-
Notifications
You must be signed in to change notification settings - Fork 258
/
translation_transformer.py
409 lines (323 loc) Β· 17.3 KB
/
translation_transformer.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
"""
``nn.Transformer`` μ torchtextλ‘ μΈμ΄ λ²μνκΈ°
======================================================
μ΄ νν 리μΌμμλ,
- Transformer(νΈλμ€ν¬λ¨Έ)λ₯Ό μ¬μ©ν λ²μ λͺ¨λΈμ λ°λ₯λΆν° νμ΅νλ λ°©λ²μ λ°°μλ³΄κ² μ΅λλ€.
- `Multi30k <http://www.statmt.org/wmt16/multimodal-task.html#task1>`__
λ°μ΄ν°μ
μ μ¬μ©νμ¬ λ
μΌμ΄(German)λ₯Ό μμ΄(English)λ‘ λ²μνλ λͺ¨λΈμ νμ΅ν΄λ³΄κ² μ΅λλ€.
"""
######################################################################
# λ°μ΄ν° ꡬνκ³ μ²λ¦¬νκΈ°
# ----------------------------
#
# `torchtext λΌμ΄λΈλ¬λ¦¬ <https://pytorch.org/text/stable/>`__\ μλ μΈμ΄ λ²μ λͺ¨λΈμ
# μμ±νκΈ° μν λ°μ΄ν°μ
μ μ½κ² λ§λ€ μ μλ λꡬλ€μ΄ μμ΅λλ€.
# μ΄ νν 리μΌμμλ torchtextμ λ΄μ₯(inbuilt) λ°μ΄ν°μ
μ μ΄λ»κ² μ¬μ©νκ³ ,
# μμ(raw) ν
μ€νΈ λ¬Έμ₯μ ν ν°ν(tokenize)νκ³ , ν ν°μ ν
μλ‘ μμΉννλ λ°©λ²μ
# μ΄ν΄λ³΄κ² μ΅λλ€. μΆλ°μ΄(source)-λμ°©μ΄(target) μμ(raw) λ¬Έμ₯μ μμ±νκΈ° μν΄μλ
# `torchtext λΌμ΄λΈλ¬λ¦¬μ Multi30k λ°μ΄ν°μ
<https://pytorch.org/text/stable/datasets.html#multi30k>`__
# μ μ¬μ©νκ² μ΅λλ€.
#
# torchtext λ°μ΄ν°μ
μ μ κ·ΌνκΈ° μ μ, https://github.com/pytorch/data μ μ°Έκ³ νμ¬ torchdataλ₯Ό μ€μΉνμκΈ° λ°λλλ€.
#
from torchtext.data.utils import get_tokenizer
from torchtext.vocab import build_vocab_from_iterator
from torchtext.datasets import multi30k, Multi30k
from typing import Iterable, List
# μλ³Έ λ°μ΄ν°μ λ§ν¬κ° λμνμ§ μμΌλ―λ‘ λ°μ΄ν°μ
μ URLμ μμ ν΄μΌ ν©λλ€.
# λ μμΈν λ΄μ©μ https://github.com/pytorch/text/issues/1756#issuecomment-1163664163 μ μ°Έκ³ ν΄μ£ΌμΈμ.
multi30k.URL["train"] = "https://raw.githubusercontent.com/neychev/small_DL_repo/master/datasets/Multi30k/training.tar.gz"
multi30k.URL["valid"] = "https://raw.githubusercontent.com/neychev/small_DL_repo/master/datasets/Multi30k/validation.tar.gz"
SRC_LANGUAGE = 'de'
TGT_LANGUAGE = 'en'
# Place-holders
token_transform = {}
vocab_transform = {}
###################################################################################
