Modelo transformador para la comprensión del lenguaje

Ver en TensorFlow.org Ejecutar en Google Colab Ver fuente en GitHubDescargar cuaderno

Este tutorial trenes al modelo de Transformer para traducir un conjunto de datos portugués al Inglés . Este es un ejemplo avanzado que supone el conocimiento de generación de texto y la atención .

La idea central detrás del modelo de Transformer es la auto-atención, la habilidad de asistir a diferentes posiciones de la secuencia de entrada para calcular una representación de esa secuencia. Transformador crea pilas de capas de auto-atención y se explica a continuación en las secciones Scaled atención producto escalar y la atención Multi-cabeza.

Una de tamaño variable mangos modelo transformador de entrada usando pilas de capas de auto-atención en lugar de RNNs o CNNs . Esta arquitectura general tiene una serie de ventajas:

  • No hace suposiciones sobre las relaciones temporales / espaciales entre los datos. Esto es ideal para el procesamiento de un conjunto de objetos (por ejemplo, unidades de StarCraft ).
  • Las salidas de capa se pueden calcular en paralelo, en lugar de una serie como un RNN.
  • Artículos distantes pueden afectar la producción de la otra sin pasar a través de muchos RNN-etapas, o capas de convolución (véase Scene Memory transformador por ejemplo).
  • Puede aprender dependencias de largo alcance. Este es un desafío en muchas tareas de secuencia.

Las desventajas de esta arquitectura son:

  • Por una serie de tiempo, la salida de un paso de tiempo se calcula a partir de toda la historia en lugar de solo las entradas y del estado escondido actual. Esto puede ser menos eficiente.
  • Si la entrada tiene una relación temporal / espacial, como texto, se debe añadir un poco de codificación posicional o el modelo será efectivamente ver una bolsa de palabras.

Después de entrenar el modelo en este cuaderno, podrá ingresar una oración en portugués y devolver la traducción al inglés.

Mapa de calor de atención

Configuración

pip install -q tensorflow_datasets
pip install -q -U tensorflow-text
import collections
import logging
import os
import pathlib
import re
import string
import sys
import time

import numpy as np
import matplotlib.pyplot as plt

import tensorflow_datasets as tfds
import tensorflow_text as text
import tensorflow as tf
2021-07-08 11:06:56.882101: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcudart.so.11.0
logging.getLogger('tensorflow').setLevel(logging.ERROR)  # suppress warnings

Descarga el conjunto de datos

Uso TensorFlow conjuntos de datos para cargar el conjunto de datos en portugués-Inglés del proyecto abierto de traducción TED Talks .

Este conjunto de datos contiene aproximadamente 50000 ejemplos de entrenamiento, 1100 ejemplos de validación y 2000 ejemplos de prueba.

examples, metadata = tfds.load('ted_hrlr_translate/pt_to_en', with_info=True,
                               as_supervised=True)
train_examples, val_examples = examples['train'], examples['validation']
2021-07-08 11:07:01.891325: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcuda.so.1
2021-07-08 11:07:02.567479: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-07-08 11:07:02.568175: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1733] Found device 0 with properties: 
pciBusID: 0000:00:05.0 name: NVIDIA Tesla V100-SXM2-16GB computeCapability: 7.0
coreClock: 1.53GHz coreCount: 80 deviceMemorySize: 15.78GiB deviceMemoryBandwidth: 836.37GiB/s
2021-07-08 11:07:02.568215: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcudart.so.11.0
2021-07-08 11:07:02.571596: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcublas.so.11
2021-07-08 11:07:02.571723: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcublasLt.so.11
2021-07-08 11:07:02.572867: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcufft.so.10
2021-07-08 11:07:02.573248: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcurand.so.10
2021-07-08 11:07:02.574372: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcusolver.so.11
2021-07-08 11:07:02.575325: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcusparse.so.11
2021-07-08 11:07:02.575514: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcudnn.so.8
2021-07-08 11:07:02.575612: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-07-08 11:07:02.576333: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-07-08 11:07:02.576942: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1871] Adding visible gpu devices: 0
2021-07-08 11:07:02.577664: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 AVX512F FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2021-07-08 11:07:02.578246: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-07-08 11:07:02.578904: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1733] Found device 0 with properties: 
pciBusID: 0000:00:05.0 name: NVIDIA Tesla V100-SXM2-16GB computeCapability: 7.0
coreClock: 1.53GHz coreCount: 80 deviceMemorySize: 15.78GiB deviceMemoryBandwidth: 836.37GiB/s
2021-07-08 11:07:02.578980: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-07-08 11:07:02.579656: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-07-08 11:07:02.580279: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1871] Adding visible gpu devices: 0
2021-07-08 11:07:02.580323: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcudart.so.11.0
2021-07-08 11:07:03.190409: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1258] Device interconnect StreamExecutor with strength 1 edge matrix:
2021-07-08 11:07:03.190462: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1264]      0 
2021-07-08 11:07:03.190470: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1277] 0:   N 
2021-07-08 11:07:03.190677: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-07-08 11:07:03.191419: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-07-08 11:07:03.192079: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-07-08 11:07:03.192729: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1418] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 14646 MB memory) -> physical GPU (device: 0, name: NVIDIA Tesla V100-SXM2-16GB, pci bus id: 0000:00:05.0, compute capability: 7.0)

