Spaces:
Sleeping
Sleeping
File size: 10,367 Bytes
f2f112a |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 |
#NOTE: Hanning, Transformer Decoder
import torch
import numpy as np
import copy
import torch.nn.functional as F
def _get_clones(module, N):
return torch.nn.ModuleList([copy.deepcopy(module) for i in range(N)])
def _get_activation_fn(activation):
if activation == "relu":
return F.relu
if activation == "gelu":
return F.gelu
if activation == "glu":
return F.glu
raise RuntimeError(f"activation should be relu/gelu, not {activation}.")
class TransformerDecoder(torch.nn.Module):
def __init__(self, decoder_layer, num_layers, norm=None, return_intermediate=False):
super().__init__()
self.layers = _get_clones(decoder_layer, num_layers)
self.num_layers = num_layers
self.norm = norm
self.return_intermediate = return_intermediate
def forward(
self,
tgt,
memory,
text_memory,
tgt_mask = None,
memory_mask = None,
text_memory_key_padding_mask = None,
tgt_key_padding_mask = None,
memory_key_padding_mask = None,
pos = None,
query_pos = None,
):
output = tgt
intermediate = []
for layer in self.layers:
output, memory = layer(
output,
memory,
text_memory=text_memory,
tgt_mask=tgt_mask,
memory_mask=memory_mask,
text_memory_key_padding_mask=text_memory_key_padding_mask,
tgt_key_padding_mask=tgt_key_padding_mask,
memory_key_padding_mask=memory_key_padding_mask,
pos=pos,
query_pos=query_pos,
)
if self.return_intermediate:
intermediate.append(self.norm(output))
if self.norm is not None:
output = self.norm(output)
memory = self.norm(memory)
if self.return_intermediate:
intermediate.pop()
intermediate.append(output)
if self.return_intermediate:
return torch.stack(intermediate)
return output, memory
class TransformerDecoderLayer(torch.nn.Module):
def __init__(self, d_model, nhead, dim_feedforward=2048, dropout=0.1, activation="relu", normalize_before=False):
super().__init__()
self.self_attn_text = torch.nn.MultiheadAttention(d_model, nhead, dropout=dropout)
self.self_attn_image = torch.nn.MultiheadAttention(d_model, nhead, dropout=dropout)
self.cross_attn_image = torch.nn.MultiheadAttention(d_model, nhead, dropout=dropout)
# Implementation of Feedforward model
self.linear1 = torch.nn.Linear(d_model, dim_feedforward)
self.dropout = torch.nn.Dropout(dropout)
self.linear2 = torch.nn.Linear(dim_feedforward, d_model)
self.norm1 = torch.nn.LayerNorm(d_model)
# self.norm2 = nn.LayerNorm(d_model)
self.norm3 = torch.nn.LayerNorm(d_model)
self.norm4 = torch.nn.LayerNorm(d_model)
self.dropout1 = torch.nn.Dropout(dropout)
# self.dropout2 = nn.Dropout(dropout)
self.dropout3 = torch.nn.Dropout(dropout)
self.dropout4 = torch.nn.Dropout(dropout)
self.activation = _get_activation_fn(activation)
self.normalize_before = normalize_before
def with_pos_embed(self, tensor, pos):
return tensor if pos is None else tensor + pos
# For now, trying one version where its self attn -> cross attn text -> cross attn image -> FFN
def forward_post(
self,
tgt,
memory,
text_memory,
tgt_mask = None,
memory_mask = None,
text_memory_key_padding_mask = None,
tgt_key_padding_mask = None,
memory_key_padding_mask = None,
pos = None,
query_pos = None,
):
#NOTE: tgt is vision and memory is text
q_vision = self.with_pos_embed(tgt, query_pos)
k_vision = self.with_pos_embed(tgt, query_pos)
tgt2 = self.self_attn_image(q_vision, k_vision, value=tgt, attn_mask=tgt_mask, key_padding_mask=tgt_key_padding_mask)[0]
tgt = tgt + self.dropout1(tgt2)
tgt = self.norm1(tgt)
#NOTE: memory 2 is None, need to figure out
q_text = self.with_pos_embed(memory,query_pos)
k_text = self.with_pos_embed(memory,query_pos)
memory2 = self.self_attn_text(q_text,k_text,value=memory,attn_mask=tgt_mask, key_padding_mask=tgt_key_padding_mask)[0]
memory = memory + self.dropout1(memory2)
memory = self.norm1(memory)
# Cross attention to image
tgt2 = self.cross_attn_image(
query=self.with_pos_embed(tgt, query_pos),
key=self.with_pos_embed(memory, pos),
value=memory,
attn_mask=memory_mask,
key_padding_mask=memory_key_padding_mask,
)[0]
tgt = tgt + self.dropout3(tgt2)
tgt = self.norm3(tgt)
# FFN
tgt2 = self.linear2(self.dropout(self.activation(self.linear1(tgt))))
