-
Notifications
You must be signed in to change notification settings - Fork 1
/
UNET_v3.py
523 lines (432 loc) · 21 KB
/
UNET_v3.py
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
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
import torch.nn as nn
import torch
import torch.nn.functional as F
import math
import warnings
import matplotlib.pyplot as plt
warnings.simplefilter("ignore")
""" To incorporate timestep embedding alongwith unet of 4 tranfomers in enc-dec """
class Embedding(nn.Module):
def __init__(self, vocab_size, embed_dim):
"""
Args:
vocab_size: size of vocabulary
embed_dim: dimension of embeddings
"""
super(Embedding, self).__init__()
self.embed = nn.Embedding(vocab_size, embed_dim)
def forward(self, x):
"""
Args:
x: input vector
Returns:
out: embedding vector
"""
out = self.embed(x)
return out
class PositionalEmbedding(nn.Module):
def __init__(self,max_seq_len,embed_model_dim):
"""
Args:
seq_len: length of input sequence
embed_model_dim: demension of embedding
"""
super(PositionalEmbedding, self).__init__()
self.embed_dim = embed_model_dim
pe = torch.zeros(max_seq_len,self.embed_dim)
for pos in range(max_seq_len):
for i in range(0,self.embed_dim,2):
pe[pos, i] = math.sin(pos / (10000 ** ((2 * i)/self.embed_dim)))
pe[pos, i + 1] = math.cos(pos / (10000 ** ((2 * (i + 1))/self.embed_dim)))
pe = pe.unsqueeze(0)
self.register_buffer('pe', pe)
def forward(self, x):
"""
Args:
x: input vector
Returns:
x: output
"""
# make embeddings relatively larger
x = x * math.sqrt(self.embed_dim)
if x.ndim == 2 :
x = x.unsqueeze(0)
seq_len = x.size(1)
x = x + torch.autograd.Variable(self.pe[:,:seq_len], requires_grad=False)
return x
class MultiHeadAttention(nn.Module):
def __init__(self, embed_dim=512, n_heads=4):
"""
Args:
embed_dim: dimension of embeding vector output
n_heads: number of self attention heads
"""
super(MultiHeadAttention, self).__init__()
self.embed_dim = embed_dim #512 dim
self.n_heads = n_heads #8
self.single_head_dim = int(self.embed_dim / self.n_heads) #512/8 = 64 . each key,query, value will be of 64d
#key,query and value matrixes #64 x 64
self.query_matrix = nn.Linear(self.single_head_dim , self.single_head_dim ,bias=False) # single key matrix for all 8 keys #512x512
self.key_matrix = nn.Linear(self.single_head_dim , self.single_head_dim, bias=False)
self.value_matrix = nn.Linear(self.single_head_dim ,self.single_head_dim , bias=False)
self.out = nn.Linear(self.n_heads*self.single_head_dim ,self.embed_dim)
def forward(self,key,query,value,mask=None): #batch_size x sequence_length x embedding_dim # 32 x 10 x 512
"""
Args:
key : key vector
query : query vector
value : value vector
mask: mask for decoder
Returns:
output vector from multihead attention
"""
batch_size = key.size(0)
seq_length = key.size(1)