El tf.data.Dataset objeto devuelto por TensorFlow conjuntos de datos produce pares de ejemplos de texto:

for pt_examples, en_examples in train_examples.batch(3).take(1):
  for pt in pt_examples.numpy():
    print(pt.decode('utf-8'))

  print()

  for en in en_examples.numpy():
    print(en.decode('utf-8'))
2021-07-08 11:07:03.295521: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:176] None of the MLIR Optimization Passes are enabled (registered 2)
2021-07-08 11:07:03.296195: I tensorflow/core/platform/profile_utils/cpu_utils.cc:114] CPU Frequency: 2000170000 Hz
e quando melhoramos a procura , tiramos a \xfanica vantagem da impress\xe3o , que \xe9 a serendipidade .
mas e se estes fatores fossem ativos ?
mas eles n\xe3o tinham a curiosidade de me testar .

and when you improve searchability , you actually take away the one advantage of print , which is serendipity .
but what if it were active ?
but they did n't test for curiosity .

Tokenización y destokenización de texto

No puede entrenar un modelo directamente en el texto. El texto debe convertirse primero a alguna representación numérica. Normalmente, convierte el texto en secuencias de ID de token, que se utilizan como índices en una incrustación.

Una aplicación muy popular se demuestra en la palabra parcial tokenizer tutorial se basa tokenizers palabra parcial ( text.BertTokenizer ) optimizados para este conjunto de datos y los exporta en un saved_model .

Descargar y descomprimir e importar el saved_model :

model_name = "ted_hrlr_translate_pt_en_converter"
tf.keras.utils.get_file(
    f"{model_name}.zip",
    f"https://storage.googleapis.com/download.tensorflow.org/models/{model_name}.zip",
    cache_dir='.', cache_subdir='', extract=True
)
Downloading data from https://storage.googleapis.com/download.tensorflow.org/models/ted_hrlr_translate_pt_en_converter.zip
188416/184801 [==============================] - 0s 0us/step
'./ted_hrlr_translate_pt_en_converter.zip'
tokenizers = tf.saved_model.load(model_name)

El tf.saved_model contiene dos tokenizers de texto, uno para Inglés y otro para el portugués. Ambos tienen los mismos métodos:

[item for item in dir(tokenizers.en) if not item.startswith('_')]
['detokenize',
 'get_reserved_tokens',
 'get_vocab_path',
 'get_vocab_size',
 'lookup',
 'tokenize',
 'tokenizer',
 'vocab']

El tokenize método convierte un lote de cadenas a un acolchado-lote de ID de testigo. Este método divide la puntuación, las minúsculas y normaliza unicode la entrada antes de la tokenización. Esa estandarización no es visible aquí porque los datos de entrada ya están estandarizados.

for en in en_examples.numpy():
  print(en.decode('utf-8'))
and when you improve searchability , you actually take away the one advantage of print , which is serendipity .
but what if it were active ?
but they did n't test for curiosity .
encoded = tokenizers.en.tokenize(en_examples)

for row in encoded.to_list():
  print(row)
[2, 72, 117, 79, 1259, 1491, 2362, 13, 79, 150, 184, 311, 71, 103, 2308, 74, 2679, 13, 148, 80, 55, 4840, 1434, 2423, 540, 15, 3]
[2, 87, 90, 107, 76, 129, 1852, 30, 3]
[2, 87, 83, 149, 50, 9, 56, 664, 85, 2512, 15, 3]

Los detokenize intentos método para convertir estos ID de testigo volver al texto legible por humanos:

round_trip = tokenizers.en.detokenize(encoded)
for line in round_trip.numpy():
  print(line.decode('utf-8'))
and when you improve searchability , you actually take away the one advantage of print , which is serendipity .
but what if it were active ?
but they did n ' t test for curiosity .

El nivel inferior lookup conversos método de token-IDs a texto token:

tokens = tokenizers.en.lookup(encoded)
tokens
<tf.RaggedTensor [[b'[START]', b'and', b'when', b'you', b'improve', b'search', b'##ability', b',', b'you', b'actually', b'take', b'away', b'the', b'one', b'advantage', b'of', b'print', b',', b'which', b'is', b's', b'##ere', b'##nd', b'##ip', b'##ity', b'.', b'[END]'], [b'[START]', b'but', b'what', b'if', b'it', b'were', b'active', b'?', b'[END]'], [b'[START]', b'but', b'they', b'did', b'n', b"'", b't', b'test', b'for', b'curiosity', b'.', b'[END]']]>

Aquí puede ver el aspecto de "subpalabra" de los tokenizadores. La palabra "capacidad de búsqueda" se descompone en "búsqueda ## capacidad" y la palabra "serindipity" en "s ## ere ## nd ## ip ## ity"

Configurar canalización de entrada

Para crear una canalización de entrada adecuada para el entrenamiento, aplicará algunas transformaciones al conjunto de datos.

Esta función se utilizará para codificar los lotes de texto sin formato:

def tokenize_pairs(pt, en):
    pt = tokenizers.pt.tokenize(pt)
    # Convert from ragged to dense, padding with zeros.
    pt = pt.to_tensor()

    en = tokenizers.en.tokenize(en)
    # Convert from ragged to dense, padding with zeros.
    en = en.to_tensor()
    return pt, en

Aquí hay una canalización de entrada simple que procesa, baraja y agrupa los datos:

BUFFER_SIZE = 20000
BATCH_SIZE = 64
def make_batches(ds):
  return (
      ds
      .cache()
      .shuffle(BUFFER_SIZE)
      .batch(BATCH_SIZE)
      .map(tokenize_pairs, num_parallel_calls=tf.data.AUTOTUNE)
      .prefetch(tf.data.AUTOTUNE))


train_batches = make_batches(train_examples)
val_batches = make_batches(val_examples)

Codificación posicional

Dado que este modelo no contiene ninguna recurrencia o convolución, se agrega codificación posicional para darle al modelo alguna información sobre la posición relativa de las palabras en la oración.

