# Copyright (c) 2023, Teriks
#
# dgenerate is distributed under the following BSD 3-Clause License
#
# Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
#
# 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
#
# 2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in
# the documentation and/or other materials provided with the distribution.
#
# 3. Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
# ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
import transformers
import dgenerate.hfhub as _hfhub
import dgenerate.memoize as _d_memoize
import dgenerate.memory as _memory
import dgenerate.messages as _messages
import dgenerate.pipelinewrapper.enums as _enums
import dgenerate.pipelinewrapper.util as _pipelinewrapper_util
import dgenerate.textprocessing as _textprocessing
import dgenerate.types as _types
from dgenerate.memoize import memoize as _memoize
from dgenerate.pipelinewrapper import constants as _constants
from dgenerate.pipelinewrapper.uris import exceptions as _exceptions
from dgenerate.pipelinewrapper.uris import util as _util
from dgenerate.pipelinewrapper import models as _models
_image_encoder_uri_parser = _textprocessing.ConceptUriParser(
'ImageEncoder', ['revision', 'variant', 'subfolder', 'dtype'])
_image_encoder_cache = _d_memoize.create_object_cache(
'image_encoder', cache_type=_memory.SizedConstrainedObjectCache
)
[docs]
class ImageEncoderUri:
"""
Representation of ``--image-encoder`` URI.
"""
# pipelinewrapper.uris.util.get_uri_accepted_args_schema metadata
NAMES = ['Image Encoder']
[docs]
@staticmethod
def help():
import dgenerate.arguments as _a
return _a.get_raw_help_text('--image-encoder')
OPTION_ARGS = {
'dtype': ['float16', 'bfloat16', 'float32']
}
FILE_ARGS = {
'model': {'mode': 'dir'}
}
# ===
@property
def model(self) -> str:
"""
Model path, huggingface slug, file path, or blob link
"""
return self._model
@property
def revision(self) -> _types.OptionalString:
"""
Model repo revision
"""
return self._revision
@property
def variant(self) -> _types.OptionalString:
"""
Model repo revision
"""
return self._variant
@property
def subfolder(self) -> _types.OptionalPath:
"""
Model repo subfolder
"""
return self._subfolder
@property
def dtype(self) -> _enums.DataType | None:
"""
Model dtype (precision)
"""
return self._dtype
[docs]
def __init__(self,
model: str,
revision: _types.OptionalString = None,
variant: _types.OptionalString = None,
subfolder: _types.OptionalString = None,
dtype: _enums.DataType | str | None = None):
"""
:param model: model path
:param revision: model revision (branch name)
:param variant: model variant, for example ``fp16``
:param subfolder: model subfolder
:param dtype: model data type (precision)
:raises InvalidImageEncoderUriError: If ``model`` points to a single file,
single file loads are not supported. Or if ``dtype`` is passed an
invalid data type string.
"""
if _hfhub.is_single_file_model_load(model):
raise _exceptions.InvalidImageEncoderUriError(
'Loading an Image Encoder from a single file is not supported.')
self._model = model
self._revision = revision
self._variant = variant
try:
self._dtype = _enums.get_data_type_enum(dtype) if dtype else None
except ValueError:
raise _exceptions.InvalidImageEncoderUriError(
f'invalid dtype string, must be one of: {_textprocessing.oxford_comma(_enums.supported_data_type_strings(), "or")}')
self._subfolder = subfolder
[docs]
def load(self,
dtype_fallback: _enums.DataType = _enums.DataType.AUTO,
use_auth_token: _types.OptionalString = None,
local_files_only: bool = False,
no_cache: bool = False,
image_encoder_class:
type[transformers.CLIPVisionModelWithProjection] |
type[_models.SiglipImageEncoder] = transformers.CLIPVisionModelWithProjection) \
-> type[transformers.CLIPVisionModelWithProjection] | type[_models.SiglipImageEncoder]:
"""
Load an Image Encoder Model of type :py:class:`transformers.CLIPVisionModelWithProjection`
:param dtype_fallback: If the URI does not specify a dtype, use this dtype.
:param use_auth_token: optional huggingface auth token.
