configuration_utils.py 29.1 KB
Newer Older
1
# coding=utf-8
Patrick von Platen's avatar
Patrick von Platen committed
2
# Copyright 2023 The HuggingFace Inc. team.
3
4
5
6
7
8
9
10
11
12
13
14
15
# Copyright (c) 2022, NVIDIA CORPORATION.  All rights reserved.
#
# 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.
16
""" ConfigMixin base class and utilities."""
17
import dataclasses
18
import functools
19
import importlib
Patrick von Platen's avatar
improve  
Patrick von Platen committed
20
import inspect
21
22
23
import json
import os
import re
Patrick von Platen's avatar
Patrick von Platen committed
24
from collections import OrderedDict
25
from pathlib import PosixPath
26
27
from typing import Any, Dict, Tuple, Union

28
import numpy as np
29
from huggingface_hub import hf_hub_download
30
from huggingface_hub.utils import EntryNotFoundError, RepositoryNotFoundError, RevisionNotFoundError
Patrick von Platen's avatar
Patrick von Platen committed
31
from requests import HTTPError
32

Patrick von Platen's avatar
Patrick von Platen committed
33
from . import __version__
34
35
36
37
38
39
40
41
42
from .utils import (
    DIFFUSERS_CACHE,
    HUGGINGFACE_CO_RESOLVE_ENDPOINT,
    DummyObject,
    deprecate,
    extract_commit_hash,
    http_user_agent,
    logging,
)
43

44

45
46
47
48
49
logger = logging.get_logger(__name__)

_re_configuration_file = re.compile(r"config\.(.*)\.json")


50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
class FrozenDict(OrderedDict):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)

        for key, value in self.items():
            setattr(self, key, value)

        self.__frozen = True

    def __delitem__(self, *args, **kwargs):
        raise Exception(f"You cannot use ``__delitem__`` on a {self.__class__.__name__} instance.")

    def setdefault(self, *args, **kwargs):
        raise Exception(f"You cannot use ``setdefault`` on a {self.__class__.__name__} instance.")

    def pop(self, *args, **kwargs):
        raise Exception(f"You cannot use ``pop`` on a {self.__class__.__name__} instance.")

    def update(self, *args, **kwargs):
        raise Exception(f"You cannot use ``update`` on a {self.__class__.__name__} instance.")

    def __setattr__(self, name, value):
        if hasattr(self, "__frozen") and self.__frozen:
            raise Exception(f"You cannot use ``__setattr__`` on a {self.__class__.__name__} instance.")
        super().__setattr__(name, value)

    def __setitem__(self, name, value):
        if hasattr(self, "__frozen") and self.__frozen:
            raise Exception(f"You cannot use ``__setattr__`` on a {self.__class__.__name__} instance.")
        super().__setitem__(name, value)


Patrick von Platen's avatar
Patrick von Platen committed
82
class ConfigMixin:
83
    r"""
Patrick von Platen's avatar
Patrick von Platen committed
84
85
86
87
88
89
90
    Base class for all configuration classes. Stores all configuration parameters under `self.config` Also handles all
    methods for loading/downloading/saving classes inheriting from [`ConfigMixin`] with
        - [`~ConfigMixin.from_config`]
        - [`~ConfigMixin.save_config`]

    Class attributes:
        - **config_name** (`str`) -- A filename under which the config should stored when calling
91
          [`~ConfigMixin.save_config`] (should be overridden by parent class).
Patrick von Platen's avatar
Patrick von Platen committed
92
        - **ignore_for_config** (`List[str]`) -- A list of attributes that should not be saved in the config (should be
93
94
95
96
97
          overridden by subclass).
        - **has_compatibles** (`bool`) -- Whether the class has compatible classes (should be overridden by subclass).
        - **_deprecated_kwargs** (`List[str]`) -- Keyword arguments that are deprecated. Note that the init function
          should only have a `kwargs` argument if at least one argument is deprecated (should be overridden by
          subclass).
98
    """
99
    config_name = None
Patrick von Platen's avatar
Patrick von Platen committed
100
    ignore_for_config = []
101
    has_compatibles = False
102