# μΆλ°μ΄(source)μ λͺ©μ μ΄(target)μ ν ν¬λμ΄μ (tokenizer)λ₯Ό μμ±ν©λλ€.
# μλ νμ μ¬ν(dependency)μ λͺ¨λ μ€μΉν΄μ£ΌμΈμ.
#
# .. code-block:: python
#
# pip install -U torchdata
# pip install -U spacy
# python -m spacy download en_core_web_sm
# python -m spacy download de_core_news_sm
token_transform[SRC_LANGUAGE] = get_tokenizer('spacy', language='de_core_news_sm')
token_transform[TGT_LANGUAGE] = get_tokenizer('spacy', language='en_core_web_sm')
# ν ν° λͺ©λ‘μ μμ±νκΈ° μν ν¬νΌ(helper) ν¨μ
def yield_tokens(data_iter: Iterable, language: str) -> List[str]:
language_index = {SRC_LANGUAGE: 0, TGT_LANGUAGE: 1}
for data_sample in data_iter:
yield token_transform[language](data_sample[language_index[language]])
# νΉμ κΈ°νΈ(symbol)μ μΈλ±μ€λ₯Ό μ μν©λλ€
UNK_IDX, PAD_IDX, BOS_IDX, EOS_IDX = 0, 1, 2, 3
# ν ν°λ€μ΄ μ΄νμ§(vocab)μ μΈλ±μ€ μμλλ‘ μ μ½μ
λμ΄ μλμ§ νμΈν©λλ€
special_symbols = ['<unk>', '<pad>', '<bos>', '<eos>']
for ln in [SRC_LANGUAGE, TGT_LANGUAGE]:
# νμ΅μ© λ°μ΄ν° λ°λ³΅μ(iterator)
train_iter = Multi30k(split='train', language_pair=(SRC_LANGUAGE, TGT_LANGUAGE))
# torchtextμ Vocab(μ΄νμ§) κ°μ²΄ μμ±
vocab_transform[ln] = build_vocab_from_iterator(yield_tokens(train_iter, ln),
min_freq=1,
specials=special_symbols,
special_first=True)
# ``UNK_IDX`` λ₯Ό κΈ°λ³Έ μΈλ±μ€λ‘ μ€μ ν©λλ€. μ΄ μΈλ±μ€λ ν ν°μ μ°Ύμ§ λͺ»νλ κ²½μ°μ λ°νλ©λλ€.
# λ§μ½ κΈ°λ³Έ μΈλ±μ€λ₯Ό μ€μ νμ§ μμΌλ©΄ μ΄νμ§(Vocabulary)μμ ν ν°μ μ°Ύμ§ λͺ»νλ κ²½μ°
# ``RuntimeError`` κ° λ°μν©λλ€.
for ln in [SRC_LANGUAGE, TGT_LANGUAGE]:
vocab_transform[ln].set_default_index(UNK_IDX)
######################################################################
# Transformerλ₯Ό μ¬μ©ν μνμ€-ν¬-μνμ€(Seq2Seq) μ κ²½λ§
# ------------------------------------------------------------
#
# Transformer(νΈλμ€ν¬λ¨Έ)λ κΈ°κ³λ²μ μμ
(task)μ μν΄
# `"Attention is all you need" <https://papers.nips.cc/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf>`__
# λ
Όλ¬Έμ μκ°λ Seq2Seq λͺ¨λΈμ
λλ€.
# μλμμ Transformerλ₯Ό μ¬μ©ν Seq2Seq μ κ²½λ§μ λ§λ€μ΄λ³΄κ² μ΅λλ€.
# μ κ²½λ§μ μΈ λΆλΆμΌλ‘ ꡬμ±λλλ°, 첫λ²μ§Έ λΆλΆμ μλ² λ© κ³μΈ΅(embedding layer)μ
λλ€.
# μ΄ κ³μΈ΅μ μ
λ ₯ μΈλ±μ€μ ν
μλ₯Ό μ
λ ₯ μλ² λ©μ ν΄λΉνλ ν
μλ‘ λ³νν©λλ€.