# query dimension can change in decoder during inference.
# so we cant take general seq_length
seq_length_query = query.size(1)
# 32x10x512
key = key.view(batch_size, seq_length, self.n_heads, self.single_head_dim) #batch_size x sequence_length x n_heads x single_head_dim = (32x10x8x64)
query = query.view(batch_size, seq_length_query, self.n_heads, self.single_head_dim) #(32x10x8x64)
value = value.view(batch_size, seq_length, self.n_heads, self.single_head_dim) #(32x10x8x64)
k = self.key_matrix(key) # (32x10x8x64)
q = self.query_matrix(query)
v = self.value_matrix(value)
q = q.transpose(1,2) # (batch_size, n_heads, seq_len, single_head_dim) # (32 x 8 x 10 x 64)
k = k.transpose(1,2) # (batch_size, n_heads, seq_len, single_head_dim)
v = v.transpose(1,2) # (batch_size, n_heads, seq_len, single_head_dim)
# computes attention
# adjust key for matrix multiplication
k_adjusted = k.transpose(-1,-2) #(batch_size, n_heads, single_head_dim, seq_ken) #(32 x 8 x 64 x 10)
product = torch.matmul(q, k_adjusted) #(32 x 8 x 10 x 64) x (32 x 8 x 64 x 10) = #(32x8x10x10)
# fill those positions of product matrix as (-1e20) where mask positions are 0
if mask is not None:
product = product.masked_fill(mask == 0, float("-1e20"))
#divising by square root of key dimension
product = product / math.sqrt(self.single_head_dim) # / sqrt(64)
#applying softmax
scores = F.softmax(product, dim=-1)
#mutiply with value matrix
scores = torch.matmul(scores, v) ##(32x8x 10x 10) x (32 x 8 x 10 x 64) = (32 x 8 x 10 x 64)
#concatenated output
concat = scores.transpose(1,2).contiguous().view(batch_size, seq_length_query, self.single_head_dim*self.n_heads) # (32x8x10x64) -> (32x10x8x64) -> (32,10,512)
output = self.out(concat) #(32,10,512) -> (32,10,512)
return output
class DecoderBlock(nn.Module):
def __init__(self, embed_dim, expansion_factor=2, n_heads=4):
super(DecoderBlock, self).__init__()
"""
Args:
embed_dim: dimension of the embedding
expansion_factor: fator ehich determines output dimension of linear layer
n_heads: number of attention heads
"""
self.attention = MultiHeadAttention(embed_dim, n_heads=8)
self.norm = nn.LayerNorm(embed_dim)
self.dropout = nn.Dropout(0.2)
self.transformer_block = TransformerBlock(embed_dim, expansion_factor, n_heads)
def forward(self, key, query, x,mask):
"""
Args:
key: key vector
query: query vector
value: value vector
mask: mask to be given for multi head attention
Returns:
out: output of transformer block
"""
#we need to pass mask mask only to fst attention
attention = self.attention(x,x,x,mask=mask) #32x10x512
value = self.dropout(self.norm(attention + x))
out = self.transformer_block(key, query, value)
return out
class TransformerDecoder(nn.Module):
def __init__(self, target_vocab_size, embed_dim, seq_len, num_layers=2, expansion_factor=2, n_heads=4):
super(TransformerDecoder, self).__init__()
"""
Args:
target_vocab_size: vocabulary size of taget
embed_dim: dimension of embedding
seq_len : length of input sequence
num_layers: number of encoder layers
expansion_factor: factor which determines number of linear layers in feed forward layer
n_heads: number of heads in multihead attention
"""
self.word_embedding = nn.Linear(1, embed_dim)
self.position_embedding = PositionalEmbedding(seq_len, embed_dim)
self.layers = nn.ModuleList(
[
DecoderBlock(embed_dim, expansion_factor=2, n_heads=4)
for _ in range(num_layers)
]
)
self.fc_out = nn.Linear(embed_dim, target_vocab_size)
self.dropout = nn.Dropout(0.2)
def forward(self, x, enc_out, mask):
"""
Args:
x: input vector from target
enc_out : output from encoder layer
trg_mask: mask for decoder self attention
Returns:
out: output vector
"""
# print(x.shape,'text ka')
# print(enc_out.shape,'enc opth')
x = x[:,:,None]
