Skip to content

unitorch.cli.models.clip¤

ClipProcessor¤

Tip

core/process/clip is the section for configuration of ClipProcessor.

Bases: ClipProcessor

Processor for the CLIP model.

Initialize the ClipProcessor.

Parameters:

Name Type Description Default
vocab_path str

The path to the vocabulary file.

 required
merge_path str

The path to the BPE merge file.

 required
vision_config_path str

The path to the vision configuration file.

 required
max_seq_length int

The maximum sequence length. Defaults to 128.

 128
position_start_id int

The position start ID. Defaults to 0.

 0
Source code in src/unitorch/cli/models/clip/processing.py 
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
def __init__(
    self,
    vocab_path: str,
    merge_path: str,
    vision_config_path: str,
    max_seq_length: Optional[int] = 128,
    position_start_id: Optional[int] = 0,
):
    """
    Initialize the ClipProcessor.

    Args:
        vocab_path (str): The path to the vocabulary file.
        merge_path (str): The path to the BPE merge file.
        vision_config_path (str): The path to the vision configuration file.
        max_seq_length (int, optional): The maximum sequence length. Defaults to 128.
        position_start_id (int, optional): The position start ID. Defaults to 0.
    """
    super().__init__(
        vocab_path=vocab_path,
        merge_path=merge_path,
        vision_config_path=vision_config_path,
        max_seq_length=max_seq_length,
        position_start_id=position_start_id,
    )

from_core_configure classmethod ¤

from_core_configure(config, **kwargs)

Create an instance of ClipProcessor from a core configuration.

Parameters:

Name Type Description Default
config

The core configuration.

 required
**kwargs

Additional keyword arguments.

 {}

Returns:

Name Type Description
dict

A dictionary containing the processor's initialization arguments.
Source code in src/unitorch/cli/models/clip/processing.py 
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
@classmethod
@add_default_section_for_init("core/process/clip")
def from_core_configure(cls, config, **kwargs):
    """
    Create an instance of ClipProcessor from a core configuration.

    Args:
        config: The core configuration.
        **kwargs: Additional keyword arguments.

    Returns:
        dict: A dictionary containing the processor's initialization arguments.
    """
    config.set_default_section("core/process/clip")
    pretrained_name = config.getoption("pretrained_name", "clip-vit-base-patch16")
    vocab_path = config.getoption("vocab_path", None)
    vocab_path = pop_value(
        vocab_path,
        nested_dict_value(pretrained_clip_infos, pretrained_name, "vocab"),
    )
    vocab_path = cached_path(vocab_path)

    merge_path = config.getoption("merge_path", None)
    merge_path = pop_value(
        merge_path,
        nested_dict_value(pretrained_clip_infos, pretrained_name, "merge"),
    )
    merge_path = cached_path(merge_path)

    vision_config_path = config.getoption("vision_config_path", None)
    vision_config_path = pop_value(
        vision_config_path,
        nested_dict_value(pretrained_clip_infos, pretrained_name, "vision_config"),
    )

    vision_config_path = cached_path(vision_config_path)

    return {
        "vocab_path": vocab_path,
        "merge_path": merge_path,
        "vision_config_path": vision_config_path,
    }

ClipForPretrain¤

Tip

core/model/pretrain/clip is the section for configuration of ClipForPretrain.

Bases: ClipForPretrain

CLIP model for pretraining.

Initialize the ClipForPretrain model.

Parameters:

Name Type Description Default
config_path str

The path to the model configuration file.

 required
projection_dim int

The dimension of the projection head. Defaults to 512.

 512
freeze_base_model bool

Whether to freeze the base model. Defaults to True.

 True
gradient_checkpointing bool

Whether to use gradient checkpointing. Defaults to False.

 False
use_all_gather bool

Whether to use all_gather operation. Defaults to True.

 True
Source code in src/unitorch/cli/models/clip/modeling.py 
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
def __init__(
    self,
    config_path: str,
    projection_dim: Optional[int] = 512,
    freeze_base_model: Optional[bool] = True,
    gradient_checkpointing: Optional[bool] = False,
    use_all_gather: Optional[bool] = True,
):
    """
    Initialize the ClipForPretrain model.