103
104
    _deprecated_kwargs = []

105
    def register_to_config(self, **kwargs):
106
107
        if self.config_name is None:
            raise NotImplementedError(f"Make sure that {self.__class__} has defined a class name `config_name`")
108
109
110
111
        # Special case for `kwargs` used in deprecation warning added to schedulers
        # TODO: remove this when we remove the deprecation warning, and the `kwargs` argument,
        # or solve in a more general way.
        kwargs.pop("kwargs", None)
Patrick von Platen's avatar
Patrick von Platen committed
112

Patrick von Platen's avatar
Patrick von Platen committed
113
114
115
116
117
118
        if not hasattr(self, "_internal_dict"):
            internal_dict = kwargs
        else:
            previous_dict = dict(self._internal_dict)
            internal_dict = {**self._internal_dict, **kwargs}
            logger.debug(f"Updating config from {previous_dict} to {internal_dict}")
119

Patrick von Platen's avatar
Patrick von Platen committed
120
        self._internal_dict = FrozenDict(internal_dict)
121

122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
    def __getattr__(self, name: str) -> Any:
        """The only reason we overwrite `getattr` here is to gracefully deprecate accessing
        config attributes directly. See https://github.com/huggingface/diffusers/pull/3129

        Tihs funtion is mostly copied from PyTorch's __getattr__ overwrite:
        https://pytorch.org/docs/stable/_modules/torch/nn/modules/module.html#Module
        """

        is_in_config = "_internal_dict" in self.__dict__ and hasattr(self.__dict__["_internal_dict"], name)
        is_attribute = name in self.__dict__

        if is_in_config and not is_attribute:
            deprecation_message = f"Accessing config attribute `{name}` directly via '{type(self).__name__}' object attribute is deprecated. Please access '{name}' over '{type(self).__name__}'s config object instead, e.g. 'scheduler.config.{name}'."
            deprecate("direct config name access", "1.0.0", deprecation_message, standard_warn=False)
            return self._internal_dict[name]

        raise AttributeError(f"'{type(self).__name__}' object has no attribute '{name}'")

140
    def save_config(self, save_directory: Union[str, os.PathLike], push_to_hub: bool = False, **kwargs):
141
142
        """
        Save a configuration object to the directory `save_directory`, so that it can be re-loaded using the
Patrick von Platen's avatar
Patrick von Platen committed
143
        [`~ConfigMixin.from_config`] class method.
144
145
146
147
148
149
150
151
152
153

        Args:
            save_directory (`str` or `os.PathLike`):
                Directory where the configuration JSON file will be saved (will be created if it does not exist).
        """
        if os.path.isfile(save_directory):
            raise AssertionError(f"Provided path ({save_directory}) should be a directory, not a file")

        os.makedirs(save_directory, exist_ok=True)

154
155
        # If we save using the predefined names, we can load using `from_config`
        output_config_file = os.path.join(save_directory, self.config_name)
156

157
        self.to_json_file(output_config_file)
Pedro Cuenca's avatar
Pedro Cuenca committed
158
        logger.info(f"Configuration saved in {output_config_file}")
159

160
    @classmethod
161
    def from_config(cls, config: Union[FrozenDict, Dict[str, Any]] = None, return_unused_kwargs=False, **kwargs):
Patrick von Platen's avatar
Patrick von Platen committed
162
        r"""
Steven Liu's avatar
Steven Liu committed
163
        Instantiate a Python class from a config dictionary.
164
165
166
167
168
169
170
171
172

        Parameters:
            config (`Dict[str, Any]`):
                A config dictionary from which the Python class will be instantiated. Make sure to only load
                configuration files of compatible classes.
            return_unused_kwargs (`bool`, *optional*, defaults to `False`):
                Whether kwargs that are not consumed by the Python class should be returned or not.

            kwargs (remaining dictionary of keyword arguments, *optional*):
Steven Liu's avatar
Steven Liu committed
173
174
175
176
177
178
179
                Can be used to update the configuration object (after it is loaded) and initiate the Python class.
                `**kwargs` are directly passed to the underlying scheduler/model's `__init__` method and eventually
                overwrite same named arguments in `config`.

