File size: 12,604 Bytes
2cd560a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# --------------------------------------------------------
# Copyright (c) 2021 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# Modified by RainbowSecret from:
#   https://github.com/pytorch/pytorch/blob/master/torch/nn/modules/activation.py#L852
# --------------------------------------------------------

import copy
import math
import warnings
import torch
import torch.nn.functional as F
from torch import nn, Tensor
from torch.nn.modules.module import Module
from torch._jit_internal import Optional, Tuple
from torch.overrides import has_torch_function, handle_torch_function
from torch.nn.functional import linear, pad, softmax, dropout


class MultiheadAttention(Module):
    bias_k: Optional[torch.Tensor]
    bias_v: Optional[torch.Tensor]

    def __init__(
        self,
        embed_dim,
        num_heads,
        dropout=0.0,
        bias=True,
        add_bias_kv=False,
        add_zero_attn=False,
        kdim=None,
        vdim=None,
    ):
        super(MultiheadAttention, self).__init__()
        self.embed_dim = embed_dim
        self.kdim = kdim if kdim is not None else embed_dim
        self.vdim = vdim if vdim is not None else embed_dim
        self._qkv_same_embed_dim = self.kdim == embed_dim and self.vdim == embed_dim

        self.num_heads = num_heads
        self.dropout = dropout
        self.head_dim = embed_dim // num_heads
        assert (
            self.head_dim * num_heads == self.embed_dim
        ), "embed_dim must be divisible by num_heads"

        self.k_proj = nn.Linear(self.kdim, embed_dim, bias=bias)
        self.v_proj = nn.Linear(self.vdim, embed_dim, bias=bias)
        self.q_proj = nn.Linear(embed_dim, embed_dim, bias=bias)
        self.out_proj = nn.Linear(embed_dim, embed_dim)

        self.in_proj_bias = None
        self.in_proj_weight = None
        self.bias_k = self.bias_v = None
        self.q_proj_weight = None
        self.k_proj_weight = None
        self.v_proj_weight = None
        self.add_zero_attn = add_zero_attn

    def __setstate__(self, state):
        # Support loading old MultiheadAttention checkpoints generated by v1.1.0
        if "_qkv_same_embed_dim" not in state:
            state["_qkv_same_embed_dim"] = True

        super(MultiheadAttention, self).__setstate__(state)

    def forward(
        self,
        query,
        key,
        value,
        key_padding_mask=None,
        need_weights=False,
        attn_mask=None,
        residual_attn=None,
    ):
        if not self._qkv_same_embed_dim:
            return self.multi_head_attention_forward(
                query,
                key,
                value,
                self.embed_dim,
                self.num_heads,
                self.in_proj_weight,
                self.in_proj_bias,
                self.bias_k,
                self.bias_v,
                self.add_zero_attn,
                self.dropout,
                self.out_proj.weight,
                self.out_proj.bias,
                training=self.training,
                key_padding_mask=key_padding_mask,
                need_weights=need_weights,
                attn_mask=attn_mask,
                use_separate_proj_weight=True,
                q_proj_weight=self.q_proj_weight,
                k_proj_weight=self.k_proj_weight,
                v_proj_weight=self.v_proj_weight,
                out_dim=self.vdim,
                residual_attn=residual_attn,
            )
        else:
            return self.multi_head_attention_forward(
                query,
                key,
                value,
                self.embed_dim,
                self.num_heads,
                self.in_proj_weight,
                self.in_proj_bias,
                self.bias_k,
                self.bias_v,
                self.add_zero_attn,
                self.dropout,
                self.out_proj.weight,
                self.out_proj.bias,
                training=self.training,
                key_padding_mask=key_padding_mask,
                need_weights=need_weights,
                attn_mask=attn_mask,
                out_dim=self.vdim,
                residual_attn=residual_attn,
            )

    def multi_head_attention_forward(
        self,
        query: Tensor,
        key: Tensor,
        value: Tensor,
        embed_dim_to_check: int,
        num_heads: int,
        in_proj_weight: Tensor,
        in_proj_bias: Tensor,
        bias_k: Optional[Tensor],
        bias_v: Optional[Tensor],
        add_zero_attn: bool,
        dropout_p: float,
        out_proj_weight: Tensor,
        out_proj_bias: Tensor,
        training: bool = True,
        key_padding_mask: Optional[Tensor] = None,
        need_weights: bool = False,
        attn_mask: Optional[Tensor] = None,
        use_separate_proj_weight: bool = False,
        q_proj_weight: Optional[Tensor] = None,
        k_proj_weight: Optional[Tensor] = None,
        v_proj_weight: Optional[Tensor] = None,
        static_k: Optional[Tensor] = None,
        static_v: Optional[Tensor] = None,
        out_dim: Optional[Tensor] = None,
        residual_attn: Optional[Tensor] = None,
    ) -> Tuple[Tensor, Optional[Tensor]]:
        if not torch.jit.is_scripting():
            tens_ops = (
                query,
                key,
                value,
                in_proj_weight,
                in_proj_bias,
                bias_k,
                bias_v,
                out_proj_weight,
                out_proj_bias,
            )
            if any([type(t) is not Tensor for t in tens_ops]) and has_torch_function(
                tens_ops
            ):
                return handle_torch_function(
                    multi_head_attention_forward,
                    tens_ops,
                    query,
                    key,
                    value,
                    embed_dim_to_check,
                    num_heads,
                    in_proj_weight,
                    in_proj_bias,
                    bias_k,
                    bias_v,
                    add_zero_attn,
                    dropout_p,
                    out_proj_weight,
                    out_proj_bias,
                    training=training,
                    key_padding_mask=key_padding_mask,
                    need_weights=need_weights,
                    attn_mask=attn_mask,
                    use_separate_proj_weight=use_separate_proj_weight,
                    q_proj_weight=q_proj_weight,
                    k_proj_weight=k_proj_weight,
                    v_proj_weight=v_proj_weight,
                    static_k=static_k,
                    static_v=static_v,
                )
        tgt_len, bsz, embed_dim = query.size()
        key = query if key is None else key
        value = query if value is None else value