# μ΄λ¬ν μλ² λ©μ μ
λ ₯ ν ν°μ μμΉ μ 보(position information)λ₯Ό λͺ¨λΈμ μ λ¬νκΈ° μν΄
# μμΉ μΈμ½λ©(positional encoding)μ μΆκ°ν©λλ€.
# λλ²μ§Έ λΆλΆμ μ€μ `Transformer <https://pytorch.org/docs/stable/generated/torch.nn.Transformer.html>`__ λͺ¨λΈμ
λλ€.
# λ§μ§λ§μΌλ‘ Transformer λͺ¨λΈμ μΆλ ₯μ μ ν κ³μΈ΅μ ν΅κ³ΌμμΌ λμ°©μ΄μ κ° ν ν°μ λν μ κ·νλμ§ μμ
# νλ₯ (un-normalized probability)λ‘ μ 곡ν©λλ€.
#
from torch import Tensor
import torch
import torch.nn as nn
from torch.nn import Transformer
import math
DEVICE = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# λ¨μ΄ μμ κ°λ
(notion)μ ν ν° μλ² λ©μ λμ
νκΈ° μν μμΉ μΈμ½λ©(positional encoding)μ μν ν¬νΌ λͺ¨λ(Module)
class PositionalEncoding(nn.Module):
def __init__(self,
emb_size: int,
dropout: float,
maxlen: int = 5000):
super(PositionalEncoding, self).__init__()
den = torch.exp(- torch.arange(0, emb_size, 2)* math.log(10000) / emb_size)
pos = torch.arange(0, maxlen).reshape(maxlen, 1)
pos_embedding = torch.zeros((maxlen, emb_size))
pos_embedding[:, 0::2] = torch.sin(pos * den)
pos_embedding[:, 1::2] = torch.cos(pos * den)
pos_embedding = pos_embedding.unsqueeze(-2)
self.dropout = nn.Dropout(dropout)
self.register_buffer('pos_embedding', pos_embedding)
def forward(self, token_embedding: Tensor):
return self.dropout(token_embedding + self.pos_embedding[:token_embedding.size(0), :])
# μ
λ ₯ μΈλ±μ€μ ν
μλ₯Ό ν΄λΉνλ ν ν° μλ² λ©μ ν
μλ‘ λ³ννκΈ° μν ν¬νΌ λͺ¨λ(Module)
class TokenEmbedding(nn.Module):
def __init__(self, vocab_size: int, emb_size):
super(TokenEmbedding, self).__init__()
self.embedding = nn.Embedding(vocab_size, emb_size)
self.emb_size = emb_size
def forward(self, tokens: Tensor):
return self.embedding(tokens.long()) * math.sqrt(self.emb_size)
# Seq2Seq μ κ²½λ§
class Seq2SeqTransformer(nn.Module):
def __init__(self,
num_encoder_layers: int,
num_decoder_layers: int,
emb_size: int,
nhead: int,
src_vocab_size: int,
tgt_vocab_size: int,
dim_feedforward: int = 512,
dropout: float = 0.1):
super(Seq2SeqTransformer, self).__init__()
self.transformer = Transformer(d_model=emb_size,
nhead=nhead,
num_encoder_layers=num_encoder_layers,
num_decoder_layers=num_decoder_layers,
dim_feedforward=dim_feedforward,
dropout=dropout)
self.generator = nn.Linear(emb_size, tgt_vocab_size)
self.src_tok_emb = TokenEmbedding(src_vocab_size, emb_size)
self.tgt_tok_emb = TokenEmbedding(tgt_vocab_size, emb_size)
self.positional_encoding = PositionalEncoding(
emb_size, dropout=dropout)
def forward(self,
src: Tensor,
trg: Tensor,
src_mask: Tensor,
tgt_mask: Tensor,
src_padding_mask: Tensor,
tgt_padding_mask: Tensor,
memory_key_padding_mask: Tensor):
src_emb = self.positional_encoding(self.src_tok_emb(src))
