File size: 7,088 Bytes
093b0a5 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 | import math
import torch
import torch.nn as nn
import torch.nn.functional as F
# --------------------------
# Utilities / Norm / Activations
# --------------------------
class RMSNorm(nn.Module):
"""RMSNorm with learnable weight. Drop-in for LayerNorm when using Pre-Norm."""
def __init__(self, d_model, eps=1e-8):
super().__init__()
self.eps = eps
self.weight = nn.Parameter(torch.ones(d_model))
def forward(self, x):
# x: (..., D)
norm = x.pow(2).mean(dim=-1, keepdim=True).add(self.eps).rsqrt()
return self.weight * x * norm
def get_activation(name: str):
name = (name or "relu").lower()
if name == "relu":
return nn.ReLU()
if name in ("gelu", "geglu"):
return nn.GELU()
if name in ("silu", "swish"):
return nn.SiLU()
return nn.ReLU()
class SwiGLU(nn.Module):
"""SwiGLU FFN: proj( SiLU(a) * b ), a,b from linear split."""
def __init__(self, d_model, d_ff):
super().__init__()
self.w12 = nn.Linear(d_model, 2 * d_ff, bias=True)
self.proj = nn.Linear(d_ff, d_model, bias=True)
def forward(self, x):
a, b = self.w12(x).chunk(2, dim=-1)
return self.proj(F.silu(a) * b)
# --------------------------
# Conv Layer (kept signature)
# --------------------------
class ConvLayer(nn.Module):
def __init__(self, c_in):
super(ConvLayer, self).__init__()
padding = 1 if torch.__version__ >= "1.5.0" else 2
self.downConv = nn.Conv1d(
in_channels=c_in,
out_channels=c_in,
kernel_size=3,
padding=padding,
padding_mode="circular",
)
self.norm = nn.BatchNorm1d(c_in)
self.activation = nn.ELU()
self.maxPool = nn.MaxPool1d(kernel_size=3, stride=2, padding=1)
def forward(self, x):
# x: [B, L, D]
x = x.permute(0, 2, 1) # B, D, L
y = self.downConv(x)
y = self.norm(y)
y = self.activation(y)
y = self.maxPool(y)
y = y.transpose(1, 2).contiguous() # B, L', D
return y
# --------------------------
# Encoder Layer (kept signature)
# --------------------------
class EncoderLayer(nn.Module):
"""
Keep the same signature:
__init__(self, attention, d_model, d_ff=None, dropout=0.1, activation="relu", ln_mode="pre", conv_layer=False)
forward(self, x, attn_mask=None)
Internals:
- Pre-Norm by default (ln_mode="pre")
- RMSNorm (instead of LN) but callable doesn’t change
- Residual scaling 1/sqrt(2)
- FFN uses SwiGLU, dropout after branch
- Attention module is expected to have signature (q, k, v, attn_mask=None) -> (new_x, attn)
and internally do q *= 1/sqrt(d_head)
"""
def __init__(
self,
attention,
d_model,
d_ff=None,
dropout=0.1,
activation="relu",
ln_mode="pre",
conv_layer=False,
):
super(EncoderLayer, self).__init__()
self.attention = attention
self.conv_layer = ConvLayer(d_model) if conv_layer else None
self.dropout = nn.Dropout(dropout)
self.activation = get_activation(activation)
self.ln_mode = ln_mode # will honor "pre" / "post" without changing signature
# Core hyperparams
self.d_model = d_model
self.d_ff = d_ff or 4 * d_model
self.res_scale = 1.0 / math.sqrt(2.0)
# Use RMSNorm but keep object names norm1/norm2 to avoid API change elsewhere
self.norm1 = RMSNorm(d_model)
self.norm2 = RMSNorm(d_model)
# FFN: use SwiGLU for better stability/accuracy
self.ff = SwiGLU(d_model, self.d_ff)
# In case some pipeline expects LayerNorm instance, we also keep a post-norm if ln_mode="post"
# (But the actual normalization used is RMSNorm above; this is just to respect the mode)
if self.ln_mode == "post":
self.post_ln1 = nn.LayerNorm(d_model)
self.post_ln2 = nn.LayerNorm(d_model)
def forward(self, x, attn_mask=None):
# x: [B, L, D]
if self.conv_layer is not None:
x = x + self.dropout(self.conv_layer(x)) * self.res_scale
if self.ln_mode == "post":
# -------- Post-LN path (kept behavior but more explicit/clean) --------
new_x, attn = self.attention(x, x, x, attn_mask=attn_mask)
x = x + self.dropout(new_x) * self.res_scale
x = self.post_ln1(x)
y = self.ff(x)
x = x + self.dropout(y) * self.res_scale
x = self.post_ln2(x)
return x, attn
# -------- Default: Pre-LN (recommended) --------
# Attention branch (Pre-Norm)
h, attn = self.attention(self.norm1(x), self.norm1(x), self.norm1(x), attn_mask=attn_mask)
x = x + self.dropout(h) * self.res_scale
# FFN branch (Pre-Norm)
y = self.ff(self.norm2(x))
x = x + self.dropout(y) * self.res_scale
return x, attn
# --------------------------
# Encoder (kept signature)
# --------------------------
class Encoder(nn.Module):
"""
Keep the same signature:
__init__(self, attn_layers, conv_layers=None, norm_layer=None)
forward(self, x, attn_mask=None)
"""
def __init__(self, attn_layers, conv_layers=None, norm_layer=None):
super(Encoder, self).__init__()
self.attn_layers = nn.ModuleList(attn_layers)
self.conv_layers = nn.ModuleList(conv_layers) if conv_layers is not None else None
self.norm = norm_layer # can be None or nn.Module
def forward(self, x, attn_mask=None):
# x: [B, L, D]
attns = []
for i, attn_layer in enumerate(self.attn_layers):
x, attn = attn_layer(x, attn_mask=attn_mask)
attns.append(attn)
if self.conv_layers is not None and i < len(self.conv_layers):
x = self.conv_layers[i](x)
if self.norm is not None:
x = self.norm(x)
return x, attns
# --------------------------
# Encoder Stack (kept signature)
# --------------------------
class EncoderStack(nn.Module):
"""
Keep the same signature:
__init__(self, encoders, inp_lens, d_model)
forward(self, x, attn_mask=None)
"""
def __init__(self, encoders, inp_lens, d_model):
super(EncoderStack, self).__init__()
self.encoders = nn.ModuleList(encoders)
self.inp_lens = inp_lens
self.d_model = d_model
def forward(self, x, attn_mask=None):
# x: [B, L, D]
x_stack = []
attns = []
# For each pyramid level, take the tail part of the sequence
for i_len, encoder in zip(self.inp_lens, self.encoders):
inp_len = x.shape[1] // (2 ** i_len)
x_s, attn = encoder(x[:, -inp_len:, :], attn_mask=attn_mask)
x_stack.append(x_s)
attns.append(attn)
x_stack = torch.cat(x_stack, dim=-2) # concat on sequence length axis
return x_stack, attns
|