File size: 5,469 Bytes
3d1c0e1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Copyright (c) 2025 FoundationVision
# SPDX-License-Identifier: MIT

import torch
import torch.nn as nn
import torch.nn.functional as F
from einops import rearrange


def get_norm(norm_type):
    if norm_type == "spatial-group":
        return SpatialGroupNorm
    elif norm_type == "rms":
        return RMS_norm
    elif norm_type == "group":
        return nn.GroupNorm
    else:
        raise NotImplementedError

class RMS_norm(nn.Module):

    def __init__(self, num_channels, channel_first=True, bias=False, **kwargs):
        super().__init__()
        broadcastable_dims = (1, 1, 1)
        shape = (num_channels, *broadcastable_dims)

        self.channel_first = channel_first
        self.scale = num_channels**0.5
        self.gamma = nn.Parameter(torch.ones(shape))
        self.bias = nn.Parameter(torch.zeros(shape)) if bias else 0.

    def forward(self, x):
        return F.normalize(
            x, dim=(1 if self.channel_first else
                    -1)) * self.scale * self.gamma + self.bias

class SpatialGroupNorm(nn.GroupNorm):
    def __init__(self, *args, **kwargs):
        super(SpatialGroupNorm, self).__init__(*args, **kwargs)
    
    def shard_norm(self, x):
        dtype = x.dtype
        x = x.to(torch.float32)
        with torch.amp.autocast("cuda", torch.float32):
            for _i in range(x.shape[0]):
                x[_i:_i+1,...] = super(SpatialGroupNorm, self).forward(x[_i:_i+1,...])
        x = x.to(dtype=dtype)
        return x

    def forward(self, x):
        dtype = x.dtype
        x = x.to(torch.float32)
        assert x.ndim == 5
        T = x.shape[2]
        x = rearrange(x, "B C T H W -> (B T) C H W")
        try:
            x = super(SpatialGroupNorm, self).forward(x)
        except:
            x = self.shard_norm(x) # shard norm if OOM fallback
        x = rearrange(x, "(B T) C H W -> B C T H W", T=T)
        x = x.to(dtype=dtype)
        return x

class Normalize(nn.Module):
    def __init__(self, in_channels, norm_type, norm_axis="spatial"):
        super().__init__()
        self.norm_axis = norm_axis
        assert norm_type in ['group', 'batch', "no"]
        if norm_type == 'group':
            if in_channels % 32 == 0:
                self.norm = nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True)
            elif in_channels % 24 == 0: 
                self.norm = nn.GroupNorm(num_groups=24, num_channels=in_channels, eps=1e-6, affine=True)
            else:
                raise NotImplementedError
        elif norm_type == 'batch':
            self.norm = nn.SyncBatchNorm(in_channels, track_running_stats=False) # Runtime Error: grad inplace if set track_running_stats to True
        elif norm_type == 'no':
            self.norm = nn.Identity()
    
    def _norm(self, x):
        try:
            x = self.norm(x)
        except:
            device = x.device
            self.norm_cpu = self.norm.cpu()
            x = self.norm_cpu(x.cpu().pin_memory()).to(device=device)
        return x

    def shard_norm(self, x):
        dtype = x.dtype
        x = x.to(torch.float32)
        with torch.amp.autocast("cuda", torch.float32):
            for _i in range(x.shape[0]):
                x[_i:_i+1,...] = self.norm(x[_i:_i+1,...])
        x = x.to(dtype=dtype)
        return x

    def forward(self, x):
        if self.norm_axis == "spatial":
            if type(x) == list:
                for i in range(len(x)):
                    x[i] = self.norm(x[i])
                return x
            if x.ndim == 4:
                try:
                    x = self.norm(x)
                except:
                    x = self.shard_norm(x)
            else:
                B, C, T, H, W = x.shape
                x = rearrange(x, "B C T H W -> (B T) C H W")
                # x = self.shard_norm(x)
                try:
                    x = self.norm(x)
                except:
                    x = self.shard_norm(x)
                x = rearrange(x, "(B T) C H W -> B C T H W", T=T)
        elif self.norm_axis == "spatial-temporal":
            x = self._norm(x)
        else:
            raise NotImplementedError
        return x

def l2norm(t):
    return F.normalize(t, dim=-1)

class LayerNorm(nn.Module):
    def __init__(self, dim):
        super().__init__()
        self.gamma = nn.Parameter(torch.ones(dim))
        self.register_buffer("beta", torch.zeros(dim))

    def forward(self, x):
        return F.layer_norm(x, x.shape[-1:], self.gamma, self.beta)

# https://github.com/huggingface/transformers/blob/2f12e408225b1ebceb0d2f701ce419d46678dc31/src/transformers/models/llama/modeling_llama.py#L76
class RMSNorm(nn.Module):
    def __init__(self, hidden_size, eps=1e-6):
        """
        LlamaRMSNorm is equivalent to T5LayerNorm
        """
        super().__init__()
        self.weight = nn.Parameter(torch.ones(hidden_size))
        self.variance_epsilon = eps

    def forward(self, hidden_states, sp_slice=None):
        input_dtype = hidden_states.dtype
        hidden_states = hidden_states.to(torch.float32)
        variance = hidden_states.pow(2).mean(-1, keepdim=True)
        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
        if sp_slice is None:
            return (self.weight * hidden_states).to(input_dtype)
        else:
            return (self.weight[sp_slice] * hidden_states).to(input_dtype)  # torch.float32 * torchbfloat16 in DDP will cast to torch.float32