El vector de codificación posicional se agrega al vector de incrustación. Las incrustaciones representan un token en un espacio d-dimensional donde los tokens con un significado similar estarán más cerca unos de otros. Pero las incrustaciones no codifican la posición relativa de las palabras en una oración. Así que después de la adición de la codificación posicional, palabras estarán más cerca el uno al otro sobre la base de la similitud de su significado y de su posición en la oración, en el espacio d-dimensional.

La fórmula para calcular la codificación posicional es la siguiente:

$$\Large{PE_{(pos, 2i)} = \sin(pos / 10000^{2i / d_{model} })} $$
$$\Large{PE_{(pos, 2i+1)} = \cos(pos / 10000^{2i / d_{model} })} $$
def get_angles(pos, i, d_model):
  angle_rates = 1 / np.power(10000, (2 * (i//2)) / np.float32(d_model))
  return pos * angle_rates
def positional_encoding(position, d_model):
  angle_rads = get_angles(np.arange(position)[:, np.newaxis],
                          np.arange(d_model)[np.newaxis, :],
                          d_model)

  # apply sin to even indices in the array; 2i
  angle_rads[:, 0::2] = np.sin(angle_rads[:, 0::2])

  # apply cos to odd indices in the array; 2i+1
  angle_rads[:, 1::2] = np.cos(angle_rads[:, 1::2])

  pos_encoding = angle_rads[np.newaxis, ...]

  return tf.cast(pos_encoding, dtype=tf.float32)
n, d = 2048, 512
pos_encoding = positional_encoding(n, d)
print(pos_encoding.shape)
pos_encoding = pos_encoding[0]

# Juggle the dimensions for the plot
pos_encoding = tf.reshape(pos_encoding, (n, d//2, 2))
pos_encoding = tf.transpose(pos_encoding, (2, 1, 0))
pos_encoding = tf.reshape(pos_encoding, (d, n))

plt.pcolormesh(pos_encoding, cmap='RdBu')
plt.ylabel('Depth')
plt.xlabel('Position')
plt.colorbar()
plt.show()
(1, 2048, 512)

png

Enmascaramiento

Enmascare todas las fichas de pad en el lote de secuencia. Asegura que el modelo no trate el relleno como entrada. La máscara indica que el valor de la almohadilla de 0 está presente: se da salida a un 1 en esos lugares, y un 0 de otro modo.

def create_padding_mask(seq):
  seq = tf.cast(tf.math.equal(seq, 0), tf.float32)

  # add extra dimensions to add the padding
  # to the attention logits.
  return seq[:, tf.newaxis, tf.newaxis, :]  # (batch_size, 1, 1, seq_len)
x = tf.constant([[7, 6, 0, 0, 1], [1, 2, 3, 0, 0], [0, 0, 0, 4, 5]])
create_padding_mask(x)
<tf.Tensor: shape=(3, 1, 1, 5), dtype=float32, numpy=
array([[[[0., 0., 1., 1., 0.]]],


       [[[0., 0., 0., 1., 1.]]],


       [[[1., 1., 1., 0., 0.]]]], dtype=float32)>

La máscara de anticipación se utiliza para enmascarar los tokens futuros en una secuencia. En otras palabras, la máscara indica qué entradas no deben usarse.

Esto significa que para predecir la tercera palabra, solo se utilizarán la primera y la segunda palabra. De manera similar, para predecir la cuarta palabra, solo se usarán la primera, la segunda y la tercera palabra y así sucesivamente.

def create_look_ahead_mask(size):
  mask = 1 - tf.linalg.band_part(tf.ones((size, size)), -1, 0)
  return mask  # (seq_len, seq_len)
x = tf.random.uniform((1, 3))
temp = create_look_ahead_mask(x.shape[1])
temp
<tf.Tensor: shape=(3, 3), dtype=float32, numpy=
array([[0., 1., 1.],
       [0., 0., 1.],
       [0., 0., 0.]], dtype=float32)>

Atención de productos escalados

scaled_dot_product_attention

La función de atención que utiliza el transformador toma tres entradas: Q (consulta), K (tecla), V (valor). La ecuación utilizada para calcular los pesos de atención es:

$$\Large{Attention(Q, K, V) = softmax_k\left(\frac{QK^T}{\sqrt{d_k} }\right) V} $$

La atención del producto escalado se escala por un factor de raíz cuadrada de la profundidad. Esto se hace porque para valores grandes de profundidad, el producto escalar aumenta en magnitud empujando la función softmax donde tiene pequeños gradientes dando como resultado un softmax muy duro.

Por ejemplo, considere que Q y K tienen una media de 0 y una varianza de 1. Su multiplicación de matrices tendrán una media de 0 y una varianza de dk . Así que la raíz cuadrada de dk se utiliza para escalar, por lo que se obtiene una varianza constante, independientemente del valor de dk . Si la variación es demasiado baja, la salida puede ser demasiado plana para optimizarla de manera eficaz. Si la varianza es demasiado alta, el softmax puede saturarse al inicio, lo que dificulta el aprendizaje.

