import functools
import importlib
import inspect
import io
import logging
import multiprocessing
import os
import random
import re
import struct
import sys
import tempfile
import time
import unittest
import urllib.parse
from contextlib import contextmanager
from distutils.util import strtobool
from io import BytesIO, StringIO
from pathlib import Path
from typing import List, Optional, Union
import numpy as np
import PIL.Image
import PIL.ImageOps
import requests
from numpy.linalg import norm
from packaging import version
from .import_utils import (
BACKENDS_MAPPING,
is_compel_available,
is_flax_available,
is_note_seq_available,
is_onnx_available,
is_opencv_available,
is_peft_available,
is_torch_available,
is_torch_version,
is_torchsde_available,
is_transformers_available,
)
from .logging import get_logger
global_rng = random.Random()
logger = get_logger(__name__)
_required_peft_version = is_peft_available() and version.parse(
version.parse(importlib.metadata.version("peft")).base_version
) > version.parse("0.5")
_required_transformers_version = is_transformers_available() and version.parse(
version.parse(importlib.metadata.version("transformers")).base_version
) > version.parse("4.33")
USE_PEFT_BACKEND = _required_peft_version and _required_transformers_version
if is_torch_available():
import torch
if "DIFFUSERS_TEST_DEVICE" in os.environ:
torch_device = os.environ["DIFFUSERS_TEST_DEVICE"]
try:
# try creating device to see if provided device is valid
_ = torch.device(torch_device)
except RuntimeError as e:
raise RuntimeError(
f"Unknown testing device specified by environment variable `DIFFUSERS_TEST_DEVICE`: {torch_device}"
) from e
logger.info(f"torch_device overrode to {torch_device}")
else:
torch_device = "cuda" if torch.cuda.is_available() else "cpu"
is_torch_higher_equal_than_1_12 = version.parse(
version.parse(torch.__version__).base_version
) >= version.parse("1.12")
if is_torch_higher_equal_than_1_12:
# Some builds of torch 1.12 don't have the mps backend registered. See #892 for more details
mps_backend_registered = hasattr(torch.backends, "mps")
torch_device = "mps" if (mps_backend_registered and torch.backends.mps.is_available()) else torch_device
def torch_all_close(a, b, *args, **kwargs):
if not is_torch_available():
raise ValueError("PyTorch needs to be installed to use this function.")
if not torch.allclose(a, b, *args, **kwargs):
assert False, f"Max diff is absolute {(a - b).abs().max()}. Diff tensor is {(a - b).abs()}."
return True
def numpy_cosine_similarity_distance(a, b):
similarity = np.dot(a, b) / (norm(a) * norm(b))
distance = 1.0 - similarity.mean()
return distance
def print_tensor_test(tensor, filename="test_corrections.txt", expected_tensor_name="expected_slice"):
test_name = os.environ.get("PYTEST_CURRENT_TEST")
if not torch.is_tensor(tensor):
tensor = torch.from_numpy(tensor)
tensor_str = str(tensor.detach().cpu().flatten().to(torch.float32)).replace("\n", "")
# format is usually:
# expected_slice = np.array([-0.5713, -0.3018, -0.9814, 0.04663, -0.879, 0.76, -1.734, 0.1044, 1.161])
output_str = tensor_str.replace("tensor", f"{expected_tensor_name} = np.array")
test_file, test_class, test_fn = test_name.split("::")
test_fn = test_fn.split()[0]
with open(filename, "a") as f:
print(";".join([test_file, test_class, test_fn, output_str]), file=f)
def get_tests_dir(append_path=None):
"""
Args:
append_path: optional path to append to the tests dir path
Return:
The full path to the `tests` dir, so that the tests can be invoked from anywhere. Optionally `append_path` is
joined after the `tests` dir the former is provided.
"""
# this function caller's __file__
caller__file__ = inspect.stack()[1][1]
tests_dir = os.path.abspath(os.path.dirname(caller__file__))
while not tests_dir.endswith("tests"):
tests_dir = os.path.dirname(tests_dir)
if append_path:
return os.path.join(tests_dir, append_path)
else:
return tests_dir
def parse_flag_from_env(key, default=False):
try:
value = os.environ[key]
except KeyError:
# KEY isn't set, default to `default`.
_value = default
else:
# KEY is set, convert it to True or False.
try:
_value = strtobool(value)
except ValueError:
# More values are supported, but let's keep the message simple.
raise ValueError(f"If set, {key} must be yes or no.")
return _value
_run_slow_tests = parse_flag_from_env("RUN_SLOW", default=False)
_run_nightly_tests = parse_flag_from_env("RUN_NIGHTLY", default=False)
def floats_tensor(shape, scale=1.0, rng=None, name=None):
"""Creates a random float32 tensor"""
if rng is None:
rng = global_rng
total_dims = 1
for dim in shape:
total_dims *= dim
values = []
for _ in range(total_dims):
values.append(rng.random() * scale)
return torch.tensor(data=values, dtype=torch.float).view(shape).contiguous()
def slow(test_case):
"""
Decorator marking a test as slow.