x = self.word_embedding(x) #32x10x512
x = self.position_embedding(x) #32x10x512
# print(x.shape,'text pos emb')
x = self.dropout(x)
# print(x.shape,'text dr')
for layer in self.layers:
x = layer(enc_out, x, enc_out, mask)
# print(x.shape,'text dec')
out = F.softmax(self.fc_out(x))
# print(out.shape,'text ksify a')
return out
class TransformerBlock(nn.Module):
def __init__(self, embed_dim, expansion_factor=2, n_heads=4):
super(TransformerBlock, self).__init__()
"""
Args:
embed_dim: dimension of the embedding
expansion_factor: fator ehich determines output dimension of linear layer
n_heads: number of attention heads
"""
self.attention = MultiHeadAttention(embed_dim, n_heads)
self.norm1 = nn.LayerNorm(embed_dim)
# self.norm2 = nn.LayerNorm(embed_dim)
self.FF1 = nn.Linear(embed_dim, expansion_factor*embed_dim)
self.FF2 = nn.ReLU()
self.FF3 = nn.Linear(expansion_factor*embed_dim, embed_dim)
# self.feed_forward = nn.Sequential(
# nn.Linear(embed_dim, expansion_factor*embed_dim),
# nn.ReLU(),
# nn.Linear(expansion_factor*embed_dim, embed_dim)
# )
self.dropout2 = nn.Dropout(0.2)
def forward(self,key,query,value):
"""
Args:
key: key vector
query: query vector
value: value vector
norm2_out: output of transformer block
"""
attention_out = self.attention(key,query,value) #32x10x512
attention_residual_out = attention_out + value #32x10x512
norm1_out = self.dropout2(self.norm1(attention_residual_out)) #32x10x512
feed_fwd_out = self.FF3(self.FF2(self.FF1(norm1_out))) #32x10x512 -> #32x10x2048 -> 32x10x512
feed_fwd_residual_out = feed_fwd_out + norm1_out #32x10x512
norm2_out = self.dropout2(self.norm1(feed_fwd_residual_out)) #32x10x512
return norm2_out
class TransformerEncoder(nn.Module):
"""
Args:
seq_len : length of input sequence
embed_dim: dimension of embedding
num_layers: number of encoder layers
expansion_factor: factor which determines number of linear layers in feed forward layer
n_heads: number of heads in multihead attention
Returns:
out: output of the encoder
"""
def __init__(self, seq_len, vocab_size, embed_dim, num_layers=2, expansion_factor=2, n_heads=4):
super(TransformerEncoder, self).__init__()
self.vocab_size = vocab_size
self.embedding_layer = nn.Linear(1, embed_dim)
self.positional_encoder = PositionalEmbedding(seq_len, embed_dim)
self.Lin = nn.Linear(vocab_size,1)
self.layers = nn.ModuleList([TransformerBlock(embed_dim, expansion_factor, n_heads) for i in range(num_layers)])
def forward(self, x):
# print(x.shape,'enc ip')
x = x[:,:,:,None]#.view(-1)
# x = torch.zeros(x.size(0), self.vocab_size)
# x.scatter_(1, x.unsqueeze(1), 1)
embed_out = self.embedding_layer(x)
# print(embed_out.shape,'embed')
embed_out = embed_out.permute(0,1,3,2)
embed_out = self.Lin(embed_out).squeeze()
# print(embed_out.shape,'s')
out = self.positional_encoder(embed_out)
# print(out.shape,'posenc')
# print(out.shape,'reshaped')
for layer in self.layers:
out = layer(out,out,out)
# print(out.shape,'final enc op')
return out #32x10x512
class SinusoidalPositionEmbeddings(nn.Module):
#gives the time embeddings as input in Unet
def __init__(self, total_time_steps=1000, time_emb_dims=128, time_emb_dims_exp=512):
super().__init__()
half_dim = time_emb_dims // 2
emb = math.log(10000) / (half_dim - 1)
emb = torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb)
ts = torch.arange(total_time_steps, dtype=torch.float32)
emb = torch.unsqueeze(ts, dim=-1) * torch.unsqueeze(emb, dim=0)
self.emb = torch.cat((emb.sin(), emb.cos()), dim=-1)
self.a = nn.Embedding.from_pretrained(emb)
self.b = nn.Linear(in_features=time_emb_dims, out_features=time_emb_dims_exp)
self.c = nn.Linear(in_features=time_emb_dims_exp, out_features=time_emb_dims_exp)
self.time_blocks = nn.Sequential(
nn.Embedding.from_pretrained(emb),
