File size: 4,154 Bytes
1d409a9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
from contextlib import contextmanager
import ldm_patched.modules.model_management

def cast_bias_weight(s, input):
    bias = None
    non_blocking = ldm_patched.modules.model_management.device_supports_non_blocking(input.device)
    if s.bias is not None:
        bias = s.bias.to(device=input.device, dtype=input.dtype, non_blocking=non_blocking)
    weight = s.weight.to(device=input.device, dtype=input.dtype, non_blocking=non_blocking)
    return weight, bias


class disable_weight_init:
    class Linear(torch.nn.Linear):
        ldm_patched_cast_weights = False
        def reset_parameters(self):
            return None

        def forward_ldm_patched_cast_weights(self, input):
            weight, bias = cast_bias_weight(self, input)
            return torch.nn.functional.linear(input, weight, bias)

        def forward(self, *args, **kwargs):
            if self.ldm_patched_cast_weights:
                return self.forward_ldm_patched_cast_weights(*args, **kwargs)
            else:
                return super().forward(*args, **kwargs)

    class Conv2d(torch.nn.Conv2d):
        ldm_patched_cast_weights = False
        def reset_parameters(self):
            return None

        def forward_ldm_patched_cast_weights(self, input):
            weight, bias = cast_bias_weight(self, input)
            return self._conv_forward(input, weight, bias)

        def forward(self, *args, **kwargs):
            if self.ldm_patched_cast_weights:
                return self.forward_ldm_patched_cast_weights(*args, **kwargs)
            else:
                return super().forward(*args, **kwargs)

    class Conv3d(torch.nn.Conv3d):
        ldm_patched_cast_weights = False
        def reset_parameters(self):
            return None

        def forward_ldm_patched_cast_weights(self, input):
            weight, bias = cast_bias_weight(self, input)
            return self._conv_forward(input, weight, bias)

        def forward(self, *args, **kwargs):
            if self.ldm_patched_cast_weights:
                return self.forward_ldm_patched_cast_weights(*args, **kwargs)
            else:
                return super().forward(*args, **kwargs)

    class GroupNorm(torch.nn.GroupNorm):
        ldm_patched_cast_weights = False
        def reset_parameters(self):
            return None

        def forward_ldm_patched_cast_weights(self, input):
            weight, bias = cast_bias_weight(self, input)
            return torch.nn.functional.group_norm(input, self.num_groups, weight, bias, self.eps)

        def forward(self, *args, **kwargs):
            if self.ldm_patched_cast_weights:
                return self.forward_ldm_patched_cast_weights(*args, **kwargs)
            else:
                return super().forward(*args, **kwargs)


    class LayerNorm(torch.nn.LayerNorm):
        ldm_patched_cast_weights = False
        def reset_parameters(self):
            return None

        def forward_ldm_patched_cast_weights(self, input):
            weight, bias = cast_bias_weight(self, input)
            return torch.nn.functional.layer_norm(input, self.normalized_shape, weight, bias, self.eps)

        def forward(self, *args, **kwargs):
            if self.ldm_patched_cast_weights:
                return self.forward_ldm_patched_cast_weights(*args, **kwargs)
            else:
                return super().forward(*args, **kwargs)

    @classmethod
    def conv_nd(s, dims, *args, **kwargs):
        if dims == 2:
            return s.Conv2d(*args, **kwargs)
        elif dims == 3:
            return s.Conv3d(*args, **kwargs)
        else:
            raise ValueError(f"unsupported dimensions: {dims}")


class manual_cast(disable_weight_init):
    class Linear(disable_weight_init.Linear):
        ldm_patched_cast_weights = True

    class Conv2d(disable_weight_init.Conv2d):
        ldm_patched_cast_weights = True

    class Conv3d(disable_weight_init.Conv3d):
        ldm_patched_cast_weights = True

    class GroupNorm(disable_weight_init.GroupNorm):
        ldm_patched_cast_weights = True

    class LayerNorm(disable_weight_init.LayerNorm):
        ldm_patched_cast_weights = True