        Returns:
            [`ModelMixin`] or [`SchedulerMixin`]:
                A model or scheduler object instantiated from a config dictionary.
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227

        Examples:

        ```python
        >>> from diffusers import DDPMScheduler, DDIMScheduler, PNDMScheduler

        >>> # Download scheduler from huggingface.co and cache.
        >>> scheduler = DDPMScheduler.from_pretrained("google/ddpm-cifar10-32")

        >>> # Instantiate DDIM scheduler class with same config as DDPM
        >>> scheduler = DDIMScheduler.from_config(scheduler.config)

        >>> # Instantiate PNDM scheduler class with same config as DDPM
        >>> scheduler = PNDMScheduler.from_config(scheduler.config)
        ```
        """
        # <===== TO BE REMOVED WITH DEPRECATION
        # TODO(Patrick) - make sure to remove the following lines when config=="model_path" is deprecated
        if "pretrained_model_name_or_path" in kwargs:
            config = kwargs.pop("pretrained_model_name_or_path")

        if config is None:
            raise ValueError("Please make sure to provide a config as the first positional argument.")
        # ======>

        if not isinstance(config, dict):
            deprecation_message = "It is deprecated to pass a pretrained model name or path to `from_config`."
            if "Scheduler" in cls.__name__:
                deprecation_message += (
                    f"If you were trying to load a scheduler, please use {cls}.from_pretrained(...) instead."
                    " Otherwise, please make sure to pass a configuration dictionary instead. This functionality will"
                    " be removed in v1.0.0."
                )
            elif "Model" in cls.__name__:
                deprecation_message += (
                    f"If you were trying to load a model, please use {cls}.load_config(...) followed by"
                    f" {cls}.from_config(...) instead. Otherwise, please make sure to pass a configuration dictionary"
                    " instead. This functionality will be removed in v1.0.0."
                )
            deprecate("config-passed-as-path", "1.0.0", deprecation_message, standard_warn=False)
            config, kwargs = cls.load_config(pretrained_model_name_or_path=config, return_unused_kwargs=True, **kwargs)

        init_dict, unused_kwargs, hidden_dict = cls.extract_init_dict(config, **kwargs)

        # Allow dtype to be specified on initialization
        if "dtype" in unused_kwargs:
            init_dict["dtype"] = unused_kwargs.pop("dtype")

228
229
230
231
        # add possible deprecated kwargs
        for deprecated_kwarg in cls._deprecated_kwargs:
            if deprecated_kwarg in unused_kwargs:
                init_dict[deprecated_kwarg] = unused_kwargs.pop(deprecated_kwarg)
232

233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
        # Return model and optionally state and/or unused_kwargs
        model = cls(**init_dict)

        # make sure to also save config parameters that might be used for compatible classes
        model.register_to_config(**hidden_dict)

        # add hidden kwargs of compatible classes to unused_kwargs
        unused_kwargs = {**unused_kwargs, **hidden_dict}

        if return_unused_kwargs:
            return (model, unused_kwargs)
        else:
            return model

    @classmethod
    def get_config_dict(cls, *args, **kwargs):
        deprecation_message = (
            f" The function get_config_dict is deprecated. Please use {cls}.load_config instead. This function will be"
            " removed in version v1.0.0"
        )
        deprecate("get_config_dict", "1.0.0", deprecation_message, standard_warn=False)
        return cls.load_config(*args, **kwargs)

    @classmethod
    def load_config(
258
259
260
261
262
        cls,
        pretrained_model_name_or_path: Union[str, os.PathLike],
        return_unused_kwargs=False,
        return_commit_hash=False,
        **kwargs,
263
264
    ) -> Tuple[Dict[str, Any], Dict[str, Any]]:
        r"""
Steven Liu's avatar
Steven Liu committed
265
        Load a model or scheduler configuration.
Patrick von Platen's avatar
Patrick von Platen committed
266
267
268
269
270