        assert embed_dim == embed_dim_to_check
        # allow MHA to have different sizes for the feature dimension
        assert key.size(0) == value.size(0) and key.size(1) == value.size(1)

        head_dim = embed_dim // num_heads
        v_head_dim = out_dim // num_heads
        assert (
            head_dim * num_heads == embed_dim
        ), "embed_dim must be divisible by num_heads"
        scaling = float(head_dim) ** -0.5

        q = self.q_proj(query) * scaling
        k = self.k_proj(key)
        v = self.v_proj(value)

        if attn_mask is not None:
            assert (
                attn_mask.dtype == torch.float32
                or attn_mask.dtype == torch.float64
                or attn_mask.dtype == torch.float16
                or attn_mask.dtype == torch.uint8
                or attn_mask.dtype == torch.bool
            ), "Only float, byte, and bool types are supported for attn_mask, not {}".format(
                attn_mask.dtype
            )
            if attn_mask.dtype == torch.uint8:
                warnings.warn(
                    "Byte tensor for attn_mask in nn.MultiheadAttention is deprecated. Use bool tensor instead."
                )
                attn_mask = attn_mask.to(torch.bool)

            if attn_mask.dim() == 2:
                attn_mask = attn_mask.unsqueeze(0)
                if list(attn_mask.size()) != [1, query.size(0), key.size(0)]:
                    raise RuntimeError("The size of the 2D attn_mask is not correct.")
            elif attn_mask.dim() == 3:
                if list(attn_mask.size()) != [
                    bsz * num_heads,
                    query.size(0),
                    key.size(0),
                ]:
                    raise RuntimeError("The size of the 3D attn_mask is not correct.")
            else:
                raise RuntimeError(
                    "attn_mask's dimension {} is not supported".format(attn_mask.dim())
                )

        # convert ByteTensor key_padding_mask to bool
        if key_padding_mask is not None and key_padding_mask.dtype == torch.uint8:
            warnings.warn(
                "Byte tensor for key_padding_mask in nn.MultiheadAttention is deprecated. Use bool tensor instead."
            )
            key_padding_mask = key_padding_mask.to(torch.bool)

        q = q.contiguous().view(tgt_len, bsz * num_heads, head_dim).transpose(0, 1)
        if k is not None:
            k = k.contiguous().view(-1, bsz * num_heads, head_dim).transpose(0, 1)
        if v is not None:
            v = v.contiguous().view(-1, bsz * num_heads, v_head_dim).transpose(0, 1)

        src_len = k.size(1)

        if key_padding_mask is not None:
            assert key_padding_mask.size(0) == bsz
            assert key_padding_mask.size(1) == src_len

        if add_zero_attn:
            src_len += 1
            k = torch.cat(
                [
                    k,
                    torch.zeros(
                        (k.size(0), 1) + k.size()[2:], dtype=k.dtype, device=k.device
                    ),
                ],
                dim=1,
            )
            v = torch.cat(
                [
                    v,
                    torch.zeros(
                        (v.size(0), 1) + v.size()[2:], dtype=v.dtype, device=v.device
                    ),
                ],
                dim=1,
            )
            if attn_mask is not None:
                attn_mask = pad(attn_mask, (0, 1))
            if key_padding_mask is not None:
                key_padding_mask = pad(key_padding_mask, (0, 1))

        attn_output_weights = torch.bmm(q, k.transpose(1, 2))
        assert list(attn_output_weights.size()) == [bsz * num_heads, tgt_len, src_len]

        """
        Attention weight for the invalid region is -inf
        """
        if attn_mask is not None:
            if attn_mask.dtype == torch.bool:
                attn_output_weights.masked_fill_(attn_mask, float("-inf"))
            else:
                attn_output_weights += attn_mask

        if key_padding_mask is not None:
            attn_output_weights = attn_output_weights.view(
                bsz, num_heads, tgt_len, src_len
            )
            attn_output_weights = attn_output_weights.masked_fill(
                key_padding_mask.unsqueeze(1).unsqueeze(2),
                float("-inf"),
            )
            attn_output_weights = attn_output_weights.view(
                bsz * num_heads, tgt_len, src_len
            )

        if residual_attn is not None:
            attn_output_weights = attn_output_weights.view(
                bsz, num_heads, tgt_len, src_len
            )
            attn_output_weights += residual_attn.unsqueeze(0)
            attn_output_weights = attn_output_weights.view(
                bsz * num_heads, tgt_len, src_len
            )

        """
        Reweight the attention map before softmax().
        attn_output_weights: (b*n_head, n, hw)
        """
        attn_output_weights = softmax(attn_output_weights, dim=-1)
        attn_output_weights = dropout(
            attn_output_weights, p=dropout_p, training=training
        )

        attn_output = torch.bmm(attn_output_weights, v)
        assert list(attn_output.size()) == [bsz * num_heads, tgt_len, v_head_dim]
        attn_output = (
            attn_output.transpose(0, 1).contiguous().view(tgt_len, bsz, out_dim)
        )
        attn_output = linear(attn_output, out_proj_weight, out_proj_bias)

        if need_weights:
            # average attention weights over heads
            attn_output_weights = attn_output_weights.view(
                bsz, num_heads, tgt_len, src_len
            )
            return attn_output, attn_output_weights.sum(dim=1) / num_heads
        else:
            return attn_output