:param local_files_only: avoid downloading files and only look for cached files
when the model path is a huggingface slug or blob link
:param no_cache: If True, force the returned object not to be cached by the memoize decorator.
:param image_encoder_class: Image Encoder class to load.
:raises ModelNotFoundError: If the model could not be found.
:return: :py:class:`transformers.CLIPVisionModelWithProjection`
"""
def cache_all(e):
raise _exceptions.ImageEncoderUriLoadError(
f'error loading Image Encoder "{self.model}": {e}') from e
with _hfhub.with_hf_errors_as_model_not_found(cache_all):
args = locals()
args.pop('self')
args.pop('cache_all')
return self._load(**args)
@staticmethod
def _enforce_cache_size(new_image_encoder_size):
_image_encoder_cache.enforce_cpu_mem_constraints(
_constants.IMAGE_ENCODER_CACHE_MEMORY_CONSTRAINTS,
size_var='image_encoder_size',
new_object_size=new_image_encoder_size)
@_memoize(_image_encoder_cache,
exceptions={'local_files_only'},
hasher=lambda args: _d_memoize.args_cache_key(args, {'self': _d_memoize.property_hasher}),
on_hit=lambda key, hit: _d_memoize.simple_cache_hit_debug("ImageEncoder", key, hit),
on_create=lambda key, new: _d_memoize.simple_cache_miss_debug("ImageEncoder", key, new))
def _load(self,
dtype_fallback: _enums.DataType = _enums.DataType.AUTO,
use_auth_token: _types.OptionalString = None,
local_files_only: bool = False,
no_cache: bool = False,
image_encoder_class:
type[transformers.CLIPVisionModelWithProjection] |
type[_models.SiglipImageEncoder] = transformers.CLIPVisionModelWithProjection) \
-> type[transformers.CLIPVisionModelWithProjection] | type[_models.SiglipImageEncoder]:
if self.dtype is None:
torch_dtype = _enums.get_torch_dtype(dtype_fallback)
else:
torch_dtype = _enums.get_torch_dtype(self.dtype)
path = self.model
estimated_memory_use = _pipelinewrapper_util.estimate_model_memory_use(
repo_id=path,
revision=self.revision,
variant=self.variant,
subfolder=self.subfolder,
local_files_only=local_files_only,
use_auth_token=use_auth_token
)
self._enforce_cache_size(estimated_memory_use)
if self.subfolder:
extra_args = {'subfolder': self.subfolder}
else:
# flux null reference bug
extra_args = dict()
image_encoder = image_encoder_class.from_pretrained(
path,
revision=self.revision,
variant=self.variant,
torch_dtype=torch_dtype,
token=use_auth_token,
local_files_only=local_files_only,
**extra_args)
_messages.debug_log('Estimated Image Encoder Memory Use:',
_memory.bytes_best_human_unit(estimated_memory_use))
self._enforce_cache_size(estimated_memory_use)
_util._patch_module_to_for_sized_cache(_image_encoder_cache, image_encoder)
# noinspection PyTypeChecker
return image_encoder, _d_memoize.CachedObjectMetadata(
size=estimated_memory_use,
skip=no_cache
)
[docs]
@staticmethod
def parse(uri: _types.Uri) -> 'ImageEncoderUri':
"""
Parse a ``--image-encoder`` uri and return an object representing its constituents
:param uri: string with ``--image-encoder`` uri syntax
:raise InvalidImageEncoderUriError:
:return: :py:class:`.ImageEncoderUri`
"""
try:
r = _image_encoder_uri_parser.parse(uri)
dtype = r.args.get('dtype')
supported_dtypes = _enums.supported_data_type_strings()
if dtype is not None and dtype not in supported_dtypes:
raise _exceptions.InvalidImageEncoderUriError(
f'Image Encoder "dtype" must be {", ".join(supported_dtypes)}, '
f'or left undefined, received: {dtype}')
return ImageEncoderUri(
model=r.concept,
revision=r.args.get('revision', None),
variant=r.args.get('variant', None),
dtype=dtype,
subfolder=r.args.get('subfolder', None))
except _textprocessing.ConceptUriParseError as e:
raise _exceptions.InvalidImageEncoderUriError(e) from e