File size: 3,264 Bytes
4845d25
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.

# References:
#   https://github.com/facebookresearch/dino/blob/master/vision_transformer.py
#   https://github.com/rwightman/pytorch-image-models/tree/master/timm/models/vision_transformer.py

import logging
import torch.nn.functional as F
from torch import Tensor, nn

logger = logging.getLogger("dinov2")


class Attention(nn.Module):
    def __init__(
        self,
        dim: int,
        num_heads: int = 8,
        qkv_bias: bool = False,
        proj_bias: bool = True,
        attn_drop: float = 0.0,
        proj_drop: float = 0.0,
        norm_layer: nn.Module = nn.LayerNorm,
        qk_norm: bool = False,
        fused_attn: bool = True,  # use F.scaled_dot_product_attention or not
        rope=None,
    ) -> None:
        super().__init__()
        assert dim % num_heads == 0, "dim should be divisible by num_heads"
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = head_dim**-0.5
        self.fused_attn = fused_attn

        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
        self.q_norm = norm_layer(head_dim) if qk_norm else nn.Identity()
        self.k_norm = norm_layer(head_dim) if qk_norm else nn.Identity()
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(dim, dim, bias=proj_bias)
        self.proj_drop = nn.Dropout(proj_drop)
        self.rope = rope

    def forward(self, x: Tensor, pos=None, attn_mask=None) -> Tensor:
        B, N, C = x.shape
        qkv = (
            self.qkv(x)
            .reshape(B, N, 3, self.num_heads, C // self.num_heads)
            .permute(2, 0, 3, 1, 4)
        )
        q, k, v = qkv[0], qkv[1], qkv[2]
        q, k = self.q_norm(q), self.k_norm(k)
        if self.rope is not None and pos is not None:
            q = self.rope(q, pos)
            k = self.rope(k, pos)
        if self.fused_attn:
            x = F.scaled_dot_product_attention(
                q,
                k,
                v,
                dropout_p=self.attn_drop.p if self.training else 0.0,
                attn_mask=(
                    (attn_mask)[:, None].repeat(1, self.num_heads, 1, 1)
                    if attn_mask is not None
                    else None
                ),
            )
        else:
            q = q * self.scale
            attn = q @ k.transpose(-2, -1)
            attn = attn.softmax(dim=-1)
            attn = self.attn_drop(attn)
            x = attn @ v

        x = x.transpose(1, 2).reshape(B, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x

    def _forward(self, x: Tensor) -> Tensor:
        B, N, C = x.shape
        qkv = (
            self.qkv(x)
            .reshape(B, N, 3, self.num_heads, C // self.num_heads)
            .permute(2, 0, 3, 1, 4)
        )

        q, k, v = qkv[0] * self.scale, qkv[1], qkv[2]
        attn = q @ k.transpose(-2, -1)

        attn = attn.softmax(dim=-1)
        attn = self.attn_drop(attn)

        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x