nn.Linear(in_features=time_emb_dims, out_features=time_emb_dims_exp),
nn.SiLU(),
nn.Linear(in_features=time_emb_dims_exp, out_features=time_emb_dims_exp),
)
def forward(self, time):
print(self.emb)
print(time.shape,'*')
time = self.a(time)
print(time.shape,'**')
time = self.b(time)
print(time.shape,'***')
time = self.c(time)
print(time.shape,'*****')
return self.a(time)
class Transformer(nn.Module):
def __init__(self, embed_dim, src_vocab_size, target_vocab_size, seq_length,num_layers=2, expansion_factor=2, n_heads=4):
super(Transformer, self).__init__()
"""
Args:
embed_dim: dimension of embedding
src_vocab_size: vocabulary size of source
target_vocab_size: vocabulary size of target
seq_length : length of input sequence
num_layers: number of encoder layers
expansion_factor: factor which determines number of linear layers in feed forward layer
n_heads: number of heads in multihead attention
"""
self.target_vocab_size = target_vocab_size
self.Lin = nn.Linear(target_vocab_size,1)
self.encoder = TransformerEncoder(seq_length, src_vocab_size, embed_dim, num_layers=num_layers, expansion_factor=expansion_factor, n_heads=n_heads)
self.decoder = TransformerDecoder(target_vocab_size, embed_dim, seq_length, num_layers=num_layers, expansion_factor=expansion_factor, n_heads=n_heads)
time_emb_dims_exp = embed_dim * src_vocab_size
self.time_embeddings = SinusoidalPositionEmbeddings(time_emb_dims=embed_dim, time_emb_dims_exp=time_emb_dims_exp)
def make_trg_mask(self, trg):
"""
Args:
trg: target sequence
Returns:
trg_mask: target mask
"""
batch_size, trg_len = trg.shape
# returns the lower triangular part of matrix filled with ones
trg_mask = torch.tril(torch.ones((trg_len, trg_len))).expand(
batch_size, 1, trg_len, trg_len
)
return trg_mask
def decode(self,src,trg):
"""
for inference
Args:
src: input to encoder
trg: input to decoder
out:
out_labels : returns final prediction of sequence
"""
trg_mask = self.make_trg_mask(trg)
enc_out = self.encoder(src)
out_labels = []
batch_size,seq_len = src.shape[0],src.shape[1]
#outputs = torch.zeros(seq_len, batch_size, self.target_vocab_size)
out = trg
for i in range(seq_len): #10
out = self.decoder(out,enc_out,trg_mask) #bs x seq_len x vocab_dim
# taking the last token
out = out[:,-1,:]
out = out.argmax(-1)
out_labels.append(out.item())
out = torch.unsqueeze(out,axis=0)
return out_labels
def forward(self, src, trg):
"""
Args:
src: input to encoder
trg: input to decoder
out:
out: final vector which returns probabilities of each target word
"""
src = src.permute(0,2,1)
# print(src.shape)
trg_mask = self.make_trg_mask(trg)
enc_out = self.encoder(src)
# print(enc_out.shape,'enco ouput')
outputs = self.decoder(trg, enc_out, trg_mask).permute(0,2,1)
# print(outputs.shape)
return outputs
class Unet(nn.Module):
def __init__(self, embed_dim=128, src_vocab_size=2, target_vocab_size=80, seq_length=800,num_layers=2, expansion_factor=2, n_heads=4,KernelSizeList = [13,9,5]):
super().__init__()
self.embed_dim = embed_dim
self.src_vocab_size = src_vocab_size
self.target_vocab_size = target_vocab_size
self.seq_length = seq_length
self.num_layers=num_layers
self.expansion_factor=expansion_factor
self.n_heads=n_heads
self.KernelSizeList = KernelSizeList
self.t1 = Transformer(self.embed_dim,self.src_vocab_size,self.src_vocab_size,self.seq_length,self.num_layers,self.expansion_factor,self.n_heads)
self.seq_length1 = self.seq_length-self.KernelSizeList[0]+1
self.t2 = Transformer(self.embed_dim,self.src_vocab_size*4,self.src_vocab_size*4,self.seq_length1,self.num_layers,self.expansion_factor,self.n_heads)
self.seq_length2 = self.seq_length-self.KernelSizeList[0]+1-self.KernelSizeList[1]+1