    Args:
        config_path (str): The path to the model configuration file.
        projection_dim (int, optional): The dimension of the projection head. Defaults to 512.
        freeze_base_model (bool, optional): Whether to freeze the base model. Defaults to True.
        gradient_checkpointing (bool, optional): Whether to use gradient checkpointing. Defaults to False.
        use_all_gather (bool, optional): Whether to use all_gather operation. Defaults to True.
    """
    super().__init__(
        config_path=config_path,
        projection_dim=projection_dim,
        freeze_base_model=freeze_base_model,
        gradient_checkpointing=gradient_checkpointing,
        use_all_gather=use_all_gather,
    )

forward ¤

forward(
    input_ids: Tensor,
    pixel_values: Tensor,
    attention_mask: Optional[Tensor] = None,
    position_ids: Optional[Tensor] = None,
)

Perform a forward pass through the model.

Parameters:

Name Type Description Default
input_ids Tensor

Input token IDs.

 required
pixel_values Tensor

Input pixel values.

 required
attention_mask Tensor

Attention mask. Defaults to None.

 None
position_ids Tensor

Position IDs. Defaults to None.

 None

Returns:

Name Type Description
LossOutputs

The loss outputs.
Source code in src/unitorch/cli/models/clip/modeling.py 
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
@autocast(device_type=("cuda" if torch.cuda.is_available() else "cpu"))
def forward(
    self,
    input_ids: torch.Tensor,
    pixel_values: torch.Tensor,
    attention_mask: Optional[torch.Tensor] = None,
    position_ids: Optional[torch.Tensor] = None,
):
    """
    Perform a forward pass through the model.

    Args:
        input_ids (torch.Tensor): Input token IDs.
        pixel_values (torch.Tensor): Input pixel values.
        attention_mask (torch.Tensor, optional): Attention mask. Defaults to None.
        position_ids (torch.Tensor, optional): Position IDs. Defaults to None.

    Returns:
        LossOutputs: The loss outputs.
    """
    outputs = super().forward(
        input_ids=input_ids,
        pixel_values=pixel_values,
        attention_mask=attention_mask,
        position_ids=position_ids,
    )
    return LossOutputs(loss=outputs)

from_core_configure classmethod ¤

from_core_configure(config, **kwargs)

Create an instance of ClipForPretrain from a core configuration.

Parameters:

Name Type Description Default
config

The core configuration.

 required
**kwargs

Additional keyword arguments.

 {}

Returns:

Name Type Description
ClipForPretrain

An instance of the ClipForPretrain model.
Source code in src/unitorch/cli/models/clip/modeling.py 
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
@classmethod
@add_default_section_for_init("core/model/pretrain/clip")
def from_core_configure(cls, config, **kwargs):
    """
    Create an instance of ClipForPretrain from a core configuration.

    Args:
        config: The core configuration.
        **kwargs: Additional keyword arguments.

    Returns:
        ClipForPretrain: An instance of the ClipForPretrain model.
    """
    config.set_default_section("core/model/pretrain/clip")
    pretrained_name = config.getoption("pretrained_name", "clip-vit-base-patch16")
    config_path = config.getoption("config_path", None)
    config_path = pop_value(
        config_path,
        nested_dict_value(pretrained_clip_infos, pretrained_name, "config"),
    )

    config_path = cached_path(config_path)

    projection_dim = config.getoption("projection_dim", 512)
    freeze_base_model = config.getoption("freeze_base_model", True)
    gradient_checkpointing = config.getoption("gradient_checkpointing", False)
    use_all_gather = config.getoption("use_all_gather", True)

    inst = cls(
        config_path=config_path,
        projection_dim=projection_dim,
        freeze_base_model=freeze_base_model,
        gradient_checkpointing=gradient_checkpointing,
        use_all_gather=use_all_gather,
    )
    pretrained_weight_path = config.getoption("pretrained_weight_path", None)
    weight_path = pop_value(
        pretrained_weight_path,
        nested_dict_value(pretrained_clip_infos, pretrained_name, "weight"),
        check_none=False,
    )
    if weight_path is not None:
        inst.from_pretrained(weight_path)

    return inst

ClipForClassification¤

Tip

core/model/classification/clip is the section for configuration of ClipForClassification.

Bases: ClipForClassification

CLIP model for classification.

Initialize the ClipForClassification model.

Parameters:

Name Type Description Default
config_path str

The path to the model configuration file.

 required
projection_dim int

The dimension of the projection head. Defaults to 512.

 512
num_classes int

The number of output classes. Defaults to 1.

 1
freeze_base_model bool

Whether to freeze the base model. Defaults to True.

 True
gradient_checkpointing bool

Whether to use gradient checkpointing. Defaults to False.