        Parameters:
            pretrained_model_name_or_path (`str` or `os.PathLike`, *optional*):
                Can be either:

Steven Liu's avatar
Steven Liu committed
271
272
273
274
                    - A string, the *model id* (for example `google/ddpm-celebahq-256`) of a pretrained model hosted on
                      the Hub.
                    - A path to a *directory* (for example `./my_model_directory`) containing model weights saved with
                      [`~ConfigMixin.save_config`].
Patrick von Platen's avatar
Patrick von Platen committed
275
276

            cache_dir (`Union[str, os.PathLike]`, *optional*):
Steven Liu's avatar
Steven Liu committed
277
278
                Path to a directory where a downloaded pretrained model configuration is cached if the standard cache
                is not used.
Patrick von Platen's avatar
Patrick von Platen committed
279
280
281
282
            force_download (`bool`, *optional*, defaults to `False`):
                Whether or not to force the (re-)download of the model weights and configuration files, overriding the
                cached versions if they exist.
            resume_download (`bool`, *optional*, defaults to `False`):
Steven Liu's avatar
Steven Liu committed
283
284
                Whether or not to resume downloading the model weights and configuration files. If set to False, any
                incompletely downloaded files are deleted.
Patrick von Platen's avatar
Patrick von Platen committed
285
            proxies (`Dict[str, str]`, *optional*):
Steven Liu's avatar
Steven Liu committed
286
                A dictionary of proxy servers to use by protocol or endpoint, for example, `{'http': 'foo.bar:3128',
Patrick von Platen's avatar
Patrick von Platen committed
287
288
                'http://hostname': 'foo.bar:4012'}`. The proxies are used on each request.
            output_loading_info(`bool`, *optional*, defaults to `False`):
289
                Whether or not to also return a dictionary containing missing keys, unexpected keys and error messages.
Patrick von Platen's avatar
Patrick von Platen committed
290
            local_files_only(`bool`, *optional*, defaults to `False`):
Steven Liu's avatar
Steven Liu committed
291
292
                Whether to only load local model weights and configuration files or not. If set to True, the model
                won’t be downloaded from the Hub.
Patrick von Platen's avatar
Patrick von Platen committed
293
            use_auth_token (`str` or *bool*, *optional*):
Steven Liu's avatar
Steven Liu committed
294
295
                The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                `diffusers-cli login` (stored in `~/.huggingface`) is used.
Patrick von Platen's avatar
Patrick von Platen committed
296
            revision (`str`, *optional*, defaults to `"main"`):
Steven Liu's avatar
Steven Liu committed
297
298
                The specific model version to use. It can be a branch name, a tag name, a commit id, or any identifier
                allowed by Git.
299
            subfolder (`str`, *optional*, defaults to `""`):
Steven Liu's avatar
Steven Liu committed
300
                The subfolder location of a model file within a larger model repository on the Hub or locally.
301
            return_unused_kwargs (`bool`, *optional*, defaults to `False):
Steven Liu's avatar
Steven Liu committed
302
                Whether unused keyword arguments of the config are returned.
303
            return_commit_hash (`bool`, *optional*, defaults to `False):
Steven Liu's avatar
Steven Liu committed
304
                Whether the `commit_hash` of the loaded configuration are returned.
Patrick von Platen's avatar
Patrick von Platen committed
305

Steven Liu's avatar
Steven Liu committed
306
307
308
        Returns:
            `dict`:
                A dictionary of all the parameters stored in a JSON configuration file.
Patrick von Platen's avatar
Patrick von Platen committed
309
310
311

        <Tip>

Steven Liu's avatar
Steven Liu committed
312
313
314
315
        To use private or [gated models](https://huggingface.co/docs/hub/models-gated#gated-models), log-in with
        `huggingface-cli login`. You can also activate the special
        ["offline-mode"](https://huggingface.co/transformers/installation.html#offline-mode) to use this method in a
        firewalled environment.
Patrick von Platen's avatar
Patrick von Platen committed
316
317
318

