File size: 20,484 Bytes
4c65bff
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
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
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
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
228
229
230
231
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
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
# coding=utf-8
# Copyright 2021 The HuggingFace Inc. team. 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.
""" CLIP model configuration"""

import os
from collections import OrderedDict
from typing import TYPE_CHECKING, Any, Mapping, Optional, Union


if TYPE_CHECKING:
    from ...processing_utils import ProcessorMixin
    from ...utils import TensorType

from ...configuration_utils import PretrainedConfig
from ...onnx import OnnxConfig
from ...utils import logging


logger = logging.get_logger(__name__)

CLIP_PRETRAINED_CONFIG_ARCHIVE_MAP = {
    "openai/clip-vit-base-patch32": "https://huggingface.co/openai/clip-vit-base-patch32/resolve/main/config.json",
    # See all CLIP models at https://huggingface.co/models?filter=clip
}


class CLIPTextConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`CLIPTextModel`]. It is used to instantiate a CLIP
    text encoder according to the specified arguments, defining the model architecture. Instantiating a configuration
    with the defaults will yield a similar configuration to that of the text encoder of the CLIP
    [openai/clip-vit-base-patch32](https://huggingface.co/openai/clip-vit-base-patch32) architecture.

    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
    documentation from [`PretrainedConfig`] for more information.

    Args:
        vocab_size (`int`, *optional*, defaults to 49408):
            Vocabulary size of the CLIP text model. Defines the number of different tokens that can be represented by
            the `inputs_ids` passed when calling [`CLIPModel`].
        hidden_size (`int`, *optional*, defaults to 512):
            Dimensionality of the encoder layers and the pooler layer.
        intermediate_size (`int`, *optional*, defaults to 2048):
            Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
        num_hidden_layers (`int`, *optional*, defaults to 12):
            Number of hidden layers in the Transformer encoder.
        num_attention_heads (`int`, *optional*, defaults to 8):
            Number of attention heads for each attention layer in the Transformer encoder.
        max_position_embeddings (`int`, *optional*, defaults to 77):
            The maximum sequence length that this model might ever be used with. Typically set this to something large
            just in case (e.g., 512 or 1024 or 2048).
        hidden_act (`str` or `function`, *optional*, defaults to `"quick_gelu"`):
            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
            `"relu"`, `"selu"` and `"gelu_new"` `"quick_gelu"` are supported.
        layer_norm_eps (`float`, *optional*, defaults to 1e-5):
            The epsilon used by the layer normalization layers.
        attention_dropout (`float`, *optional*, defaults to 0.0):
            The dropout ratio for the attention probabilities.
        initializer_range (`float`, *optional*, defaults to 0.02):
            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
        initializer_factor (`float`, *optional*, defaults to 1):
            A factor for initializing all weight matrices (should be kept to 1, used internally for initialization
            testing).

    Example:

    ```python
    >>> from transformers import CLIPTextConfig, CLIPTextModel

    >>> # Initializing a CLIPTextConfig with openai/clip-vit-base-patch32 style configuration
    >>> configuration = CLIPTextConfig()

    >>> # Initializing a CLIPTextModel (with random weights) from the openai/clip-vit-base-patch32 style configuration
    >>> model = CLIPTextModel(configuration)

    >>> # Accessing the model configuration
    >>> configuration = model.config
    ```"""
    model_type = "clip_text_model"

    def __init__(
        self,
        vocab_size=49408,
        hidden_size=512,
        intermediate_size=2048,
        projection_dim=512,
        num_hidden_layers=12,
        num_attention_heads=8,
        max_position_embeddings=77,
        hidden_act="quick_gelu",
        layer_norm_eps=1e-5,
        attention_dropout=0.0,
        initializer_range=0.02,
        initializer_factor=1.0,
        # This differs from `CLIPTokenizer`'s default and from openai/clip
        # See https://github.com/huggingface/transformers/pull/24773#issuecomment-1632287538
        pad_token_id=1,
        bos_token_id=49406,
        eos_token_id=49407,
        **kwargs,
    ):
        super().__init__(pad_token_id=pad_token_id, bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)

        self.vocab_size = vocab_size
        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        self.projection_dim = projection_dim
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.max_position_embeddings = max_position_embeddings
        self.layer_norm_eps = layer_norm_eps
        self.hidden_act = hidden_act
        self.initializer_range = initializer_range
        self.initializer_factor = initializer_factor
        self.attention_dropout = attention_dropout