Slow tests are skipped by default. Set the RUN_SLOW environment variable to a truthy value to run them.
"""
return unittest.skipUnless(_run_slow_tests, "test is slow")(test_case)
def nightly(test_case):
"""
Decorator marking a test that runs nightly in the diffusers CI.
Slow tests are skipped by default. Set the RUN_NIGHTLY environment variable to a truthy value to run them.
"""
return unittest.skipUnless(_run_nightly_tests, "test is nightly")(test_case)
def require_torch(test_case):
"""
Decorator marking a test that requires PyTorch. These tests are skipped when PyTorch isn't installed.
"""
return unittest.skipUnless(is_torch_available(), "test requires PyTorch")(test_case)
def require_torch_2(test_case):
"""
Decorator marking a test that requires PyTorch 2. These tests are skipped when it isn't installed.
"""
return unittest.skipUnless(is_torch_available() and is_torch_version(">=", "2.0.0"), "test requires PyTorch 2")(
test_case
)
def require_torch_gpu(test_case):
"""Decorator marking a test that requires CUDA and PyTorch."""
return unittest.skipUnless(is_torch_available() and torch_device == "cuda", "test requires PyTorch+CUDA")(
test_case
)
def skip_mps(test_case):
"""Decorator marking a test to skip if torch_device is 'mps'"""
return unittest.skipUnless(torch_device != "mps", "test requires non 'mps' device")(test_case)
def require_flax(test_case):
"""
Decorator marking a test that requires JAX & Flax. These tests are skipped when one / both are not installed
"""
return unittest.skipUnless(is_flax_available(), "test requires JAX & Flax")(test_case)
def require_compel(test_case):
"""
Decorator marking a test that requires compel: https://github.com/damian0815/compel. These tests are skipped when
the library is not installed.
"""
return unittest.skipUnless(is_compel_available(), "test requires compel")(test_case)
def require_onnxruntime(test_case):
"""
Decorator marking a test that requires onnxruntime. These tests are skipped when onnxruntime isn't installed.
"""
return unittest.skipUnless(is_onnx_available(), "test requires onnxruntime")(test_case)
def require_note_seq(test_case):
"""
Decorator marking a test that requires note_seq. These tests are skipped when note_seq isn't installed.
"""
return unittest.skipUnless(is_note_seq_available(), "test requires note_seq")(test_case)
def require_torchsde(test_case):
"""
Decorator marking a test that requires torchsde. These tests are skipped when torchsde isn't installed.
"""
return unittest.skipUnless(is_torchsde_available(), "test requires torchsde")(test_case)
def require_peft_backend(test_case):
"""
Decorator marking a test that requires PEFT backend, this would require some specific versions of PEFT and
transformers.
"""
return unittest.skipUnless(USE_PEFT_BACKEND, "test requires PEFT backend")(test_case)
def deprecate_after_peft_backend(test_case):
"""
Decorator marking a test that will be skipped after PEFT backend
"""
return unittest.skipUnless(not USE_PEFT_BACKEND, "test skipped in favor of PEFT backend")(test_case)
def require_python39_or_higher(test_case):
def python39_available():
sys_info = sys.version_info
major, minor = sys_info.major, sys_info.minor
return major == 3 and minor >= 9
return unittest.skipUnless(python39_available(), "test requires Python 3.9 or higher")(test_case)
def load_numpy(arry: Union[str, np.ndarray], local_path: Optional[str] = None) -> np.ndarray:
if isinstance(arry, str):
# local_path = "/home/patrick_huggingface_co/"
if local_path is not None:
# local_path can be passed to correct images of tests
return os.path.join(local_path, "/".join([arry.split("/")[-5], arry.split("/")[-2], arry.split("/")[-1]]))
elif arry.startswith("http://") or arry.startswith("https://"):
response = requests.get(arry)
response.raise_for_status()
arry = np.load(BytesIO(response.content))
elif os.path.isfile(arry):
arry = np.load(arry)
else:
raise ValueError(
f"Incorrect path or url, URLs must start with `http://` or `https://`, and {arry} is not a valid path"
)
elif isinstance(arry, np.ndarray):
pass
else:
raise ValueError(
"Incorrect format used for numpy ndarray. Should be an url linking to an image, a local path, or a"
" ndarray."
)
return arry
def load_pt(url: str):
response = requests.get(url)
response.raise_for_status()
arry = torch.load(BytesIO(response.content))
return arry
def load_image(image: Union[str, PIL.Image.Image]) -> PIL.Image.Image:
"""
Loads `image` to a PIL Image.