 False
Source code in src/unitorch/cli/models/clip/modeling.py 
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
def __init__(
    self,
    config_path: str,
    projection_dim: Optional[int] = 512,
    num_classes: Optional[int] = 1,
    freeze_base_model: Optional[bool] = True,
    gradient_checkpointing: Optional[bool] = False,
):
    """
    Initialize the ClipForClassification model.

    Args:
        config_path (str): The path to the model configuration file.
        projection_dim (int, optional): The dimension of the projection head. Defaults to 512.
        num_classes (int, optional): The number of output classes. Defaults to 1.
        freeze_base_model (bool, optional): Whether to freeze the base model. Defaults to True.
        gradient_checkpointing (bool, optional): Whether to use gradient checkpointing. Defaults to False.
    """
    super().__init__(
        config_path=config_path,
        projection_dim=projection_dim,
        num_classes=num_classes,
        freeze_base_model=freeze_base_model,
        gradient_checkpointing=gradient_checkpointing,
    )

forward ¤

forward(
    input_ids: Tensor,
    pixel_values: Tensor,
    attention_mask: Optional[Tensor] = None,
    position_ids: Optional[Tensor] = None,
)

Perform a forward pass through the model.

Parameters:

Name Type Description Default
input_ids Tensor

Input token IDs.

 required
pixel_values Tensor

Input pixel values.

 required
attention_mask Tensor

Attention mask. Defaults to None.

 None
position_ids Tensor

Position IDs. Defaults to None.

 None

Returns:

Name Type Description
ClassificationOutputs

The classification outputs.
Source code in src/unitorch/cli/models/clip/modeling.py 
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
@autocast(device_type=("cuda" if torch.cuda.is_available() else "cpu"))
def forward(
    self,
    input_ids: torch.Tensor,
    pixel_values: torch.Tensor,
    attention_mask: Optional[torch.Tensor] = None,
    position_ids: Optional[torch.Tensor] = None,
):
    """
    Perform a forward pass through the model.

    Args:
        input_ids (torch.Tensor): Input token IDs.
        pixel_values (torch.Tensor): Input pixel values.
        attention_mask (torch.Tensor, optional): Attention mask. Defaults to None.
        position_ids (torch.Tensor, optional): Position IDs. Defaults to None.

    Returns:
        ClassificationOutputs: The classification outputs.
    """
    outputs = super().forward(
        input_ids=input_ids,
        pixel_values=pixel_values,
        attention_mask=attention_mask,
        position_ids=position_ids,
    )
    return ClassificationOutputs(outputs=outputs)

from_core_configure classmethod ¤

from_core_configure(config, **kwargs)

Create an instance of ClipForClassification from a core configuration.

Parameters:

Name Type Description Default
config

The core configuration.

 required
**kwargs

Additional keyword arguments.

 {}

Returns:

Name Type Description
ClipForClassification

An instance of the ClipForClassification model.
Source code in src/unitorch/cli/models/clip/modeling.py 
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
@classmethod
@add_default_section_for_init("core/model/classification/clip")
def from_core_configure(cls, config, **kwargs):
    """
    Create an instance of ClipForClassification from a core configuration.

    Args:
        config: The core configuration.
        **kwargs: Additional keyword arguments.

    Returns:
        ClipForClassification: An instance of the ClipForClassification model.
    """
    config.set_default_section("core/model/classification/clip")
    pretrained_name = config.getoption("pretrained_name", "clip-vit-base-patch16")
    config_path = config.getoption("config_path", None)
    config_path = pop_value(
        config_path,
        nested_dict_value(pretrained_clip_infos, pretrained_name, "config"),
    )

    config_path = cached_path(config_path)

    projection_dim = config.getoption("projection_dim", 512)
    num_classes = config.getoption("num_classes", 1)
    freeze_base_model = config.getoption("freeze_base_model", True)
    gradient_checkpointing = config.getoption("gradient_checkpointing", False)

    inst = cls(
        config_path=config_path,
        projection_dim=projection_dim,
        num_classes=num_classes,
        freeze_base_model=freeze_base_model,
        gradient_checkpointing=gradient_checkpointing,
    )
    pretrained_weight_path = config.getoption("pretrained_weight_path", None)
    weight_path = pop_value(
        pretrained_weight_path,
        nested_dict_value(pretrained_clip_infos, pretrained_name, "weight"),
        check_none=False,
    )
    if weight_path is not None:
        inst.from_pretrained(weight_path)

    return inst

ClipForTextClassification¤

Tip

core/model/classification/clip/text is the section for configuration of ClipForTextClassification.

Bases: ClipForTextClassification

CLIP model for text classification.

Initialize the ClipForTextClassification model.

