File size: 14,768 Bytes
9505bbc
 
 
 
 
 
 
 
 
 
 
 
 
 
b1b2476
9505bbc
 
 
 
 
 
 
b1b2476
9505bbc
 
 
b1b2476
9505bbc
 
b1b2476
9505bbc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b1b2476
9505bbc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b1b2476
9505bbc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b1b2476
9505bbc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b1b2476
9505bbc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b1b2476
9505bbc
b1b2476
 
9505bbc
 
 
b1b2476
9505bbc
b1b2476
 
9505bbc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# coding=utf-8
# Copyright 2023 Mistral AI and 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.
""" VMistral model configuration"""
from transformers.configuration_utils import PretrainedConfig
from transformers.utils import logging


logger = logging.get_logger(__name__)

MISTRAL_PRETRAINED_CONFIG_ARCHIVE_MAP = {
    "HuggingFaceM4/VLM_WebSight_finetuned": "https://huggingface.co/HuggingFaceM4/VLM_WebSight_finetuned/resolve/main/config.json",
}


class VMistralVisionConfig(PretrainedConfig):
    r"""
    """
    model_type = "vmistral"

    def __init__(
        self,
        hidden_size=768,
        intermediate_size=3072,
        num_hidden_layers=12,
        num_attention_heads=12,
        num_channels=3,
        image_size=224,
        patch_size=32,
        hidden_act="gelu_pytorch_tanh",
        layer_norm_eps=1e-6,
        attention_dropout=0.0,
        initializer_range=0.02,
        initializer_factor=1.0,
        _flash_attn_2_enabled=True,
        **kwargs,
    ):
        super().__init__(**kwargs)

        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        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
        self._flash_attn_2_enabled = _flash_attn_2_enabled


class VMistralPerceiverConfig(PretrainedConfig):
    r"""
    TThis is the configuration class to store the configuration of a [`MistralModel`]. It is used to instantiate an
    Mistral model according to the specified arguments, defining the model architecture. Instantiating a configuration
    with the defaults will yield a similar configuration to that of the Mistral-7B-v0.1 or Mistral-7B-Instruct-v0.1.

    [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1)
    [mistralai/Mistral-7B-Instruct-v0.1](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1)

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

    Args:
        use_resampler (`bool`, *optional*, defaults to `False`):
            Whether or not to use the resampler
        resampler_n_latents (`int`, *optional*, defaults to ):
            Number of latent embeddings to resample ("compress") the input sequence to (usually < 128).
        resampler_depth (`int`, *optional*, defaults to 6):
            Depth of the Perceiver Resampler (Transformer w/ cross attention). Should be shallow (< 3).
        resampler_n_heads (`int`, *optional*, defaults to 16):
            Number of heads in each Transformer block (for multi-headed self-attention).
        resampler_head_dim (`int`, *optional*, defaults to 96):
            Dimensionality of each head projection in the Transformer block.
        qk_layer_norms_perceiver (`bool`, *optional*, defaults to `False`):
            Whether or not to use qk layer norms in perceiver
    """
    model_type = "vmistral"

    def __init__(
        self,
        resampler_n_latents=64,
        resampler_depth=6,
        resampler_n_heads=16,
        resampler_head_dim=96,
        qk_layer_norms_perceiver=False,
        **kwargs,
    ):
        self.resampler_n_latents = resampler_n_latents
        self.resampler_depth = resampler_depth
        self.resampler_n_heads = resampler_n_heads
        self.resampler_head_dim = resampler_head_dim
        self.qk_layer_norms_perceiver = qk_layer_norms_perceiver

        super().__init__(**kwargs)


class VMistralConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`MistralModel`]. It is used to instantiate an
    Mistral model according to the specified arguments, defining the model architecture. Instantiating a configuration
    with the defaults will yield a similar configuration to that of the Mistral-7B-v0.1 or Mistral-7B-Instruct-v0.1.