La máscara se multiplica por -1e9 (cerca del infinito negativo). Esto se hace porque la máscara se suma con la multiplicación de matriz escalada de Q y K y se aplica inmediatamente antes de un softmax. El objetivo es poner a cero estas celdas, y las grandes entradas negativas a softmax están cerca de cero en la salida.

def scaled_dot_product_attention(q, k, v, mask):
  """Calculate the attention weights.
  q, k, v must have matching leading dimensions.
  k, v must have matching penultimate dimension, i.e.: seq_len_k = seq_len_v.
  The mask has different shapes depending on its type(padding or look ahead)
  but it must be broadcastable for addition.

  Args:
    q: query shape == (..., seq_len_q, depth)
    k: key shape == (..., seq_len_k, depth)
    v: value shape == (..., seq_len_v, depth_v)
    mask: Float tensor with shape broadcastable
          to (..., seq_len_q, seq_len_k). Defaults to None.

  Returns:
    output, attention_weights
  """

  matmul_qk = tf.matmul(q, k, transpose_b=True)  # (..., seq_len_q, seq_len_k)

  # scale matmul_qk
  dk = tf.cast(tf.shape(k)[-1], tf.float32)
  scaled_attention_logits = matmul_qk / tf.math.sqrt(dk)

  # add the mask to the scaled tensor.
  if mask is not None:
    scaled_attention_logits += (mask * -1e9)

  # softmax is normalized on the last axis (seq_len_k) so that the scores
  # add up to 1.
  attention_weights = tf.nn.softmax(scaled_attention_logits, axis=-1)  # (..., seq_len_q, seq_len_k)

  output = tf.matmul(attention_weights, v)  # (..., seq_len_q, depth_v)

  return output, attention_weights

A medida que la normalización softmax se realiza en K, sus valores deciden la importancia que se le da a Q.

La salida representa la multiplicación de los pesos de atención y el vector V (valor). Esto asegura que las palabras en las que desea enfocarse se mantengan como están y que las palabras irrelevantes se eliminen.

def print_out(q, k, v):
  temp_out, temp_attn = scaled_dot_product_attention(
      q, k, v, None)
  print('Attention weights are:')
  print(temp_attn)
  print('Output is:')
  print(temp_out)
np.set_printoptions(suppress=True)

temp_k = tf.constant([[10, 0, 0],
                      [0, 10, 0],
                      [0, 0, 10],
                      [0, 0, 10]], dtype=tf.float32)  # (4, 3)

temp_v = tf.constant([[1, 0],
                      [10, 0],
                      [100, 5],
                      [1000, 6]], dtype=tf.float32)  # (4, 2)

# This `query` aligns with the second `key`,
# so the second `value` is returned.
temp_q = tf.constant([[0, 10, 0]], dtype=tf.float32)  # (1, 3)
print_out(temp_q, temp_k, temp_v)
2021-07-08 11:07:08.703095: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcublas.so.11
Attention weights are:
tf.Tensor([[0. 1. 0. 0.]], shape=(1, 4), dtype=float32)
Output is:
tf.Tensor([[10.  0.]], shape=(1, 2), dtype=float32)
2021-07-08 11:07:09.081686: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcublasLt.so.11
# This query aligns with a repeated key (third and fourth),
# so all associated values get averaged.
temp_q = tf.constant([[0, 0, 10]], dtype=tf.float32)  # (1, 3)
print_out(temp_q, temp_k, temp_v)
Attention weights are:
tf.Tensor([[0.  0.  0.5 0.5]], shape=(1, 4), dtype=float32)
Output is:
tf.Tensor([[550.    5.5]], shape=(1, 2), dtype=float32)
# This query aligns equally with the first and second key,
# so their values get averaged.
temp_q = tf.constant([[10, 10, 0]], dtype=tf.float32)  # (1, 3)
print_out(temp_q, temp_k, temp_v)
Attention weights are:
tf.Tensor([[0.5 0.5 0.  0. ]], shape=(1, 4), dtype=float32)
Output is:
tf.Tensor([[5.5 0. ]], shape=(1, 2), dtype=float32)

Pase todas las consultas juntas.

temp_q = tf.constant([[0, 0, 10],
                      [0, 10, 0],
                      [10, 10, 0]], dtype=tf.float32)  # (3, 3)
print_out(temp_q, temp_k, temp_v)
Attention weights are:
tf.Tensor(
[[0.  0.  0.5 0.5]
 [0.  1.  0.  0. ]
 [0.5 0.5 0.  0. ]], shape=(3, 4), dtype=float32)
Output is:
tf.Tensor(
[[550.    5.5]
 [ 10.    0. ]
 [  5.5   0. ]], shape=(3, 2), dtype=float32)

Atención multicabezal

atención de múltiples cabezas

La atención de varios cabezales consta de cuatro partes:

  • Capas lineales y divididas en cabezas.
  • Atención de productos escalados.
  • Concatenación de cabezas.
  • Capa lineal final.

Cada bloque de atención de varios cabezales recibe tres entradas; Q (consulta), K (clave), V (valor). Estos se colocan en capas lineales (densas) y se dividen en varias cabezas.