tgt = tgt + self.dropout4(tgt2)
tgt = self.norm4(tgt)
return tgt, memory
def forward(
self,
tgt,
memory,
text_memory,
tgt_mask = None,
memory_mask = None,
text_memory_key_padding_mask = None,
tgt_key_padding_mask = None,
memory_key_padding_mask = None,
pos = None,
query_pos = None,
):
return self.forward_post(
tgt,
memory,
text_memory,
tgt_mask,
memory_mask,
text_memory_key_padding_mask,
tgt_key_padding_mask,
memory_key_padding_mask,
pos,
query_pos,
)
class TransformerCrossDecoderLayer(torch.nn.Module):
def __init__(self, d_model, nhead, dim_feedforward=2048, dropout=0.1, activation="relu", normalize_before=False):
super().__init__()
self.self_attn_text = torch.nn.MultiheadAttention(d_model, nhead, dropout=dropout)
self.self_attn_image = torch.nn.MultiheadAttention(d_model, nhead, dropout=dropout)
self.cross_attn_image = torch.nn.MultiheadAttention(d_model, nhead, dropout=dropout)
self.cross_attn_text = torch.nn.MultiheadAttention(d_model, nhead, dropout=dropout)
# Implementation of Feedforward model
self.linear1 = torch.nn.Linear(d_model, dim_feedforward)
self.linear2 = torch.nn.Linear(dim_feedforward, d_model)
self.linear3 = torch.nn.Linear(d_model, dim_feedforward)
self.linear4 = torch.nn.Linear(dim_feedforward, d_model)
self.norm0 = torch.nn.LayerNorm(d_model)
self.norm1 = torch.nn.LayerNorm(d_model)
self.norm2 = torch.nn.LayerNorm(d_model)
self.norm3 = torch.nn.LayerNorm(d_model)
self.norm4 = torch.nn.LayerNorm(d_model)
self.norm5 = torch.nn.LayerNorm(d_model)
self.dropout0 = torch.nn.Dropout(dropout)
self.dropout1 = torch.nn.Dropout(dropout)
self.dropout2 = torch.nn.Dropout(dropout)
self.dropout3 = torch.nn.Dropout(dropout)
self.dropout4 = torch.nn.Dropout(dropout)
self.dropout5 = torch.nn.Dropout(dropout)
self.activation = _get_activation_fn(activation)
self.normalize_before = normalize_before
def with_pos_embed(self, tensor, pos):
return tensor if pos is None else tensor + pos
# For now, trying one version where its self attn -> cross attn text -> cross attn image -> FFN
def forward_post(
self,
tgt,
memory,
text_memory,
tgt_mask = None,
memory_mask = None,
text_memory_key_padding_mask = None,
tgt_key_padding_mask = None,
memory_key_padding_mask = None,
pos = None,
query_pos = None,
):
#NOTE: tgt is vision and memory is text
q_vision = self.with_pos_embed(tgt, query_pos)
k_vision = self.with_pos_embed(tgt, query_pos)
tgt2 = self.self_attn_image(q_vision, k_vision, value=tgt, attn_mask=tgt_mask, key_padding_mask=tgt_key_padding_mask)[0]
tgt = tgt + self.dropout0(tgt2)
tgt = self.norm0(tgt)
#NOTE: memory 2 is None, need to figure out
q_text = self.with_pos_embed(memory,query_pos)
k_text = self.with_pos_embed(memory,query_pos)
memory2 = self.self_attn_text(q_text,k_text,value=memory,attn_mask=tgt_mask, key_padding_mask=tgt_key_padding_mask)[0]
memory = memory + self.dropout1(memory2)
memory = self.norm1(memory)
# Cross attention to image
tgt2 = self.cross_attn_image(
query=self.with_pos_embed(tgt, query_pos),
key=self.with_pos_embed(memory, pos),
value=memory,
attn_mask=memory_mask,
key_padding_mask=memory_key_padding_mask,
)[0]
tgt = tgt + self.dropout2(tgt2)
tgt = self.norm2(tgt)
# Cross attention to text
memory2 = self.cross_attn_image(
query=self.with_pos_embed(memory, query_pos),
key=self.with_pos_embed(tgt, pos),
value=tgt,
attn_mask=memory_mask,
key_padding_mask=memory_key_padding_mask,
)[0]
memory = memory + self.dropout3(memory2)
memory = self.norm3(memory)
# FFN
tgt2 = self.linear2(self.dropout4(self.activation(self.linear1(tgt))))
tgt = tgt + self.dropout4(tgt2)
tgt = self.norm4(tgt)
memory2 = self.linear4(self.dropout5(self.activation(self.linear3(memory))))
memory = memory + self.dropout5(memory2)
memory = self.norm5(memory)
return tgt, memory
def forward(
self,
tgt,
memory,
text_memory,
tgt_mask = None,
memory_mask = None,
text_memory_key_padding_mask = None,
tgt_key_padding_mask = None,
memory_key_padding_mask = None,
pos = None,
query_pos = None,
):
return self.forward_post(
tgt,
memory,
text_memory,
tgt_mask,
memory_mask,
text_memory_key_padding_mask,
tgt_key_padding_mask,
memory_key_padding_mask,
pos,
query_pos,
) |