        </Tip>
        """
319
        cache_dir = kwargs.pop("cache_dir", DIFFUSERS_CACHE)
320
321
322
323
324
325
        force_download = kwargs.pop("force_download", False)
        resume_download = kwargs.pop("resume_download", False)
        proxies = kwargs.pop("proxies", None)
        use_auth_token = kwargs.pop("use_auth_token", None)
        local_files_only = kwargs.pop("local_files_only", False)
        revision = kwargs.pop("revision", None)
326
        _ = kwargs.pop("mirror", None)
Patrick von Platen's avatar
Patrick von Platen committed
327
        subfolder = kwargs.pop("subfolder", None)
328
        user_agent = kwargs.pop("user_agent", {})
329

330
331
        user_agent = {**user_agent, "file_type": "config"}
        user_agent = http_user_agent(user_agent)
332
333
334

        pretrained_model_name_or_path = str(pretrained_model_name_or_path)

335
336
337
338
339
340
        if cls.config_name is None:
            raise ValueError(
                "`self.config_name` is not defined. Note that one should not load a config from "
                "`ConfigMixin`. Please make sure to define `config_name` in a class inheriting from `ConfigMixin`"
            )

341
342
343
344
345
346
        if os.path.isfile(pretrained_model_name_or_path):
            config_file = pretrained_model_name_or_path
        elif os.path.isdir(pretrained_model_name_or_path):
            if os.path.isfile(os.path.join(pretrained_model_name_or_path, cls.config_name)):
                # Load from a PyTorch checkpoint
                config_file = os.path.join(pretrained_model_name_or_path, cls.config_name)
Patrick von Platen's avatar
Patrick von Platen committed
347
348
349
350
            elif subfolder is not None and os.path.isfile(
                os.path.join(pretrained_model_name_or_path, subfolder, cls.config_name)
            ):
                config_file = os.path.join(pretrained_model_name_or_path, subfolder, cls.config_name)
351
            else:
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
                raise EnvironmentError(
                    f"Error no file named {cls.config_name} found in directory {pretrained_model_name_or_path}."
                )
        else:
            try:
                # Load from URL or cache if already cached
                config_file = hf_hub_download(
                    pretrained_model_name_or_path,
                    filename=cls.config_name,
                    cache_dir=cache_dir,
                    force_download=force_download,
                    proxies=proxies,
                    resume_download=resume_download,
                    local_files_only=local_files_only,
                    use_auth_token=use_auth_token,
                    user_agent=user_agent,
Patrick von Platen's avatar
Patrick von Platen committed
368
                    subfolder=subfolder,
369
                    revision=revision,
370
                )
371
372
            except RepositoryNotFoundError:
                raise EnvironmentError(
Patrick von Platen's avatar
Patrick von Platen committed
373
374
375
                    f"{pretrained_model_name_or_path} is not a local folder and is not a valid model identifier"
                    " listed on 'https://huggingface.co/models'\nIf this is a private repository, make sure to pass a"
                    " token having permission to this repo with `use_auth_token` or log in with `huggingface-cli"
376
                    " login`."
377
378
379
                )
            except RevisionNotFoundError:
                raise EnvironmentError(
Patrick von Platen's avatar
Patrick von Platen committed
380
381
382
                    f"{revision} is not a valid git identifier (branch name, tag name or commit id) that exists for"
                    " this model name. Check the model page at"
                    f" 'https://huggingface.co/{pretrained_model_name_or_path}' for available revisions."
383
384
385
386
387
388
389
                )
            except EntryNotFoundError:
                raise EnvironmentError(
                    f"{pretrained_model_name_or_path} does not appear to have a file named {cls.config_name}."
                )
            except HTTPError as err:
                raise EnvironmentError(
Patrick von Platen's avatar
Patrick von Platen committed
390
391
                    "There was a specific connection error when trying to load"
                    f" {pretrained_model_name_or_path}:\n{err}"
392
393
394
                )
            except ValueError:
                raise EnvironmentError(
Patrick von Platen's avatar
Patrick von Platen committed
395
396
397
398
399
                    f"We couldn't connect to '{HUGGINGFACE_CO_RESOLVE_ENDPOINT}' to load this model, couldn't find it"
                    f" in the cached files and it looks like {pretrained_model_name_or_path} is not the path to a"
                    f" directory containing a {cls.config_name} file.\nCheckout your internet connection or see how to"
                    " run the library in offline mode at"
                    " 'https://huggingface.co/docs/diffusers/installation#offline-mode'."
400
401
402
403
404
405
406
407
                )
            except EnvironmentError:
                raise EnvironmentError(
                    f"Can't load config for '{pretrained_model_name_or_path}'. If you were trying to load it from "
                    "'https://huggingface.co/models', make sure you don't have a local directory with the same name. "
                    f"Otherwise, make sure '{pretrained_model_name_or_path}' is the correct path to a directory "
                    f"containing a {cls.config_name} file"
                )
408

