File size: 9,007 Bytes
2824520
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Copyright 2024 Salesforce.com, inc.
# Copyright 2024 The HuggingFace 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.
from typing import Optional, Tuple, Union

import torch
from torch import nn
from transformers import CLIPPreTrainedModel
from transformers.modeling_outputs import BaseModelOutputWithPooling
from transformers.models.clip.configuration_clip import CLIPTextConfig
from transformers.models.clip.modeling_clip import CLIPEncoder


def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
    """
    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
    """
    bsz, src_len = mask.size()
    tgt_len = tgt_len if tgt_len is not None else src_len

    expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)

    inverted_mask = 1.0 - expanded_mask

    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)


# This is a modified version of the CLIPTextModel from transformers.models.clip.modeling_clip
# Which allows for an extra input of "context embeddings", which are the query embeddings used in Qformer
# They pass through the clip model, along with the text embeddings, and interact with them using self attention
class ContextCLIPTextModel(CLIPPreTrainedModel):
    config_class = CLIPTextConfig

    _no_split_modules = ["CLIPEncoderLayer"]

    def __init__(self, config: CLIPTextConfig):
        super().__init__(config)
        self.text_model = ContextCLIPTextTransformer(config)
        # Initialize weights and apply final processing
        self.post_init()

    def forward(
        self,
        ctx_embeddings: torch.Tensor = None,
        ctx_begin_pos: list = None,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, BaseModelOutputWithPooling]:
        return self.text_model(
            ctx_embeddings=ctx_embeddings,
            ctx_begin_pos=ctx_begin_pos,
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )


class ContextCLIPTextTransformer(nn.Module):
    def __init__(self, config: CLIPTextConfig):
        super().__init__()
        self.config = config
        embed_dim = config.hidden_size
        self.embeddings = ContextCLIPTextEmbeddings(config)
        self.encoder = CLIPEncoder(config)
        self.final_layer_norm = nn.LayerNorm(embed_dim)

    def forward(
        self,
        ctx_embeddings: torch.Tensor,
        ctx_begin_pos: list,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple, BaseModelOutputWithPooling]:
        r"""
        Returns:

        """
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        if input_ids is None:
            raise ValueError("You have to specify either input_ids")

        input_shape = input_ids.size()
        input_ids = input_ids.view(-1, input_shape[-1])

        hidden_states = self.embeddings(
            input_ids=input_ids,
            position_ids=position_ids,
            ctx_embeddings=ctx_embeddings,
            ctx_begin_pos=ctx_begin_pos,
        )

        bsz, seq_len = input_shape
        if ctx_embeddings is not None:
            seq_len += ctx_embeddings.size(1)
        # CLIP's text model uses causal mask, prepare it here.
        # https://github.com/openai/CLIP/blob/cfcffb90e69f37bf2ff1e988237a0fbe41f33c04/clip/model.py#L324
        causal_attention_mask = self._build_causal_attention_mask(bsz, seq_len, hidden_states.dtype).to(
            hidden_states.device
        )
        # expand attention_mask
        if attention_mask is not None:
            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
            attention_mask = _expand_mask(attention_mask, hidden_states.dtype)

        encoder_outputs = self.encoder(
            inputs_embeds=hidden_states,
            attention_mask=attention_mask,
            causal_attention_mask=causal_attention_mask,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )

        last_hidden_state = encoder_outputs[0]
        last_hidden_state = self.final_layer_norm(last_hidden_state)

        # text_embeds.shape = [batch_size, sequence_length, transformer.width]
        # take features from the eot embedding (eot_token is the highest number in each sequence)
        # casting to torch.int for onnx compatibility: argmax doesn't support int64 inputs with opset 14
        pooled_output = last_hidden_state[
            torch.arange(last_hidden_state.shape[0], device=input_ids.device),
            input_ids.to(torch.int).argmax(dim=-1),
        ]

        if not return_dict:
            return (last_hidden_state, pooled_output) + encoder_outputs[1:]

        return BaseModelOutputWithPooling(
            last_hidden_state=last_hidden_state,
            pooler_output=pooled_output,
            hidden_states=encoder_outputs.hidden_states,
            attentions=encoder_outputs.attentions,
        )

    def _build_causal_attention_mask(self, bsz, seq_len, dtype):
        # lazily create causal attention mask, with full attention between the vision tokens
        # pytorch uses additive attention mask; fill with -inf
        mask = torch.empty(bsz, seq_len, seq_len, dtype=dtype)
        mask.fill_(torch.tensor(torch.finfo(dtype).min))
        mask.triu_(1)  # zero out the lower diagonal
        mask = mask.unsqueeze(1)  # expand mask
        return mask


class ContextCLIPTextEmbeddings(nn.Module):
    def __init__(self, config: CLIPTextConfig):
        super().__init__()
        embed_dim = config.hidden_size

        self.token_embedding = nn.Embedding(config.vocab_size, embed_dim)
        self.position_embedding = nn.Embedding(config.max_position_embeddings, embed_dim)

        # position_ids (1, len position emb) is contiguous in memory and exported when serialized
        self.register_buffer("position_ids", torch.arange(config.max_position_embeddings).expand((1, -1)))

    def forward(
        self,
        ctx_embeddings: torch.Tensor,
        ctx_begin_pos: list,
        input_ids: Optional[torch.LongTensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
    ) -> torch.Tensor:
        if ctx_embeddings is None:
            ctx_len = 0
        else:
            ctx_len = ctx_embeddings.shape[1]

        seq_length = (input_ids.shape[-1] if input_ids is not None else inputs_embeds.shape[-2]) + ctx_len

        if position_ids is None:
            position_ids = self.position_ids[:, :seq_length]

        if inputs_embeds is None:
            inputs_embeds = self.token_embedding(input_ids)

            # for each input embeddings, add the ctx embeddings at the correct position
            input_embeds_ctx = []
            bsz = inputs_embeds.shape[0]

            if ctx_embeddings is not None:
                for i in range(bsz):
                    cbp = ctx_begin_pos[i]

                    prefix = inputs_embeds[i, :cbp]
                    # remove the special token embedding
                    suffix = inputs_embeds[i, cbp:]

                    input_embeds_ctx.append(torch.cat([prefix, ctx_embeddings[i], suffix], dim=0))

                inputs_embeds = torch.stack(input_embeds_ctx, dim=0)

        position_embeddings = self.position_embedding(position_ids)
        embeddings = inputs_embeds + position_embeddings

        return embeddings