diffusers.models.resnet import Downsample1D, Downsample2D, Upsample1D, Upsample2D
-from diffusers.testing_utils import floats_tensor, slow, torch_device
+from diffusers.models.resnet import Downsample2D, Upsample2D
 
 
 torch.backends.cuda.matmul.allow_tf32 = False
@@ -219,108 +216,3 @@ def test_downsample_with_conv_out_dim(self):
         output_slice = downsampled[0, -1, -3:, -3:]
         expected_slice = torch.tensor([-0.6586, 0.5985, 0.0721, 0.1256, -0.1492, 0.4436, -0.2544, 0.5021, 1.1522])
         assert torch.allclose(output_slice.flatten(), expected_slice, atol=1e-3)
-
-
-class Upsample1DBlockTests(unittest.TestCase):
-    def test_upsample_default(self):
-        torch.manual_seed(0)
-        sample = torch.randn(1, 32, 32)
-        upsample = Upsample1D(channels=32, use_conv=False)
-        with torch.no_grad():
-            upsampled = upsample(sample)
-
-        assert upsampled.shape == (1, 32, 64)
-        output_slice = upsampled[0, -1, -8:]
-        expected_slice = torch.tensor([-1.6340, -1.6340, 0.5374, 0.5374, 1.0826, 1.0826, -1.7105, -1.7105])
-        assert torch.allclose(output_slice.flatten(), expected_slice, atol=1e-3)
-
-    def test_upsample_with_conv(self):
-        torch.manual_seed(0)
-        sample = torch.randn(1, 32, 32)
-        upsample = Upsample1D(channels=32, use_conv=True)
-        with torch.no_grad():
-            upsampled = upsample(sample)
-
-        assert upsampled.shape == (1, 32, 64)
-        output_slice = upsampled[0, -1, -8:]
-        expected_slice = torch.tensor([-0.4546, -0.5010, -0.2996, 0.2844, 0.4040, -0.7772, -0.6862, 0.3612])
-        assert torch.allclose(output_slice.flatten(), expected_slice, atol=1e-3)
-
-    def test_upsample_with_conv_out_dim(self):
-        torch.manual_seed(0)
-        sample = torch.randn(1, 32, 32)
-        upsample = Upsample1D(channels=32, use_conv=True, out_channels=64)
-        with torch.no_grad():
-            upsampled = upsample(sample)
-
-        assert upsampled.shape == (1, 64, 64)
-        output_slice = upsampled[0, -1, -8:]
-        expected_slice = torch.tensor([-0.0516, -0.0972, 0.9740, 1.1883, 0.4539, -0.5285, -0.5851, 0.1152])
-        assert torch.allclose(output_slice.flatten(), expected_slice, atol=1e-3)
-
-    def test_upsample_with_transpose(self):
-        torch.manual_seed(0)
-        sample = torch.randn(1, 32, 32)
-        upsample = Upsample1D(channels=32, use_conv=False, use_conv_transpose=True)
-        with torch.no_grad():
-            upsampled = upsample(sample)
-
-        assert upsampled.shape == (1, 32, 64)
-        output_slice = upsampled[0, -1, -8:]
-        expected_slice = torch.tensor([-0.2238, -0.5842, -0.7165, 0.6699, 0.1033, -0.4269, -0.8974, -0.3716])
-        assert torch.allclose(output_slice.flatten(), expected_slice, atol=1e-3)
-
-
-class Downsample1DBlockTests(unittest.TestCase):
-    def test_downsample_default(self):
-        torch.manual_seed(0)
-        sample = torch.randn(1, 32, 64)
-        downsample = Downsample1D(channels=32, use_conv=False)
-        with torch.no_grad():
-            downsampled = downsample(sample)
-
-        assert downsampled.shape == (1, 32, 32)
-        output_slice = downsampled[0, -1, -8:]
-        expected_slice = torch.tensor([-0.8796, 1.0945, -0.3434, 0.2910, 0.3391, -0.4488, -0.9568, -0.2909])
-        max_diff = (output_slice.flatten() - expected_slice).abs().sum().item()
-        assert max_diff <= 1e-3
-        # assert torch.allclose(output_slice.flatten(), expected_slice, atol=1e-1)
-
-    def test_downsample_with_conv(self):
-        torch.manual_seed(0)
-        sample = torch.randn(1, 32, 64)
-        downsample = Downsample1D(channels=32, use_conv=True)
-        with torch.no_grad():
-            downsampled = downsample(sample)
-
-        assert downsampled.shape == (1, 32, 32)
-        output_slice = downsampled[0, -1, -8:]
-
-        expected_slice = torch.tensor(
-            [0.1723, 0.0811, -0.6205, -0.3045, 0.0666, -0.2381, -0.0238, 0.2834],
-        )
-        assert torch.allclose(output_slice.flatten(), expected_slice, atol=1e-3)
-
-    def test_downsample_with_conv_pad1(self):
-        torch.manual_seed(0)
-        sample = torch.randn(1, 32, 64)
-        downsample = Downsample1D(channels=32, use_conv=True, padding=1)
-        with torch.no_grad():
-            downsampled = downsample(sample)
-
-        assert downsampled.shape == (1, 32, 32)
-        output_slice = downsampled[0, -1, -8:]
-        expected_slice = torch.tensor([0.1723, 0.0811, -0.6205, -0.3045, 0.0666, -0.2381, -0.0238, 0.2834])
-        assert torch.allclose(output_slice.flatten(), expected_slice, atol=1e-3)
-
-    def test_downsample_with_conv_out_dim(self):
-        torch.manual_seed(0)
-        sample = torch.randn(1, 32, 64)
-        downsample = Downsample1D(channels=32, use_conv=True, out_channels=16)
-        with torch.no_grad():
-            downsampled = downsample(sample)
-
-        assert downsampled.shape == (1, 16, 32)
-        output_slice = downsampled[0, -1, -8:]
-        expected_slice = torch.tensor([1.1067, -0.5255, -0.4451, 0.0487, -0.3664, -0.7945, -0.4495, -0.3129])
-        assert torch.allclose(output_slice.flatten(), expected_slice, atol=1e-3)