El scaled_dot_product_attention definido anteriormente se aplica a cada cabeza (transmitido por la eficiencia). Se debe usar una máscara adecuada en el paso de atención. Se concatena La salida atención para cada cabezal de entonces (utilizando tf.transpose , y tf.reshape ) y puesto a través de una final Dense capa.

En lugar de una sola cabeza de atención, Q, K y V se dividen en múltiples cabezas porque permite que el modelo atienda conjuntamente la información de diferentes subespacios de representación en diferentes posiciones. Después de la división, cada cabezal tiene una dimensionalidad reducida, por lo que el costo total de cálculo es el mismo que el de una atención de un solo cabezal con dimensionalidad completa.

class MultiHeadAttention(tf.keras.layers.Layer):
  def __init__(self, d_model, num_heads):
    super(MultiHeadAttention, self).__init__()
    self.num_heads = num_heads
    self.d_model = d_model

    assert d_model % self.num_heads == 0

    self.depth = d_model // self.num_heads

    self.wq = tf.keras.layers.Dense(d_model)
    self.wk = tf.keras.layers.Dense(d_model)
    self.wv = tf.keras.layers.Dense(d_model)

    self.dense = tf.keras.layers.Dense(d_model)

  def split_heads(self, x, batch_size):
    """Split the last dimension into (num_heads, depth).
    Transpose the result such that the shape is (batch_size, num_heads, seq_len, depth)
    """
    x = tf.reshape(x, (batch_size, -1, self.num_heads, self.depth))
    return tf.transpose(x, perm=[0, 2, 1, 3])

  def call(self, v, k, q, mask):
    batch_size = tf.shape(q)[0]

    q = self.wq(q)  # (batch_size, seq_len, d_model)
    k = self.wk(k)  # (batch_size, seq_len, d_model)
    v = self.wv(v)  # (batch_size, seq_len, d_model)

    q = self.split_heads(q, batch_size)  # (batch_size, num_heads, seq_len_q, depth)
    k = self.split_heads(k, batch_size)  # (batch_size, num_heads, seq_len_k, depth)
    v = self.split_heads(v, batch_size)  # (batch_size, num_heads, seq_len_v, depth)

    # scaled_attention.shape == (batch_size, num_heads, seq_len_q, depth)
    # attention_weights.shape == (batch_size, num_heads, seq_len_q, seq_len_k)
    scaled_attention, attention_weights = scaled_dot_product_attention(
        q, k, v, mask)

    scaled_attention = tf.transpose(scaled_attention, perm=[0, 2, 1, 3])  # (batch_size, seq_len_q, num_heads, depth)

    concat_attention = tf.reshape(scaled_attention,
                                  (batch_size, -1, self.d_model))  # (batch_size, seq_len_q, d_model)

    output = self.dense(concat_attention)  # (batch_size, seq_len_q, d_model)

    return output, attention_weights

Crear un MultiHeadAttention capa de probar. En cada lugar en la secuencia, y , la MultiHeadAttention ejecuta los 8 cabezas de atención en todos los otros lugares de la secuencia, devolviendo un nuevo vector de la misma longitud en cada ubicación.

temp_mha = MultiHeadAttention(d_model=512, num_heads=8)
y = tf.random.uniform((1, 60, 512))  # (batch_size, encoder_sequence, d_model)
out, attn = temp_mha(y, k=y, q=y, mask=None)
out.shape, attn.shape
(TensorShape([1, 60, 512]), TensorShape([1, 8, 60, 60]))

Red de avance puntual

La red de avance puntual consta de dos capas completamente conectadas con una activación ReLU en el medio.

def point_wise_feed_forward_network(d_model, dff):
  return tf.keras.Sequential([
      tf.keras.layers.Dense(dff, activation='relu'),  # (batch_size, seq_len, dff)
      tf.keras.layers.Dense(d_model)  # (batch_size, seq_len, d_model)
  ])
sample_ffn = point_wise_feed_forward_network(512, 2048)
sample_ffn(tf.random.uniform((64, 50, 512))).shape
TensorShape([64, 50, 512])

Codificador y decodificador

transformador

El modelo de transformador sigue el mismo patrón general como un estándar de secuencia a la secuencia con el modelo de atención .

  • La frase de entrada se pasa a través N capas de codificador que genera una salida para cada palabra / símbolo en la secuencia.
  • El decodificador presta atención a la salida del codificador y su propia entrada (atención propia) para predecir la siguiente palabra.

Capa de codificador

Cada capa de codificador consta de subcapas:

  1. Atención multicabezal (con máscara acolchada)
  2. Redes de avance puntual.

Cada una de estas subcapas tiene una conexión residual a su alrededor seguida de una normalización de capa. Las conexiones residuales ayudan a evitar el problema del gradiente que desaparece en las redes profundas.