Args:
image (`str` or `PIL.Image.Image`):
The image to convert to the PIL Image format.
Returns:
`PIL.Image.Image`:
A PIL Image.
"""
if isinstance(image, str):
if image.startswith("http://") or image.startswith("https://"):
image = PIL.Image.open(requests.get(image, stream=True).raw)
elif os.path.isfile(image):
image = PIL.Image.open(image)
else:
raise ValueError(
f"Incorrect path or url, URLs must start with `http://` or `https://`, and {image} is not a valid path"
)
elif isinstance(image, PIL.Image.Image):
image = image
else:
raise ValueError(
"Incorrect format used for image. Should be an url linking to an image, a local path, or a PIL image."
)
image = PIL.ImageOps.exif_transpose(image)
image = image.convert("RGB")
return image
def preprocess_image(image: PIL.Image, batch_size: int):
w, h = image.size
w, h = (x - x % 8 for x in (w, h)) # resize to integer multiple of 8
image = image.resize((w, h), resample=PIL.Image.LANCZOS)
image = np.array(image).astype(np.float32) / 255.0
image = np.vstack([image[None].transpose(0, 3, 1, 2)] * batch_size)
image = torch.from_numpy(image)
return 2.0 * image - 1.0
def export_to_gif(image: List[PIL.Image.Image], output_gif_path: str = None) -> str:
if output_gif_path is None:
output_gif_path = tempfile.NamedTemporaryFile(suffix=".gif").name
image[0].save(
output_gif_path,
save_all=True,
append_images=image[1:],
optimize=False,
duration=100,
loop=0,
)
return output_gif_path
@contextmanager
def buffered_writer(raw_f):
f = io.BufferedWriter(raw_f)
yield f
f.flush()
def export_to_ply(mesh, output_ply_path: str = None):
"""
Write a PLY file for a mesh.
"""
if output_ply_path is None:
output_ply_path = tempfile.NamedTemporaryFile(suffix=".ply").name
coords = mesh.verts.detach().cpu().numpy()
faces = mesh.faces.cpu().numpy()
rgb = np.stack([mesh.vertex_channels[x].detach().cpu().numpy() for x in "RGB"], axis=1)
with buffered_writer(open(output_ply_path, "wb")) as f:
f.write(b"ply\n")
f.write(b"format binary_little_endian 1.0\n")
f.write(bytes(f"element vertex {len(coords)}\n", "ascii"))
f.write(b"property float x\n")
f.write(b"property float y\n")
f.write(b"property float z\n")
if rgb is not None:
f.write(b"property uchar red\n")
f.write(b"property uchar green\n")
f.write(b"property uchar blue\n")
if faces is not None:
f.write(bytes(f"element face {len(faces)}\n", "ascii"))
f.write(b"property list uchar int vertex_index\n")
f.write(b"end_header\n")
if rgb is not None:
rgb = (rgb * 255.499).round().astype(int)
vertices = [
(*coord, *rgb)
for coord, rgb in zip(
coords.tolist(),
rgb.tolist(),
)
]
format = struct.Struct(">> from diffusers import logging
>>> from diffusers.testing_utils import CaptureLogger
>>> msg = "Testing 1, 2, 3"
>>> logging.set_verbosity_info()
>>> logger = logging.get_logger("diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.py")
>>> with CaptureLogger(logger) as cl:
... logger.info(msg)
>>> assert cl.out, msg + "\n"
```
"""
def __init__(self, logger):
self.logger = logger
self.io = StringIO()
self.sh = logging.StreamHandler(self.io)
self.out = ""
def __enter__(self):
self.logger.addHandler(self.sh)
return self
def __exit__(self, *exc):
self.logger.removeHandler(self.sh)
self.out = self.io.getvalue()
def __repr__(self):
return f"captured: {self.out}\n"
def enable_full_determinism():
"""
Helper function for reproducible behavior during distributed training. See
- https://pytorch.org/docs/stable/notes/randomness.html for pytorch
"""
# Enable PyTorch deterministic mode. This potentially requires either the environment
# variable 'CUDA_LAUNCH_BLOCKING' or 'CUBLAS_WORKSPACE_CONFIG' to be set,
# depending on the CUDA version, so we set them both here
os.environ["CUDA_LAUNCH_BLOCKING"] = "1"
os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":16:8"
torch.use_deterministic_algorithms(True)
# Enable CUDNN deterministic mode
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
torch.backends.cuda.matmul.allow_tf32 = False
def disable_full_determinism():
os.environ["CUDA_LAUNCH_BLOCKING"] = "0"
os.environ["CUBLAS_WORKSPACE_CONFIG"] = ""
torch.use_deterministic_algorithms(False)