409
410
411
        try:
            # Load config dict
            config_dict = cls._dict_from_json_file(config_file)
412
413

            commit_hash = extract_commit_hash(config_file)
414
        except (json.JSONDecodeError, UnicodeDecodeError):
Patrick von Platen's avatar
Patrick von Platen committed
415
            raise EnvironmentError(f"It looks like the config file at '{config_file}' is not a valid JSON file.")
416

417
418
419
420
421
        if not (return_unused_kwargs or return_commit_hash):
            return config_dict

        outputs = (config_dict,)

422
        if return_unused_kwargs:
423
424
425
426
            outputs += (kwargs,)

        if return_commit_hash:
            outputs += (commit_hash,)
427

428
        return outputs
429

430
431
432
433
    @staticmethod
    def _get_init_keys(cls):
        return set(dict(inspect.signature(cls.__init__).parameters).keys())

patil-suraj's avatar
patil-suraj committed
434
435
    @classmethod
    def extract_init_dict(cls, config_dict, **kwargs):
436
        # 0. Copy origin config dict
437
        original_dict = dict(config_dict.items())
438

439
440
        # 1. Retrieve expected config attributes from __init__ signature
        expected_keys = cls._get_init_keys(cls)
441
        expected_keys.remove("self")
Patrick von Platen's avatar
hotfix  
Patrick von Platen committed
442
443
444
        # remove general kwargs if present in dict
        if "kwargs" in expected_keys:
            expected_keys.remove("kwargs")
Yuta Hayashibe's avatar
Yuta Hayashibe committed
445
        # remove flax internal keys
446
447
448
449
        if hasattr(cls, "_flax_internal_args"):
            for arg in cls._flax_internal_args:
                expected_keys.remove(arg)

450
        # 2. Remove attributes that cannot be expected from expected config attributes
Patrick von Platen's avatar
Patrick von Platen committed
451
452
453
        # remove keys to be ignored
        if len(cls.ignore_for_config) > 0:
            expected_keys = expected_keys - set(cls.ignore_for_config)
454
455
456
457

        # load diffusers library to import compatible and original scheduler
        diffusers_library = importlib.import_module(__name__.split(".")[0])

458
459
460
461
462
        if cls.has_compatibles:
            compatible_classes = [c for c in cls._get_compatibles() if not isinstance(c, DummyObject)]
        else:
            compatible_classes = []

463
464
465
466
467
468
469
470
471
        expected_keys_comp_cls = set()
        for c in compatible_classes:
            expected_keys_c = cls._get_init_keys(c)
            expected_keys_comp_cls = expected_keys_comp_cls.union(expected_keys_c)
        expected_keys_comp_cls = expected_keys_comp_cls - cls._get_init_keys(cls)
        config_dict = {k: v for k, v in config_dict.items() if k not in expected_keys_comp_cls}