La salida de cada subcapa es LayerNorm(x + Sublayer(x)) . La normalización se realiza en el d_model eje (última). Hay N capas de codificador en el transformador.

class EncoderLayer(tf.keras.layers.Layer):
  def __init__(self, d_model, num_heads, dff, rate=0.1):
    super(EncoderLayer, self).__init__()

    self.mha = MultiHeadAttention(d_model, num_heads)
    self.ffn = point_wise_feed_forward_network(d_model, dff)

    self.layernorm1 = tf.keras.layers.LayerNormalization(epsilon=1e-6)
    self.layernorm2 = tf.keras.layers.LayerNormalization(epsilon=1e-6)

    self.dropout1 = tf.keras.layers.Dropout(rate)
    self.dropout2 = tf.keras.layers.Dropout(rate)

  def call(self, x, training, mask):

    attn_output, _ = self.mha(x, x, x, mask)  # (batch_size, input_seq_len, d_model)
    attn_output = self.dropout1(attn_output, training=training)
    out1 = self.layernorm1(x + attn_output)  # (batch_size, input_seq_len, d_model)

    ffn_output = self.ffn(out1)  # (batch_size, input_seq_len, d_model)
    ffn_output = self.dropout2(ffn_output, training=training)
    out2 = self.layernorm2(out1 + ffn_output)  # (batch_size, input_seq_len, d_model)

    return out2
sample_encoder_layer = EncoderLayer(512, 8, 2048)

sample_encoder_layer_output = sample_encoder_layer(
    tf.random.uniform((64, 43, 512)), False, None)

sample_encoder_layer_output.shape  # (batch_size, input_seq_len, d_model)
TensorShape([64, 43, 512])

Capa de decodificador

Cada capa de decodificador consta de subcapas:

  1. Atención multicabezal enmascarada (con máscara de anticipación y máscara de relleno)
  2. Atención multicabezal (con mascarilla acolchada). V (valor) y K (tecla) reciben la salida del codificador como entradas. Q (consulta) recibe la salida de la atención subcapa con varios cabezales enmascarado.
  3. Redes de avance puntual

Cada una de estas subcapas tiene una conexión residual a su alrededor seguida de una normalización de capa. La salida de cada subcapa es LayerNorm(x + Sublayer(x)) . La normalización se realiza en el d_model eje (última).

Hay N capas de decodificadores en el transformador.

Cuando Q recibe la salida del primer bloque de atención del decodificador y K recibe la salida del codificador, las ponderaciones de atención representan la importancia dada a la entrada del decodificador en función de la salida del codificador. En otras palabras, el decodificador predice la siguiente palabra mirando la salida del codificador y atendiendo su propia salida. Vea la demostración anterior en la sección de atención del producto escalado.

class DecoderLayer(tf.keras.layers.Layer):
  def __init__(self, d_model, num_heads, dff, rate=0.1):
    super(DecoderLayer, self).__init__()

    self.mha1 = MultiHeadAttention(d_model, num_heads)
    self.mha2 = MultiHeadAttention(d_model, num_heads)

    self.ffn = point_wise_feed_forward_network(d_model, dff)

    self.layernorm1 = tf.keras.layers.LayerNormalization(epsilon=1e-6)
    self.layernorm2 = tf.keras.layers.LayerNormalization(epsilon=1e-6)
    self.layernorm3 = tf.keras.layers.LayerNormalization(epsilon=1e-6)

    self.dropout1 = tf.keras.layers.Dropout(rate)
    self.dropout2 = tf.keras.layers.Dropout(rate)
    self.dropout3 = tf.keras.layers.Dropout(rate)

  def call(self, x, enc_output, training,
           look_ahead_mask, padding_mask):
    # enc_output.shape == (batch_size, input_seq_len, d_model)

    attn1, attn_weights_block1 = self.mha1(x, x, x, look_ahead_mask)  # (batch_size, target_seq_len, d_model)
    attn1 = self.dropout1(attn1, training=training)
    out1 = self.layernorm1(attn1 + x)

    attn2, attn_weights_block2 = self.mha2(
        enc_output, enc_output, out1, padding_mask)  # (batch_size, target_seq_len, d_model)
    attn2 = self.dropout2(attn2, training=training)
    out2 = self.layernorm2(attn2 + out1)  # (batch_size, target_seq_len, d_model)

    ffn_output = self.ffn(out2)  # (batch_size, target_seq_len, d_model)
    ffn_output = self.dropout3(ffn_output, training=training)
    out3 = self.layernorm3(ffn_output + out2)  # (batch_size, target_seq_len, d_model)

    return out3, attn_weights_block1, attn_weights_block2
sample_decoder_layer = DecoderLayer(512, 8, 2048)

sample_decoder_layer_output, _, _ = sample_decoder_layer(
    tf.random.uniform((64, 50, 512)), sample_encoder_layer_output,
    False, None, None)

sample_decoder_layer_output.shape  # (batch_size, target_seq_len, d_model)
TensorShape([64, 50, 512])

Codificador

El Encoder consta de:

  1. Incrustación de entrada
  2. Codificación posicional
  3. N capas de codificador

La entrada se somete a una incrustación que se suma con la codificación posicional. La salida de esta suma es la entrada a las capas del codificador. La salida del codificador es la entrada al decodificador.

class Encoder(tf.keras.layers.Layer):
  def __init__(self, num_layers, d_model, num_heads, dff, input_vocab_size,
               maximum_position_encoding, rate=0.1):
    super(Encoder, self).__init__()

    self.d_model = d_model
    self.num_layers = num_layers

    self.embedding = tf.keras.layers.Embedding(input_vocab_size, d_model)
    self.pos_encoding = positional_encoding(maximum_position_encoding,
                                            self.d_model)

    self.enc_layers = [EncoderLayer(d_model, num_heads, dff, rate)
                       for _ in range(num_layers)]

    self.dropout = tf.keras.layers.Dropout(rate)

  def call(self, x, training, mask):

    seq_len = tf.shape(x)[1]