tgt_emb = self.positional_encoding(self.tgt_tok_emb(trg))
outs = self.transformer(src_emb, tgt_emb, src_mask, tgt_mask, None,
src_padding_mask, tgt_padding_mask, memory_key_padding_mask)
return self.generator(outs)
def encode(self, src: Tensor, src_mask: Tensor):
return self.transformer.encoder(self.positional_encoding(
self.src_tok_emb(src)), src_mask)
def decode(self, tgt: Tensor, memory: Tensor, tgt_mask: Tensor):
return self.transformer.decoder(self.positional_encoding(
self.tgt_tok_emb(tgt)), memory,
tgt_mask)
######################################################################
# νμ΅νλ λμ, λͺ¨λΈμ΄ μμΈ‘ν λ μ λ΅(μ΄ν μΆννλ λ¨μ΄)μ λ³΄μ§ λͺ»νλλ‘ νλ
# νμ λ¨μ΄ λ§μ€ν¬(subsequent word mask)κ° νμν©λλ€. λν, μΆλ°μ΄μ λμ°©μ΄μ ν¨λ©(padding) ν ν°λ€
# λν μ¨κ²¨μΌ ν©λλ€. μλμ λ κ°μ§ λͺ¨λλ₯Ό μ²λ¦¬ν ν¨μλ₯Ό μ μν΄λ³΄κ² μ΅λλ€.
#
def generate_square_subsequent_mask(sz):
mask = (torch.triu(torch.ones((sz, sz), device=DEVICE)) == 1).transpose(0, 1)
mask = mask.float().masked_fill(mask == 0, float('-inf')).masked_fill(mask == 1, float(0.0))
return mask
def create_mask(src, tgt):
src_seq_len = src.shape[0]
tgt_seq_len = tgt.shape[0]
tgt_mask = generate_square_subsequent_mask(tgt_seq_len)
src_mask = torch.zeros((src_seq_len, src_seq_len),device=DEVICE).type(torch.bool)
src_padding_mask = (src == PAD_IDX).transpose(0, 1)
tgt_padding_mask = (tgt == PAD_IDX).transpose(0, 1)
return src_mask, tgt_mask, src_padding_mask, tgt_padding_mask
######################################################################
# μ΄μ λͺ¨λΈμ 맀κ°λ³μλ₯Ό μ μνκ³ κ°μ²΄λ₯Ό μμ±(instantiate)ν΄λ³΄κ² μ΅λλ€.
# μλμ²λΌ νμ΅ λ¨κ³μμ μ¬μ©ν μμ€ ν¨μ(loss function)λ₯Ό κ΅μ°¨ μνΈλ‘νΌ μμ€(cross-entropy loss)λ‘ μ μνκ³ ,
# μ΅ν°λ§μ΄μ (optimizer)λ μ μν©λλ€.
#
torch.manual_seed(0)
SRC_VOCAB_SIZE = len(vocab_transform[SRC_LANGUAGE])
TGT_VOCAB_SIZE = len(vocab_transform[TGT_LANGUAGE])
EMB_SIZE = 512
NHEAD = 8
FFN_HID_DIM = 512
BATCH_SIZE = 128
NUM_ENCODER_LAYERS = 3
NUM_DECODER_LAYERS = 3
transformer = Seq2SeqTransformer(NUM_ENCODER_LAYERS, NUM_DECODER_LAYERS, EMB_SIZE,
NHEAD, SRC_VOCAB_SIZE, TGT_VOCAB_SIZE, FFN_HID_DIM)
for p in transformer.parameters():
if p.dim() > 1:
nn.init.xavier_uniform_(p)
transformer = transformer.to(DEVICE)
loss_fn = torch.nn.CrossEntropyLoss(ignore_index=PAD_IDX)
optimizer = torch.optim.Adam(transformer.parameters(), lr=0.0001, betas=(0.9, 0.98), eps=1e-9)
######################################################################
# λμ‘°(Collation)
# -----------------
#
# μμ ``λ°μ΄ν° ꡬνκ³ μ²λ¦¬νκΈ°`` μ₯μμ λ΄€λ―μ΄, λ°μ΄ν° λ°λ³΅μ(iterator)λ μμ λ¬Έμμ΄μ μμ μμ±ν©λλ€.
# μ΄ λ¬Έμμ΄ μλ€μ μ΄μ μ μ μν ``Seq2Seq`` μ κ²½λ§μμ μ²λ¦¬ν μ μλλ‘ ν
μ λ¬Άμ(batched tensor)μΌλ‘ λ³νν΄μΌ ν©λλ€.