        # remove attributes from orig class that cannot be expected
        orig_cls_name = config_dict.pop("_class_name", cls.__name__)
472
        if orig_cls_name != cls.__name__ and hasattr(diffusers_library, orig_cls_name):
473
474
475
476
477
478
479
480
            orig_cls = getattr(diffusers_library, orig_cls_name)
            unexpected_keys_from_orig = cls._get_init_keys(orig_cls) - expected_keys
            config_dict = {k: v for k, v in config_dict.items() if k not in unexpected_keys_from_orig}

        # remove private attributes
        config_dict = {k: v for k, v in config_dict.items() if not k.startswith("_")}

        # 3. Create keyword arguments that will be passed to __init__ from expected keyword arguments
patil-suraj's avatar
patil-suraj committed
481
        init_dict = {}
Patrick von Platen's avatar
improve  
Patrick von Platen committed
482
        for key in expected_keys:
483
484
485
486
487
            # if config param is passed to kwarg and is present in config dict
            # it should overwrite existing config dict key
            if key in kwargs and key in config_dict:
                config_dict[key] = kwargs.pop(key)

Patrick von Platen's avatar
improve  
Patrick von Platen committed
488
489
            if key in kwargs:
                # overwrite key
patil-suraj's avatar
patil-suraj committed
490
491
492
493
                init_dict[key] = kwargs.pop(key)
            elif key in config_dict:
                # use value from config dict
                init_dict[key] = config_dict.pop(key)
Patrick von Platen's avatar
improve  
Patrick von Platen committed
494

495
        # 4. Give nice warning if unexpected values have been passed
496
497
498
499
500
501
502
        if len(config_dict) > 0:
            logger.warning(
                f"The config attributes {config_dict} were passed to {cls.__name__}, "
                "but are not expected and will be ignored. Please verify your "
                f"{cls.config_name} configuration file."
            )

503
        # 5. Give nice info if config attributes are initiliazed to default because they have not been passed
patil-suraj's avatar
patil-suraj committed
504
        passed_keys = set(init_dict.keys())
505
        if len(expected_keys - passed_keys) > 0:
506
            logger.info(
Patrick von Platen's avatar
improve  
Patrick von Platen committed
507
                f"{expected_keys - passed_keys} was not found in config. Values will be initialized to default values."
508
            )
509

510
511
512
        # 6. Define unused keyword arguments
        unused_kwargs = {**config_dict, **kwargs}

513
        # 7. Define "hidden" config parameters that were saved for compatible classes
514
        hidden_config_dict = {k: v for k, v in original_dict.items() if k not in init_dict}
515
516

        return init_dict, unused_kwargs, hidden_config_dict
Patrick von Platen's avatar
Patrick von Platen committed
517

518
519
520
521
522
523
    @classmethod
    def _dict_from_json_file(cls, json_file: Union[str, os.PathLike]):
        with open(json_file, "r", encoding="utf-8") as reader:
            text = reader.read()
        return json.loads(text)

anton-l's avatar
anton-l committed
524
    def __repr__(self):
525
        return f"{self.__class__.__name__} {self.to_json_string()}"
526

527
528
    @property
    def config(self) -> Dict[str, Any]:
529
530
531
532
533
534
        """
        Returns the config of the class as a frozen dictionary

        Returns:
            `Dict[str, Any]`: Config of the class.
        """
Patrick von Platen's avatar
Patrick von Platen committed
535
        return self._internal_dict
536

537
    def to_json_string(self) -> str:
538
539
540
541
542
543
        """
        Serializes this instance to a JSON string.

        Returns:
            `str`: String containing all the attributes that make up this configuration instance in JSON format.
        """
544
        config_dict = self._internal_dict if hasattr(self, "_internal_dict") else {}
545
546
547
        config_dict["_class_name"] = self.__class__.__name__
        config_dict["_diffusers_version"] = __version__

548
549
550
        def to_json_saveable(value):
            if isinstance(value, np.ndarray):
                value = value.tolist()
551
552
            elif isinstance(value, PosixPath):
                value = str(value)
553
554
555
            return value

        config_dict = {k: to_json_saveable(v) for k, v in config_dict.items()}
Patrick von Platen's avatar
Patrick von Platen committed
556
557
558
        # Don't save "_ignore_files"
        config_dict.pop("_ignore_files", None)

559
560
        return json.dumps(config_dict, indent=2, sort_keys=True) + "\n"

561
    def to_json_file(self, json_file_path: Union[str, os.PathLike]):
562
563
564
565
566
567
568
569
        """
        Save this instance to a JSON file.