    # adding embedding and position encoding.
    x = self.embedding(x)  # (batch_size, input_seq_len, d_model)
    x *= tf.math.sqrt(tf.cast(self.d_model, tf.float32))
    x += self.pos_encoding[:, :seq_len, :]

    x = self.dropout(x, training=training)

    for i in range(self.num_layers):
      x = self.enc_layers[i](x, training, mask)

    return x  # (batch_size, input_seq_len, d_model)
sample_encoder = Encoder(num_layers=2, d_model=512, num_heads=8,
                         dff=2048, input_vocab_size=8500,
                         maximum_position_encoding=10000)
temp_input = tf.random.uniform((64, 62), dtype=tf.int64, minval=0, maxval=200)

sample_encoder_output = sample_encoder(temp_input, training=False, mask=None)

print(sample_encoder_output.shape)  # (batch_size, input_seq_len, d_model)
(64, 62, 512)

Descifrador

El Decoder consiste en:

  1. Incrustación de salida
  2. Codificación posicional
  3. N capas de decodificador

El objetivo se somete a una incrustación que se suma con la codificación posicional. La salida de esta suma es la entrada a las capas del decodificador. La salida del decodificador es la entrada a la capa lineal final.

class Decoder(tf.keras.layers.Layer):
  def __init__(self, num_layers, d_model, num_heads, dff, target_vocab_size,
               maximum_position_encoding, rate=0.1):
    super(Decoder, self).__init__()

    self.d_model = d_model
    self.num_layers = num_layers

    self.embedding = tf.keras.layers.Embedding(target_vocab_size, d_model)
    self.pos_encoding = positional_encoding(maximum_position_encoding, d_model)

    self.dec_layers = [DecoderLayer(d_model, num_heads, dff, rate)
                       for _ in range(num_layers)]
    self.dropout = tf.keras.layers.Dropout(rate)

  def call(self, x, enc_output, training,
           look_ahead_mask, padding_mask):

    seq_len = tf.shape(x)[1]
    attention_weights = {}

    x = self.embedding(x)  # (batch_size, target_seq_len, d_model)
    x *= tf.math.sqrt(tf.cast(self.d_model, tf.float32))
    x += self.pos_encoding[:, :seq_len, :]

    x = self.dropout(x, training=training)

    for i in range(self.num_layers):
      x, block1, block2 = self.dec_layers[i](x, enc_output, training,
                                             look_ahead_mask, padding_mask)

      attention_weights[f'decoder_layer{i+1}_block1'] = block1
      attention_weights[f'decoder_layer{i+1}_block2'] = block2

    # x.shape == (batch_size, target_seq_len, d_model)
    return x, attention_weights
sample_decoder = Decoder(num_layers=2, d_model=512, num_heads=8,
                         dff=2048, target_vocab_size=8000,
                         maximum_position_encoding=5000)
temp_input = tf.random.uniform((64, 26), dtype=tf.int64, minval=0, maxval=200)

output, attn = sample_decoder(temp_input,
                              enc_output=sample_encoder_output,
                              training=False,
                              look_ahead_mask=None,
                              padding_mask=None)

output.shape, attn['decoder_layer2_block2'].shape
(TensorShape([64, 26, 512]), TensorShape([64, 8, 26, 62]))

Crea el transformador

El transformador consta del codificador, el decodificador y una capa lineal final. La salida del decodificador es la entrada a la capa lineal y se devuelve su salida.

class Transformer(tf.keras.Model):
  def __init__(self, num_layers, d_model, num_heads, dff, input_vocab_size,
               target_vocab_size, pe_input, pe_target, rate=0.1):
    super(Transformer, self).__init__()

    self.encoder = Encoder(num_layers, d_model, num_heads, dff,
                             input_vocab_size, pe_input, rate)

    self.decoder = Decoder(num_layers, d_model, num_heads, dff,
                           target_vocab_size, pe_target, rate)

    self.final_layer = tf.keras.layers.Dense(target_vocab_size)

  def call(self, inp, tar, training, enc_padding_mask,
           look_ahead_mask, dec_padding_mask):

    enc_output = self.encoder(inp, training, enc_padding_mask)  # (batch_size, inp_seq_len, d_model)

    # dec_output.shape == (batch_size, tar_seq_len, d_model)
    dec_output, attention_weights = self.decoder(
        tar, enc_output, training, look_ahead_mask, dec_padding_mask)

    final_output = self.final_layer(dec_output)  # (batch_size, tar_seq_len, target_vocab_size)

    return final_output, attention_weights
sample_transformer = Transformer(
    num_layers=2, d_model=512, num_heads=8, dff=2048,
    input_vocab_size=8500, target_vocab_size=8000,
    pe_input=10000, pe_target=6000)

temp_input = tf.random.uniform((64, 38), dtype=tf.int64, minval=0, maxval=200)
temp_target = tf.random.uniform((64, 36), dtype=tf.int64, minval=0, maxval=200)

fn_out, _ = sample_transformer(temp_input, temp_target, training=False,
                               enc_padding_mask=None,
                               look_ahead_mask=None,
                               dec_padding_mask=None)

fn_out.shape  # (batch_size, tar_seq_len, target_vocab_size)
TensorShape([64, 36, 8000])

Establecer hiperparámetros

Para mantener este ejemplo pequeña y relativamente rápido, se han reducido los valores de num_layers, d_model, y DFF.