    [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1)
    [mistralai/Mistral-7B-Instruct-v0.1](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1)

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

    Args:
        additional_vocab_size (`int`, *optional`, defaults to 0):
            Additional vocabulary size of the model, typically for the special "<img>" token. Additional vocab tokens
            are always trainable whereas regular vocab tokens can be frozen or not.
        vocab_size (`int`, *optional*, defaults to 32000):
            Vocabulary size of the Mistral model. Defines the number of different tokens that can be represented by the
            `inputs_ids` passed when calling [`MistralModel`]
        hidden_size (`int`, *optional*, defaults to 4096):
            Dimension of the hidden representations.
        intermediate_size (`int`, *optional*, defaults to 14336):
            Dimension of the MLP representations.
        num_hidden_layers (`int`, *optional*, defaults to 32):
            Number of hidden layers in the Transformer encoder.
        num_attention_heads (`int`, *optional*, defaults to 32):
            Number of attention heads for each attention layer in the Transformer encoder.
        num_key_value_heads (`int`, *optional*, defaults to 8):
            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
            `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
            by meanpooling all the original heads within that group. For more details checkout [this
            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to `8`.
        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
            The non-linear activation function (function or string) in the decoder.
        max_position_embeddings (`int`, *optional*, defaults to `4096*32`):
            The maximum sequence length that this model might ever be used with. Mistral's sliding window attention
            allows sequence of up to 4096*32 tokens.
        initializer_range (`float`, *optional*, defaults to 0.02):
            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
        alpha_initializer (`str`, *optional*, defaults to `"zeros"`):
            Initialization type for the alphas.
        alphas_initializer_range (`float`, *optional*, defaults to 0.0):
            The standard deviation of the truncated_normal_initializer for initializing the alphas in the Gated Cross
            Attention.
        alpha_type (`str`, *optional*, defaults to `"float"`):
            Whether the gating alphas should be vectors or single floats.
        rms_norm_eps (`float`, *optional*, defaults to 1e-06):
            The epsilon used by the rms normalization layers.
        use_cache (`bool`, *optional*, defaults to `True`):
            Whether or not the model should return the last key/values attentions (not used by all models). Only
            relevant if `config.is_decoder=True`.
        pad_token_id (`int`, *optional*):
            The id of the padding token.
        bos_token_id (`int`, *optional*, defaults to 1):
            The id of the "beginning-of-sequence" token.
        eos_token_id (`int`, *optional*, defaults to 2):
            The id of the "end-of-sequence" token.
        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
            Whether the model's input and output word embeddings should be tied.
        rope_theta (`float`, *optional*, defaults to 10000.0):
            The base period of the RoPE embeddings.
        sliding_window (`int`, *optional*, defaults to 4096):
            Sliding window attention window size. If not specified, will default to `4096`.
        cross_layer_interval (`int`, *optional*, default to 1)
            Interval for cross attention (from text to image) layers.
        qk_layer_norms (`bool`, *optional*, defaults to `False`): Whether to add layer norm after q and k
        freeze_text_layers (`bool`, *optional*, defaults to `True`): Whether to freeze text layers
        freeze_text_module_exceptions (`bool`, *optional*, defaults to `[]`):
            Exceptions to freezing text layers when `freeze_text_layers` is `True`
        freeze_lm_head (`bool`, *optional*, defaults to `False`): Whether to freeze lm head
        freeze_vision_layers (`bool`, *optional*, defaults to `True`):  Whether to freeze vision layers
        freeze_vision_module_exceptions (`bool`, *optional*, defaults to `[]`):
            Exceptions to freezing vision layers when `freeze_vision_layers` is `True`
        use_resampler (`bool`, *optional*, defaults to `False`): Whether to use the Resampler
        vision_config (`IdeficsVisionConfig`,  *optional*): Custom vision config or dict
        perceiver_config (`IdeficsPerceiverConfig`,  *optional*): Custom perceiver config or dict

    Example:
    ```python
    >>> from transformers import MistralModel, MistralConfig

    >>> # Initializing a Mistral 7B style configuration
    >>> configuration = MistralConfig()