    @classmethod
    def from_pretrained(cls, pretrained_model_name_or_path: Union[str, os.PathLike], **kwargs) -> "PretrainedConfig":
        cls._set_token_in_kwargs(kwargs)

        config_dict, kwargs = cls.get_config_dict(pretrained_model_name_or_path, **kwargs)

        # get the text config dict if we are loading from CLIPConfig
        if config_dict.get("model_type") == "clip":
            config_dict = config_dict["text_config"]

        if "model_type" in config_dict and hasattr(cls, "model_type") and config_dict["model_type"] != cls.model_type:
            logger.warning(
                f"You are using a model of type {config_dict['model_type']} to instantiate a model of type "
                f"{cls.model_type}. This is not supported for all configurations of models and can yield errors."
            )

        return cls.from_dict(config_dict, **kwargs)


class CLIPVisionConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`CLIPVisionModel`]. It is used to instantiate a
    CLIP vision encoder according to the specified arguments, defining the model architecture. Instantiating a
    configuration with the defaults will yield a similar configuration to that of the vision encoder of the CLIP
    [openai/clip-vit-base-patch32](https://huggingface.co/openai/clip-vit-base-patch32) architecture.

    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
    documentation from [`PretrainedConfig`] for more information.

    Args:
        hidden_size (`int`, *optional*, defaults to 768):
            Dimensionality of the encoder layers and the pooler layer.
        intermediate_size (`int`, *optional*, defaults to 3072):
            Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
        num_hidden_layers (`int`, *optional*, defaults to 12):
            Number of hidden layers in the Transformer encoder.
        num_attention_heads (`int`, *optional*, defaults to 12):
            Number of attention heads for each attention layer in the Transformer encoder.
        image_size (`int`, *optional*, defaults to 224):
            The size (resolution) of each image.
        patch_size (`int`, *optional*, defaults to 32):
            The size (resolution) of each patch.
        hidden_act (`str` or `function`, *optional*, defaults to `"quick_gelu"`):
            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
            `"relu"`, `"selu"` and `"gelu_new"` ``"quick_gelu"` are supported.
        layer_norm_eps (`float`, *optional*, defaults to 1e-5):
            The epsilon used by the layer normalization layers.
        attention_dropout (`float`, *optional*, defaults to 0.0):
            The dropout ratio for the attention probabilities.
        initializer_range (`float`, *optional*, defaults to 0.02):
            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
        initializer_factor (`float`, *optional*, defaults to 1):
            A factor for initializing all weight matrices (should be kept to 1, used internally for initialization
            testing).

    Example:

    ```python
    >>> from transformers import CLIPVisionConfig, CLIPVisionModel

    >>> # Initializing a CLIPVisionConfig with openai/clip-vit-base-patch32 style configuration
    >>> configuration = CLIPVisionConfig()

    >>> # Initializing a CLIPVisionModel (with random weights) from the openai/clip-vit-base-patch32 style configuration
    >>> model = CLIPVisionModel(configuration)

    >>> # Accessing the model configuration
    >>> configuration = model.config
    ```"""

    model_type = "clip_vision_model"

    def __init__(
        self,
        hidden_size=768,
        intermediate_size=3072,
        projection_dim=512,
        num_hidden_layers=12,
        num_attention_heads=12,
        num_channels=3,
        image_size=224,
        patch_size=32,
        hidden_act="quick_gelu",
        layer_norm_eps=1e-5,
        attention_dropout=0.0,
        initializer_range=0.02,
        initializer_factor=1.0,
        **kwargs,
    ):
        super().__init__(**kwargs)

        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        self.projection_dim = projection_dim
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.num_channels = num_channels
        self.patch_size = patch_size
        self.image_size = image_size
        self.initializer_range = initializer_range
        self.initializer_factor = initializer_factor
        self.attention_dropout = attention_dropout
        self.layer_norm_eps = layer_norm_eps
        self.hidden_act = hidden_act