Parameters:

Name Type Description Default
config_path str

The path to the model configuration file.

 required
projection_dim int

The dimension of the projection head. Defaults to 512.

 512
num_classes int

The number of output classes. Defaults to 1.

 1
freeze_base_model bool

Whether to freeze the base model. Defaults to True.

 True
gradient_checkpointing bool

Whether to use gradient checkpointing. Defaults to False.

 False
Source code in src/unitorch/cli/models/clip/modeling.py 
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
def __init__(
    self,
    config_path: str,
    projection_dim: Optional[int] = 512,
    num_classes: Optional[int] = 1,
    freeze_base_model: Optional[bool] = True,
    gradient_checkpointing: Optional[bool] = False,
):
    """
    Initialize the ClipForTextClassification model.

    Args:
        config_path (str): The path to the model configuration file.
        projection_dim (int, optional): The dimension of the projection head. Defaults to 512.
        num_classes (int, optional): The number of output classes. Defaults to 1.
        freeze_base_model (bool, optional): Whether to freeze the base model. Defaults to True.
        gradient_checkpointing (bool, optional): Whether to use gradient checkpointing. Defaults to False.
    """
    super().__init__(
        config_path=config_path,
        projection_dim=projection_dim,
        num_classes=num_classes,
        freeze_base_model=freeze_base_model,
        gradient_checkpointing=gradient_checkpointing,
    )

forward ¤

forward(
    input_ids=None,
    attention_mask: Optional[Tensor] = None,
    position_ids: Optional[Tensor] = None,
)

Perform a forward pass through the model.

Parameters:

Name Type Description Default
input_ids Tensor

Input token IDs. Defaults to None.

 None
attention_mask Tensor

Attention mask. Defaults to None.

 None
position_ids Tensor

Position IDs. Defaults to None.

 None

Returns:

Name Type Description
ClassificationOutputs

The classification outputs.
Source code in src/unitorch/cli/models/clip/modeling.py 
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
@autocast(device_type=("cuda" if torch.cuda.is_available() else "cpu"))
def forward(
    self,
    input_ids=None,
    attention_mask: Optional[torch.Tensor] = None,
    position_ids: Optional[torch.Tensor] = None,
):
    """
    Perform a forward pass through the model.

    Args:
        input_ids (torch.Tensor, optional): Input token IDs. Defaults to None.
        attention_mask (torch.Tensor, optional): Attention mask. Defaults to None.
        position_ids (torch.Tensor, optional): Position IDs. Defaults to None.

    Returns:
        ClassificationOutputs: The classification outputs.
    """
    outputs = super().forward(
        input_ids=input_ids,
        attention_mask=attention_mask,
        position_ids=position_ids,
    )
    return ClassificationOutputs(outputs=outputs)

from_core_configure classmethod ¤

from_core_configure(config, **kwargs)

Create an instance of ClipForTextClassification from a core configuration.

Parameters:

Name Type Description Default
config

The core configuration.

 required
**kwargs

Additional keyword arguments.

 {}

Returns:

Name Type Description
ClipForTextClassification

An instance of the ClipForTextClassification model.
Source code in src/unitorch/cli/models/clip/modeling.py 
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
@classmethod
@add_default_section_for_init("core/model/classification/clip/text")
def from_core_configure(cls, config, **kwargs):
    """
    Create an instance of ClipForTextClassification from a core configuration.

    Args:
        config: The core configuration.
        **kwargs: Additional keyword arguments.

    Returns:
        ClipForTextClassification: An instance of the ClipForTextClassification model.
    """
    config.set_default_section("core/model/classification/clip/text")
    pretrained_name = config.getoption("pretrained_name", "clip-vit-base-patch16")
    config_path = config.getoption("config_path", None)
    config_path = pop_value(
        config_path,
        nested_dict_value(pretrained_clip_infos, pretrained_name, "config"),
    )

    config_path = cached_path(config_path)

    projection_dim = config.getoption("projection_dim", 512)
    num_classes = config.getoption("num_classes", 1)
    freeze_base_model = config.getoption("freeze_base_Truemodel", True)
    gradient_checkpointing = config.getoption("gradient_checkpointing", False)

    inst = cls(
        config_path=config_path,
        projection_dim=projection_dim,
        num_classes=num_classes,
        freeze_base_model=freeze_base_model,
        gradient_checkpointing=gradient_checkpointing,
    )
    pretrained_weight_path = config.getoption("pretrained_weight_path", None)
    weight_path = pop_value(
        pretrained_weight_path,
        nested_dict_value(pretrained_clip_infos, pretrained_name, "weight"),
        check_none=False,
    )
    if weight_path is not None:
        inst.from_pretrained(weight_path)

    return inst

ClipForImageClassification¤

Tip

core/model/classification/clip/image is the section for configuration of ClipForImageClassification.