    >>> # Initializing a model from the Mistral 7B style configuration
    >>> model = MistralModel(configuration)

    >>> # Accessing the model configuration
    >>> configuration = model.config
    ```"""
    model_type = "vmistral"
    is_composition = False

    def __init__(
        self,
        additional_vocab_size=0,
        vocab_size=32000,
        hidden_size=4096,
        intermediate_size=14336,
        num_hidden_layers=32,
        num_attention_heads=32,
        num_key_value_heads=8,
        hidden_act="silu",
        max_position_embeddings=4096 * 32,
        initializer_range=0.02,
        alpha_initializer="zeros",
        alphas_initializer_range=0.0,
        alpha_type="float",
        rms_norm_eps=1e-6,
        use_cache=True,
        pad_token_id=0,  # None in the original configuration_mistral, we set it to the unk_token_id
        bos_token_id=1,
        eos_token_id=2,
        image_token_id=32_001,
        tie_word_embeddings=False,
        rope_theta=10000.0,
        sliding_window=4096,
        cross_layer_interval=1,
        qk_layer_norms=False,
        freeze_text_layers=True,
        freeze_text_module_exceptions=[],
        freeze_lm_head=False,
        freeze_vision_layers=True,
        freeze_vision_module_exceptions=[],
        attention_dropout=0.0,
        _flash_attn_2_enabled=True,
        use_resampler=False,
        vision_config=None,
        perceiver_config=None,
        **kwargs,
    ):
        self.vocab_size = vocab_size
        self.additional_vocab_size = additional_vocab_size
        self.image_token_id = image_token_id
        self.max_position_embeddings = max_position_embeddings
        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.sliding_window = sliding_window

        # for backward compatibility
        if num_key_value_heads is None:
            num_key_value_heads = num_attention_heads

        self.num_key_value_heads = num_key_value_heads
        self.hidden_act = hidden_act
        self.initializer_range = initializer_range
        self.alpha_initializer = alpha_initializer
        self.alphas_initializer_range = alphas_initializer_range
        self.alpha_type = alpha_type
        self.rms_norm_eps = rms_norm_eps
        self.use_cache = use_cache
        self.rope_theta = rope_theta

        self.cross_layer_interval = cross_layer_interval
        self.qk_layer_norms = qk_layer_norms
        self.freeze_vision_layers = freeze_vision_layers

        self.freeze_text_layers = freeze_text_layers
        self.freeze_text_module_exceptions = freeze_text_module_exceptions
        self.freeze_vision_module_exceptions = freeze_vision_module_exceptions
        self.freeze_lm_head = freeze_lm_head

        self.use_resampler = use_resampler
        self._flash_attn_2_enabled = _flash_attn_2_enabled
        self.attention_dropout = attention_dropout

        if perceiver_config is None:
            self.perceiver_config = VMistralPerceiverConfig()
        elif isinstance(perceiver_config, dict):
            self.perceiver_config = VMistralPerceiverConfig(**perceiver_config)
        elif isinstance(perceiver_config, VMistralPerceiverConfig):
            self.perceiver_config = perceiver_config

        if vision_config is None:
            self.vision_config = VMistralVisionConfig()
        elif isinstance(vision_config, dict):
            self.vision_config = VMistralVisionConfig(**vision_config)
        elif isinstance(vision_config, VMistralVisionConfig):
            self.vision_config = vision_config

        super().__init__(
            pad_token_id=pad_token_id,
            bos_token_id=bos_token_id,
            eos_token_id=eos_token_id,
            tie_word_embeddings=tie_word_embeddings,
            **kwargs,
        )

        # IMPORTANT: Do not do any __init__ args-based checks in the constructor, since
        # PretrainedConfig.from_dict first instantiates the class with the config dict and only then
        # updates the config object with `kwargs` from from_pretrained, so during the instantiation
        # of this object many attributes have default values and haven't yet been overridden.
        # Do any required checks inside `from_pretrained` once the superclass' `from_pretrained` was run.