    @classmethod
    def from_pretrained(cls, pretrained_model_name_or_path: Union[str, os.PathLike], **kwargs) -> "PretrainedConfig":
        cls._set_token_in_kwargs(kwargs)

        config_dict, kwargs = cls.get_config_dict(pretrained_model_name_or_path, **kwargs)

        # get the vision config dict if we are loading from CLIPConfig
        if config_dict.get("model_type") == "clip":
            config_dict = config_dict["vision_config"]

        if "model_type" in config_dict and hasattr(cls, "model_type") and config_dict["model_type"] != cls.model_type:
            logger.warning(
                f"You are using a model of type {config_dict['model_type']} to instantiate a model of type "
                f"{cls.model_type}. This is not supported for all configurations of models and can yield errors."
            )

        return cls.from_dict(config_dict, **kwargs)


class CLIPConfig(PretrainedConfig):
    r"""
    [`CLIPConfig`] is the configuration class to store the configuration of a [`CLIPModel`]. It is used to instantiate
    a CLIP model according to the specified arguments, defining the text model and vision model configs. Instantiating
    a configuration with the defaults will yield a similar configuration to that of the CLIP
    [openai/clip-vit-base-patch32](https://huggingface.co/openai/clip-vit-base-patch32) architecture.

    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
    documentation from [`PretrainedConfig`] for more information.

    Args:
        text_config (`dict`, *optional*):
            Dictionary of configuration options used to initialize [`CLIPTextConfig`].
        vision_config (`dict`, *optional*):
            Dictionary of configuration options used to initialize [`CLIPVisionConfig`].
        projection_dim (`int`, *optional*, defaults to 512):
            Dimentionality of text and vision projection layers.
        logit_scale_init_value (`float`, *optional*, defaults to 2.6592):
            The inital value of the *logit_scale* paramter. Default is used as per the original CLIP implementation.
        kwargs (*optional*):
            Dictionary of keyword arguments.

    Example:

    ```python
    >>> from transformers import CLIPConfig, CLIPModel

    >>> # Initializing a CLIPConfig with openai/clip-vit-base-patch32 style configuration
    >>> configuration = CLIPConfig()

    >>> # Initializing a CLIPModel (with random weights) from the openai/clip-vit-base-patch32 style configuration
    >>> model = CLIPModel(configuration)

    >>> # Accessing the model configuration
    >>> configuration = model.config

    >>> # We can also initialize a CLIPConfig from a CLIPTextConfig and a CLIPVisionConfig
    >>> from transformers import CLIPTextConfig, CLIPVisionConfig

    >>> # Initializing a CLIPText and CLIPVision configuration
    >>> config_text = CLIPTextConfig()
    >>> config_vision = CLIPVisionConfig()

    >>> config = CLIPConfig.from_text_vision_configs(config_text, config_vision)
    ```"""

    model_type = "clip"

    def __init__(
        self, text_config=None, vision_config=None, projection_dim=512, logit_scale_init_value=2.6592, **kwargs
    ):
        # If `_config_dict` exist, we use them for the backward compatibility.
        # We pop out these 2 attributes before calling `super().__init__` to avoid them being saved (which causes a lot
        # of confusion!).
        text_config_dict = kwargs.pop("text_config_dict", None)
        vision_config_dict = kwargs.pop("vision_config_dict", None)

        super().__init__(**kwargs)

        # Instead of simply assigning `[text|vision]_config_dict` to `[text|vision]_config`, we use the values in
        # `[text|vision]_config_dict` to update the values in `[text|vision]_config`. The values should be same in most
        # cases, but we don't want to break anything regarding `_config_dict` that existed before commit `8827e1b2`.
        if text_config_dict is not None:
            if text_config is None:
                text_config = {}

            # This is the complete result when using `text_config_dict`.
            _text_config_dict = CLIPTextConfig(**text_config_dict).to_dict()

            # Give a warning if the values exist in both `_text_config_dict` and `text_config` but being different.
            for key, value in _text_config_dict.items():
                if key in text_config and value != text_config[key] and key not in ["transformers_version"]:
                    # If specified in `text_config_dict`
                    if key in text_config_dict:
                        message = (
                            f"`{key}` is found in both `text_config_dict` and `text_config` but with different values. "
                            f'The value `text_config_dict["{key}"]` will be used instead.'
                        )
                    # If inferred from default argument values (just to be super careful)
                    else:
                        message = (
                            f"`text_config_dict` is provided which will be used to initialize `CLIPTextConfig`. The "
                            f'value `text_config["{key}"]` will be overriden.'
                        )
                    logger.warning(message)

