File size: 3,931 Bytes
e2bca25 |
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 |
import torch
from torch import nn
from torch.nn import functional as F
class ChannelLastConv1d(nn.Conv1d):
def forward(self, x: torch.Tensor) -> torch.Tensor:
x = x.permute(0, 2, 1)
x = super().forward(x)
x = x.permute(0, 2, 1)
return x
# https://github.com/Stability-AI/sd3-ref
class MLP(nn.Module):
def __init__(
self,
in_dim: int,
out_dim: int,
hidden_dim: int,
multiple_of: int = 256,
):
"""
Initialize the FeedForward module.
Args:
dim (int): Input dimension.
hidden_dim (int): Hidden dimension of the feedforward layer.
multiple_of (int): Value to ensure hidden dimension is a multiple of this value.
Attributes:
w1 (ColumnParallelLinear): Linear transformation for the first layer.
w2 (RowParallelLinear): Linear transformation for the second layer.
w3 (ColumnParallelLinear): Linear transformation for the third layer.
"""
super().__init__()
hidden_dim = int(2 * hidden_dim / 3)
hidden_dim = multiple_of * ((hidden_dim + multiple_of - 1) // multiple_of)
self.w1 = nn.Linear(in_dim, hidden_dim, bias=False)
self.w2 = nn.Linear(hidden_dim, out_dim, bias=False)
self.w3 = nn.Linear(in_dim, hidden_dim, bias=False)
def forward(self, x):
return self.w2(F.silu(self.w1(x)) * self.w3(x))
class ConvMLP(nn.Module):
def __init__(
self,
dim: int,
hidden_dim: int,
multiple_of: int = 256,
kernel_size: int = 3,
padding: int = 1,
):
"""
Initialize the FeedForward module.
Args:
dim (int): Input dimension.
hidden_dim (int): Hidden dimension of the feedforward layer.
multiple_of (int): Value to ensure hidden dimension is a multiple of this value.
Attributes:
w1 (ColumnParallelLinear): Linear transformation for the first layer.
w2 (RowParallelLinear): Linear transformation for the second layer.
w3 (ColumnParallelLinear): Linear transformation for the third layer.
"""
super().__init__()
hidden_dim = int(2 * hidden_dim / 3)
hidden_dim = multiple_of * ((hidden_dim + multiple_of - 1) // multiple_of)
self.w1 = ChannelLastConv1d(dim,
hidden_dim,
bias=False,
kernel_size=kernel_size,
padding=padding)
self.w2 = ChannelLastConv1d(hidden_dim,
dim,
bias=False,
kernel_size=kernel_size,
padding=padding)
self.w3 = ChannelLastConv1d(dim,
hidden_dim,
bias=False,
kernel_size=kernel_size,
padding=padding)
def forward(self, x):
return self.w2(F.silu(self.w1(x)) * self.w3(x))
class Attn_Net_Gated(nn.Module):
def __init__(self, hidden_dim):
super().__init__()
self.attention = nn.Sequential(
nn.Linear(hidden_dim, hidden_dim),
nn.Tanh(),
nn.Linear(hidden_dim, 1)
)
self.gate = nn.Sequential(
nn.Linear(hidden_dim, hidden_dim),
nn.Sigmoid()
)
def forward(self, x):
attn_scores = self.attention(x) # (B, T, 1)
gate_values = self.gate(x) # (B, T, D)
gated_attn = attn_scores * gate_values # (B, T, 1)
attn_weights = F.softmax(gated_attn, dim=1) # (B, T, 1)
pooled = torch.sum(attn_weights * x, dim=1) # (B, D)
return pooled |