# μ΄μ μμ λ¬Έμμ΄λ€μ λ¬Άμ(batch)μ ν
μ λ¬ΆμμΌλ‘ λ³ννμ¬ λͺ¨λΈμ μ§μ μ λ¬ν μ μλλ‘ νλ λμμ΄(collate) ν¨μλ₯Ό
# μ μν΄λ³΄κ² μ΅λλ€.
#
from torch.nn.utils.rnn import pad_sequence
# μμ°¨μ μΈ μμ
λ€μ νλλ‘ λ¬Άλ ν¬νΌ ν¨μ
def sequential_transforms(*transforms):
def func(txt_input):
for transform in transforms:
txt_input = transform(txt_input)
return txt_input
return func
# BOS/EOSλ₯Ό μΆκ°νκ³ μ
λ ₯ μμ(sequence) μΈλ±μ€μ λν ν
μλ₯Ό μμ±νλ ν¨μ
def tensor_transform(token_ids: List[int]):
return torch.cat((torch.tensor([BOS_IDX]),
torch.tensor(token_ids),
torch.tensor([EOS_IDX])))
# μΆλ°μ΄(src)μ λμ°©μ΄(tgt) μμ λ¬Έμμ΄λ€μ ν
μ μΈλ±μ€λ‘ λ³ννλ λ³ν(transform)
text_transform = {}
for ln in [SRC_LANGUAGE, TGT_LANGUAGE]:
text_transform[ln] = sequential_transforms(token_transform[ln], # ν ν°ν(Tokenization)
vocab_transform[ln], # μμΉν(Numericalization)
tensor_transform) # BOS/EOSλ₯Ό μΆκ°νκ³ ν
μλ₯Ό μμ±
# λ°μ΄ν°λ₯Ό ν
μλ‘ μ‘°ν©(collate)νλ ν¨μ
def collate_fn(batch):
src_batch, tgt_batch = [], []
for src_sample, tgt_sample in batch:
src_batch.append(text_transform[SRC_LANGUAGE](src_sample.rstrip("\n")))
tgt_batch.append(text_transform[TGT_LANGUAGE](tgt_sample.rstrip("\n")))
src_batch = pad_sequence(src_batch, padding_value=PAD_IDX)
tgt_batch = pad_sequence(tgt_batch, padding_value=PAD_IDX)
