File size: 4,199 Bytes
0301e15
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from typing import List, Optional, Set, Type, Union

import torch
from torch import nn


class LoraInjectedLinear(nn.Module):
    """
    Linear layer with LoRA injection.
    Taken from https://github.com/cloneofsimo/lora/blob/master/lora_diffusion/lora.py
    """
    def __init__(
        self, in_features, out_features, bias=False, r=4, dropout_p=0.1, scale=1.0
    ):
        super().__init__()

        if r > min(in_features, out_features):
            raise ValueError(
                f"LoRA rank {r} must be less or equal than {min(in_features, out_features)}"
            )
        self.r = r
        self.linear = nn.Linear(in_features, out_features, bias)
        self.lora_down = nn.Linear(in_features, r, bias=False)
        self.dropout = nn.Dropout(dropout_p)
        self.lora_up = nn.Linear(r, out_features, bias=False)
        self.scale = scale
        self.selector = nn.Identity()

        nn.init.normal_(self.lora_down.weight, std=1 / r)
        nn.init.zeros_(self.lora_up.weight)

    def forward(self, input):
        return (
            self.linear(input.float())
            + self.dropout(self.lora_up(self.selector(self.lora_down(input.float()))))
            * self.scale
        ).half()

    def realize_as_lora(self):
        return self.lora_up.weight.data * self.scale, self.lora_down.weight.data

    def set_selector_from_diag(self, diag: torch.Tensor):
        # diag is a 1D tensor of size (r,)
        assert diag.shape == (self.r,)
        self.selector = nn.Linear(self.r, self.r, bias=False)
        self.selector.weight.data = torch.diag(diag)
        self.selector.weight.data = self.selector.weight.data.to(
            self.lora_up.weight.device
        ).to(self.lora_up.weight.dtype)

class LoraInjectedConv2d(nn.Module):
    def __init__(
        self,
        in_channels: int,
        out_channels: int,
        kernel_size,
        stride=1,
        padding=0,
        dilation=1,
        groups: int = 1,
        bias: bool = True,
        r: int = 4,
        dropout_p: float = 0.1,
        scale: float = 1.0,
    ):
        super().__init__()
        if r > min(in_channels, out_channels):
            raise ValueError(
                f"LoRA rank {r} must be less or equal than {min(in_channels, out_channels)}"
            )
        self.r = r
        self.conv = nn.Conv2d(
            in_channels=in_channels,
            out_channels=out_channels,
            kernel_size=kernel_size,
            stride=stride,
            padding=padding,
            dilation=dilation,
            groups=groups,
            bias=bias,
        )

        self.lora_down = nn.Conv2d(
            in_channels=in_channels,
            out_channels=r,
            kernel_size=kernel_size,
            stride=stride,
            padding=padding,
            dilation=dilation,
            groups=groups,
            bias=False,
        )
        self.dropout = nn.Dropout(dropout_p)
        self.lora_up = nn.Conv2d(
            in_channels=r,
            out_channels=out_channels,
            kernel_size=1,
            stride=1,
            padding=0,
            bias=False,
        )
        self.selector = nn.Identity()
        self.scale = scale

        nn.init.normal_(self.lora_down.weight, std=1 / r)
        nn.init.zeros_(self.lora_up.weight)

    def forward(self, input):
        return (
            self.conv(input)
            + self.dropout(self.lora_up(self.selector(self.lora_down(input))))
            * self.scale
        )

    def realize_as_lora(self):
        return self.lora_up.weight.data * self.scale, self.lora_down.weight.data

    def set_selector_from_diag(self, diag: torch.Tensor):
        # diag is a 1D tensor of size (r,)
        assert diag.shape == (self.r,)
        self.selector = nn.Conv2d(
            in_channels=self.r,
            out_channels=self.r,
            kernel_size=1,
            stride=1,
            padding=0,
            bias=False,
        )
        self.selector.weight.data = torch.diag(diag)

        # same device + dtype as lora_up
        self.selector.weight.data = self.selector.weight.data.to(
            self.lora_up.weight.device
        ).to(self.lora_up.weight.dtype)