Bases: ClipForImageClassification

CLIP model for image classification.

Initialize the ClipForImageClassification model.

Parameters:

Name Type Description Default
config_path str

The path to the model configuration file.

 required
projection_dim int

The dimension of the projection head. Defaults to 512.

 512
num_classes int

The number of output classes. Defaults to 1.

 1
freeze_base_model bool

Whether to freeze the base model. Defaults to True.

 True
gradient_checkpointing bool

Whether to use gradient checkpointing. Defaults to False.

 False
Source code in src/unitorch/cli/models/clip/modeling.py 
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
def __init__(
    self,
    config_path: str,
    projection_dim: Optional[int] = 512,
    num_classes: Optional[int] = 1,
    freeze_base_model: Optional[bool] = True,
    gradient_checkpointing: Optional[bool] = False,
):
    """
    Initialize the ClipForImageClassification model.

    Args:
        config_path (str): The path to the model configuration file.
        projection_dim (int, optional): The dimension of the projection head. Defaults to 512.
        num_classes (int, optional): The number of output classes. Defaults to 1.
        freeze_base_model (bool, optional): Whether to freeze the base model. Defaults to True.
        gradient_checkpointing (bool, optional): Whether to use gradient checkpointing. Defaults to False.
    """
    super().__init__(
        config_path=config_path,
        projection_dim=projection_dim,
        num_classes=num_classes,
        freeze_base_model=freeze_base_model,
        gradient_checkpointing=gradient_checkpointing,
    )

forward ¤

forward(pixel_values: Tensor)

Perform a forward pass through the model.

Parameters:

Name Type Description Default
pixel_values Tensor

Input pixel values.

 required

Returns:

Name Type Description
ClassificationOutputs

The classification outputs.
Source code in src/unitorch/cli/models/clip/modeling.py 
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
@autocast(device_type=("cuda" if torch.cuda.is_available() else "cpu"))
def forward(
    self,
    pixel_values: torch.Tensor,
):
    """
    Perform a forward pass through the model.

    Args:
        pixel_values (torch.Tensor): Input pixel values.

    Returns:
        ClassificationOutputs: The classification outputs.
    """
    outputs = super().forward(pixel_values=pixel_values)
    return ClassificationOutputs(outputs=outputs)

from_core_configure classmethod ¤

from_core_configure(config, **kwargs)

Create an instance of ClipForImageClassification from a core configuration.

Parameters:

Name Type Description Default
config

The core configuration.

 required
**kwargs

Additional keyword arguments.

 {}

Returns:

Name Type Description
ClipForImageClassification

An instance of the ClipForImageClassification model.
Source code in src/unitorch/cli/models/clip/modeling.py 
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
@classmethod
@add_default_section_for_init("core/model/classification/clip/image")
def from_core_configure(cls, config, **kwargs):
    """
    Create an instance of ClipForImageClassification from a core configuration.

    Args:
        config: The core configuration.
        **kwargs: Additional keyword arguments.

    Returns:
        ClipForImageClassification: An instance of the ClipForImageClassification model.
    """
    config.set_default_section("core/model/classification/clip/image")
    pretrained_name = config.getoption("pretrained_name", "clip-vit-base-patch16")
    config_path = config.getoption("config_path", None)
    config_path = pop_value(
        config_path,
        nested_dict_value(pretrained_clip_infos, pretrained_name, "config"),
    )

    config_path = cached_path(config_path)

    projection_dim = config.getoption("projection_dim", 512)
    num_classes = config.getoption("num_classes", 1)
    freeze_base_model = config.getoption("freeze_base_model", True)
    gradient_checkpointing = config.getoption("gradient_checkpointing", False)

    inst = cls(
        config_path=config_path,
        projection_dim=projection_dim,
        num_classes=num_classes,
        freeze_base_model=freeze_base_model,
        gradient_checkpointing=gradient_checkpointing,
    )
    pretrained_weight_path = config.getoption("pretrained_weight_path", None)
    weight_path = pop_value(
        pretrained_weight_path,
        nested_dict_value(pretrained_clip_infos, pretrained_name, "weight"),
        check_none=False,
    )
    if weight_path is not None:
        inst.from_pretrained(weight_path)

    return inst