return src_batch, tgt_batch
######################################################################
# κ° μν(epoch)λ§λ€ νΈμΆν νμ΅ λ° κ²μ¦(evaluation) λ¨κ³λ₯Ό μ μν΄λ³΄κ² μ΅λλ€.
#
from torch.utils.data import DataLoader
def train_epoch(model, optimizer):
model.train()
losses = 0
train_iter = Multi30k(split='train', language_pair=(SRC_LANGUAGE, TGT_LANGUAGE))
train_dataloader = DataLoader(train_iter, batch_size=BATCH_SIZE, collate_fn=collate_fn)
for src, tgt in train_dataloader:
src = src.to(DEVICE)
tgt = tgt.to(DEVICE)
tgt_input = tgt[:-1, :]
src_mask, tgt_mask, src_padding_mask, tgt_padding_mask = create_mask(src, tgt_input)
logits = model(src, tgt_input, src_mask, tgt_mask,src_padding_mask, tgt_padding_mask, src_padding_mask)
optimizer.zero_grad()
tgt_out = tgt[1:, :]
loss = loss_fn(logits.reshape(-1, logits.shape[-1]), tgt_out.reshape(-1))
loss.backward()
optimizer.step()
losses += loss.item()
return losses / len(list(train_dataloader))
def evaluate(model):
model.eval()
losses = 0
val_iter = Multi30k(split='valid', language_pair=(SRC_LANGUAGE, TGT_LANGUAGE))
val_dataloader = DataLoader(val_iter, batch_size=BATCH_SIZE, collate_fn=collate_fn)
for src, tgt in val_dataloader:
src = src.to(DEVICE)
tgt = tgt.to(DEVICE)
tgt_input = tgt[:-1, :]
src_mask, tgt_mask, src_padding_mask, tgt_padding_mask = create_mask(src, tgt_input)
logits = model(src, tgt_input, src_mask, tgt_mask,src_padding_mask, tgt_padding_mask, src_padding_mask)
tgt_out = tgt[1:, :]
loss = loss_fn(logits.reshape(-1, logits.shape[-1]), tgt_out.reshape(-1))
losses += loss.item()
return losses / len(list(val_dataloader))
######################################################################
# μ΄μ λͺ¨λΈ νμ΅μ μν λͺ¨λ μμκ° μ€λΉλμμ΅λλ€. νμ΅μ ν΄λ³΄κ² μ΅λλ€!
#
from timeit import default_timer as timer
NUM_EPOCHS = 18
for epoch in range(1, NUM_EPOCHS+1):
start_time = timer()
train_loss = train_epoch(transformer, optimizer)
end_time = timer()
val_loss = evaluate(transformer)
print((f"Epoch: {epoch}, Train loss: {train_loss:.3f}, Val loss: {val_loss:.3f}, "f"Epoch time = {(end_time - start_time):.3f}s"))
# νμ(greedy) μκ³ λ¦¬μ¦μ μ¬μ©νμ¬ μΆλ ₯ μμ(sequence)λ₯Ό μμ±νλ ν¨μ
def greedy_decode(model, src, src_mask, max_len, start_symbol):
src = src.to(DEVICE)
src_mask = src_mask.to(DEVICE)
memory = model.encode(src, src_mask)
ys = torch.ones(1, 1).fill_(start_symbol).type(torch.long).to(DEVICE)
for i in range(max_len-1):
memory = memory.to(DEVICE)
tgt_mask = (generate_square_subsequent_mask(ys.size(0))
.type(torch.bool)).to(DEVICE)
out = model.decode(ys, memory, tgt_mask)
out = out.transpose(0, 1)
prob = model.generator(out[:, -1])
_, next_word = torch.max(prob, dim=1)
next_word = next_word.item()
ys = torch.cat([ys,
torch.ones(1, 1).type_as(src.data).fill_(next_word)], dim=0)
if next_word == EOS_IDX:
break
return ys
# μ
λ ₯ λ¬Έμ₯μ λμ°©μ΄λ‘ λ²μνλ ν¨μ
def translate(model: torch.nn.Module, src_sentence: str):
model.eval()
src = text_transform[SRC_LANGUAGE](src_sentence).view(-1, 1)
num_tokens = src.shape[0]
src_mask = (torch.zeros(num_tokens, num_tokens)).type(torch.bool)
tgt_tokens = greedy_decode(
model, src, src_mask, max_len=num_tokens + 5, start_symbol=BOS_IDX).flatten()
return " ".join(vocab_transform[TGT_LANGUAGE].lookup_tokens(list(tgt_tokens.cpu().numpy()))).replace("<bos>", "").replace("<eos>", "")
######################################################################
#
print(translate(transformer, "Eine Gruppe von Menschen steht vor einem Iglu ."))
######################################################################
# μ°Έκ³ μλ£
# ----------
#
# 1. Attention is all you need λ
Όλ¬Έ.
# https://papers.nips.cc/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf
# 2. Transformerμ λν μ€λͺ
. https://nlp.seas.harvard.edu/2018/04/03/attention.html#positional-encoding
#