self.t3 = Transformer(self.embed_dim,self.src_vocab_size*8,self.src_vocab_size*8,self.seq_length2,self.num_layers,self.expansion_factor,self.n_heads)
self.seq_length3 = self.seq_length-self.KernelSizeList[0]+1-self.KernelSizeList[1]+1-self.KernelSizeList[2]+1
self.t4 = Transformer(self.embed_dim,self.src_vocab_size*16,self.src_vocab_size*16,self.seq_length3,self.num_layers,self.expansion_factor,self.n_heads)
self.t5 = Transformer(self.embed_dim,self.src_vocab_size*16,self.src_vocab_size*16,self.seq_length3,self.num_layers,self.expansion_factor,self.n_heads)
self.t6 = Transformer(self.embed_dim,self.src_vocab_size*24,self.src_vocab_size*24,self.seq_length2,self.num_layers,self.expansion_factor,self.n_heads)
self.t7 = Transformer(self.embed_dim,self.src_vocab_size*32,self.src_vocab_size*32,self.seq_length1,self.num_layers,self.expansion_factor,self.n_heads)
self.t8 = Transformer(self.embed_dim,self.src_vocab_size*40,self.src_vocab_size*40,self.seq_length,self.num_layers,self.expansion_factor,self.n_heads)
self.Conv1 = nn.Conv1d(src_vocab_size,src_vocab_size*4,self.KernelSizeList[0])
self.Conv2 = nn.Conv1d(src_vocab_size*4,src_vocab_size*8,self.KernelSizeList[1])
self.Conv3 = nn.Conv1d(src_vocab_size*8,src_vocab_size*16,self.KernelSizeList[2])
self.Conv4 = nn.ConvTranspose1d(src_vocab_size*16,src_vocab_size*24,self.KernelSizeList[2])
self.Conv5 = nn.ConvTranspose1d(src_vocab_size*24,src_vocab_size*32,self.KernelSizeList[1])
self.Conv6 = nn.ConvTranspose1d(src_vocab_size*32,src_vocab_size*40,self.KernelSizeList[0])
self.flat = nn.Flatten()
time_emb_dims_exp = embed_dim * src_vocab_size
self.time_embeddings = SinusoidalPositionEmbeddings(time_emb_dims=embed_dim, time_emb_dims_exp=time_emb_dims_exp)
self.lin = nn.Linear(15680,1024)
def forward(self,x,text,time):
time_emb = self.time_embeddings(time)
# print(x.shape,'input')
trans1 = self.t1(x,text,time)
# print(trans1.shape,'trans op')
trans1 = self.Conv1(trans1)
# print(trans1.shape,'tueige')
trans1 = self.t2(F.normalize(trans1),text[:,:self.seq_length1],time)
# print(trans1.shape,'trans1 op')
trans1 = self.Conv2(trans1)
# print(trans1.shape,'after conc2')
trans1 = self.t3(F.normalize(trans1),text[:,:self.seq_length2])
# print(trans1.shape,'trans3 op')
trans1 = self.Conv3(trans1)
# print(trans1.shape,'after conc3')
trans1 = self.t4(F.normalize(trans1),text[:,:self.seq_length3])
LatentREp = self.lin(self.flat(trans1))
#decoder
trans1 = self.t5(F.normalize(trans1),text[:,:self.seq_length3])
# print(trans1.shape,'trans3 op117887')
trans1 = self.Conv4(trans1)
# print(trans1.shape,'after conc37777777777')
trans1 = self.t6(F.normalize(trans1),text[:,:self.seq_length2])
# print(trans1.shape,'trans3 op1111')
trans1 = self.Conv5(trans1)
# print(trans1.shape,'after conc355555555555')
trans1 = self.t7(F.normalize(trans1),text[:,:self.seq_length1])
# print(trans1.shape,'trans3 op1111111111')
trans1 = self.Conv6(trans1)
# print(trans1.shape,'after conc34444444444444444')
trans1 = self.t8(F.normalize(trans1),text[:,:self.seq_length])
# print(trans1.shape,'trans3 op333333333333333333333')
return LatentREp,trans1
# un = Unet(seq_length=512,KernelSizeList=[21,5,9])
# # un(torch.randint(70,(2,80,800)),torch.randint(10,(2,800)))
# from torchinfo import summary
# summary(un,input_data=[F.normalize(torch.randn((4,2,512))),F.normalize(torch.randn((4,512)))],depth=8)
##80 se neeche ki input value deni hai come what may nhi toh nn.embedding problem1
##embdim=256 se bhot zdaa reduce hojayega size so sorted < 50M for sure
##mha torch.nn wale s toh params double hogye the toh using their attention #mha wale mei masking ka smajh nhi aa rha vaise
##input type mei problem hai ki vo sirf randint leta hai ya normalised randn bas
##subsequent layers normlaised input leti hai why (L2 normlisation)