        Args:
            json_file_path (`str` or `os.PathLike`):
                Path to the JSON file in which this configuration instance's parameters will be saved.
        """
        with open(json_file_path, "w", encoding="utf-8") as writer:
570
            writer.write(self.to_json_string())
Patrick von Platen's avatar
Patrick von Platen committed
571
572


573
def register_to_config(init):
Patrick von Platen's avatar
Patrick von Platen committed
574
575
576
577
    r"""
    Decorator to apply on the init of classes inheriting from [`ConfigMixin`] so that all the arguments are
    automatically sent to `self.register_for_config`. To ignore a specific argument accepted by the init but that
    shouldn't be registered in the config, use the `ignore_for_config` class variable
578
579
580
581
582
583
584
585

    Warning: Once decorated, all private arguments (beginning with an underscore) are trashed and not sent to the init!
    """

    @functools.wraps(init)
    def inner_init(self, *args, **kwargs):
        # Ignore private kwargs in the init.
        init_kwargs = {k: v for k, v in kwargs.items() if not k.startswith("_")}
586
        config_init_kwargs = {k: v for k, v in kwargs.items() if k.startswith("_")}
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
        if not isinstance(self, ConfigMixin):
            raise RuntimeError(
                f"`@register_for_config` was applied to {self.__class__.__name__} init method, but this class does "
                "not inherit from `ConfigMixin`."
            )

        ignore = getattr(self, "ignore_for_config", [])
        # Get positional arguments aligned with kwargs
        new_kwargs = {}
        signature = inspect.signature(init)
        parameters = {
            name: p.default for i, (name, p) in enumerate(signature.parameters.items()) if i > 0 and name not in ignore
        }
        for arg, name in zip(args, parameters.keys()):
            new_kwargs[name] = arg

        # Then add all kwargs
        new_kwargs.update(
            {
                k: init_kwargs.get(k, default)
                for k, default in parameters.items()
                if k not in ignore and k not in new_kwargs
            }
        )
611
        new_kwargs = {**config_init_kwargs, **new_kwargs}
612
        getattr(self, "register_to_config")(**new_kwargs)
613
        init(self, *args, **init_kwargs)
614
615

    return inner_init
616
617
618
619
620
621
622
623
624
625
626
627
628
629


def flax_register_to_config(cls):
    original_init = cls.__init__

    @functools.wraps(original_init)
    def init(self, *args, **kwargs):
        if not isinstance(self, ConfigMixin):
            raise RuntimeError(
                f"`@register_for_config` was applied to {self.__class__.__name__} init method, but this class does "
                "not inherit from `ConfigMixin`."
            )

        # Ignore private kwargs in the init. Retrieve all passed attributes
630
        init_kwargs = dict(kwargs.items())
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645

        # Retrieve default values
        fields = dataclasses.fields(self)
        default_kwargs = {}
        for field in fields:
            # ignore flax specific attributes
            if field.name in self._flax_internal_args:
                continue
            if type(field.default) == dataclasses._MISSING_TYPE:
                default_kwargs[field.name] = None
            else:
                default_kwargs[field.name] = getattr(self, field.name)

        # Make sure init_kwargs override default kwargs
        new_kwargs = {**default_kwargs, **init_kwargs}
646
647
648
        # dtype should be part of `init_kwargs`, but not `new_kwargs`
        if "dtype" in new_kwargs:
            new_kwargs.pop("dtype")
649
650
651
652
653
654
655
656
657
658
659

        # Get positional arguments aligned with kwargs
        for i, arg in enumerate(args):
            name = fields[i].name
            new_kwargs[name] = arg

        getattr(self, "register_to_config")(**new_kwargs)
        original_init(self, *args, **kwargs)

    cls.__init__ = init
    return cls