            # Update all values in `text_config` with the ones in `_text_config_dict`.
            text_config.update(_text_config_dict)

        if vision_config_dict is not None:
            if vision_config is None:
                vision_config = {}

            # This is the complete result when using `vision_config_dict`.
            _vision_config_dict = CLIPVisionConfig(**vision_config_dict).to_dict()
            # convert keys to string instead of integer
            if "id2label" in _vision_config_dict:
                _vision_config_dict["id2label"] = {
                    str(key): value for key, value in _vision_config_dict["id2label"].items()
                }

            # Give a warning if the values exist in both `_vision_config_dict` and `vision_config` but being different.
            for key, value in _vision_config_dict.items():
                if key in vision_config and value != vision_config[key] and key not in ["transformers_version"]:
                    # If specified in `vision_config_dict`
                    if key in vision_config_dict:
                        message = (
                            f"`{key}` is found in both `vision_config_dict` and `vision_config` but with different "
                            f'values. The value `vision_config_dict["{key}"]` will be used instead.'
                        )
                    # If inferred from default argument values (just to be super careful)
                    else:
                        message = (
                            f"`vision_config_dict` is provided which will be used to initialize `CLIPVisionConfig`. "
                            f'The value `vision_config["{key}"]` will be overriden.'
                        )
                    logger.warning(message)

            # Update all values in `vision_config` with the ones in `_vision_config_dict`.
            vision_config.update(_vision_config_dict)

        if text_config is None:
            text_config = {}
            logger.info("`text_config` is `None`. Initializing the `CLIPTextConfig` with default values.")

        if vision_config is None:
            vision_config = {}
            logger.info("`vision_config` is `None`. initializing the `CLIPVisionConfig` with default values.")

        self.text_config = CLIPTextConfig(**text_config)
        self.vision_config = CLIPVisionConfig(**vision_config)

        self.projection_dim = projection_dim
        self.logit_scale_init_value = logit_scale_init_value
        self.initializer_factor = 1.0

    @classmethod
    def from_text_vision_configs(cls, text_config: CLIPTextConfig, vision_config: CLIPVisionConfig, **kwargs):
        r"""
        Instantiate a [`CLIPConfig`] (or a derived class) from clip text model configuration and clip vision model
        configuration.

        Returns:
            [`CLIPConfig`]: An instance of a configuration object
        """

        return cls(text_config=text_config.to_dict(), vision_config=vision_config.to_dict(), **kwargs)


class CLIPOnnxConfig(OnnxConfig):
    @property
    def inputs(self) -> Mapping[str, Mapping[int, str]]:
        return OrderedDict(
            [
                ("input_ids", {0: "batch", 1: "sequence"}),
                ("pixel_values", {0: "batch", 1: "num_channels", 2: "height", 3: "width"}),
                ("attention_mask", {0: "batch", 1: "sequence"}),
            ]
        )

    @property
    def outputs(self) -> Mapping[str, Mapping[int, str]]:
        return OrderedDict(
            [
                ("logits_per_image", {0: "batch"}),
                ("logits_per_text", {0: "batch"}),
                ("text_embeds", {0: "batch"}),
                ("image_embeds", {0: "batch"}),
            ]
        )

    @property
    def atol_for_validation(self) -> float:
        return 1e-4

    def generate_dummy_inputs(
        self,
        processor: "ProcessorMixin",
        batch_size: int = -1,
        seq_length: int = -1,
        framework: Optional["TensorType"] = None,
    ) -> Mapping[str, Any]:
        text_input_dict = super().generate_dummy_inputs(
            processor.tokenizer, batch_size=batch_size, seq_length=seq_length, framework=framework
        )
        image_input_dict = super().generate_dummy_inputs(
            processor.image_processor, batch_size=batch_size, framework=framework
        )
        return {**text_input_dict, **image_input_dict}

    @property
    def default_onnx_opset(self) -> int:
        return 14