Los valores usados ​​en el modelo base de transformador fueron; num_layers = 6, d_model = 512, dff = 2,048. Ver el papel para todas las otras versiones del transformador.

num_layers = 4
d_model = 128
dff = 512
num_heads = 8
dropout_rate = 0.1

Optimizador

Utilizar el optimizador de Adán con un programador de velocidad de aprendizaje personalizado de acuerdo a la fórmula en el papel .

$$\Large{lrate = d_{model}^{-0.5} * \min(step{\_}num^{-0.5}, step{\_}num \cdot warmup{\_}steps^{-1.5})}$$
class CustomSchedule(tf.keras.optimizers.schedules.LearningRateSchedule):
  def __init__(self, d_model, warmup_steps=4000):
    super(CustomSchedule, self).__init__()

    self.d_model = d_model
    self.d_model = tf.cast(self.d_model, tf.float32)

    self.warmup_steps = warmup_steps

  def __call__(self, step):
    arg1 = tf.math.rsqrt(step)
    arg2 = step * (self.warmup_steps ** -1.5)

    return tf.math.rsqrt(self.d_model) * tf.math.minimum(arg1, arg2)
learning_rate = CustomSchedule(d_model)

optimizer = tf.keras.optimizers.Adam(learning_rate, beta_1=0.9, beta_2=0.98,
                                     epsilon=1e-9)
temp_learning_rate_schedule = CustomSchedule(d_model)

plt.plot(temp_learning_rate_schedule(tf.range(40000, dtype=tf.float32)))
plt.ylabel("Learning Rate")
plt.xlabel("Train Step")
Text(0.5, 0, 'Train Step')

png

Pérdidas y métricas

Dado que las secuencias de destino están rellenas, es importante aplicar una máscara de relleno al calcular la pérdida.

loss_object = tf.keras.losses.SparseCategoricalCrossentropy(
    from_logits=True, reduction='none')
def loss_function(real, pred):
  mask = tf.math.logical_not(tf.math.equal(real, 0))
  loss_ = loss_object(real, pred)

  mask = tf.cast(mask, dtype=loss_.dtype)
  loss_ *= mask

  return tf.reduce_sum(loss_)/tf.reduce_sum(mask)


def accuracy_function(real, pred):
  accuracies = tf.equal(real, tf.argmax(pred, axis=2))

  mask = tf.math.logical_not(tf.math.equal(real, 0))
  accuracies = tf.math.logical_and(mask, accuracies)

  accuracies = tf.cast(accuracies, dtype=tf.float32)
  mask = tf.cast(mask, dtype=tf.float32)
  return tf.reduce_sum(accuracies)/tf.reduce_sum(mask)
train_loss = tf.keras.metrics.Mean(name='train_loss')
train_accuracy = tf.keras.metrics.Mean(name='train_accuracy')

Entrenamiento y puntos de control

transformer = Transformer(
    num_layers=num_layers,
    d_model=d_model,
    num_heads=num_heads,
    dff=dff,
    input_vocab_size=tokenizers.pt.get_vocab_size(),
    target_vocab_size=tokenizers.en.get_vocab_size(),
    pe_input=1000,
    pe_target=1000,
    rate=dropout_rate)
def create_masks(inp, tar):
  # Encoder padding mask
  enc_padding_mask = create_padding_mask(inp)

  # Used in the 2nd attention block in the decoder.
  # This padding mask is used to mask the encoder outputs.
  dec_padding_mask = create_padding_mask(inp)

  # Used in the 1st attention block in the decoder.
  # It is used to pad and mask future tokens in the input received by
  # the decoder.
  look_ahead_mask = create_look_ahead_mask(tf.shape(tar)[1])
  dec_target_padding_mask = create_padding_mask(tar)
  combined_mask = tf.maximum(dec_target_padding_mask, look_ahead_mask)

  return enc_padding_mask, combined_mask, dec_padding_mask

Cree la ruta del punto de control y el administrador del punto de control. Esto será utilizado para salvar los puestos de control cada n épocas.

checkpoint_path = "./checkpoints/train"

ckpt = tf.train.Checkpoint(transformer=transformer,
                           optimizer=optimizer)

ckpt_manager = tf.train.CheckpointManager(ckpt, checkpoint_path, max_to_keep=5)

# if a checkpoint exists, restore the latest checkpoint.
if ckpt_manager.latest_checkpoint:
  ckpt.restore(ckpt_manager.latest_checkpoint)
  print('Latest checkpoint restored!!')

El objetivo se divide en tar_inp y tar_real. tar_inp se pasa como entrada al decodificador. tar_real es que misma entrada desplazado por 1: En cada ubicación en tar_input , tar_real contiene la siguiente muestra que debe ser predicho.

Por ejemplo, sentence = "SOS Un león en la selva está durmiendo EOS"

tar_inp = "SOS Un león en la selva está durmiendo"

tar_real = "Un león en la selva está durmiendo EOS"

El transformador es un modelo autoregresivo: hace predicciones una parte a la vez y utiliza su salida hasta ahora para decidir qué hacer a continuación.

Durante el entrenamiento, este ejemplo se utiliza el maestro de forzamiento (como en el tutorial de generación de texto ). La obligación del maestro consiste en pasar la salida real al siguiente paso de tiempo independientemente de lo que predice el modelo en el paso de tiempo actual.

Como el transformador predice cada palabra, la auto-atención permite que se vea en las palabras anteriores en la secuencia de entrada para predecir mejor la siguiente palabra.

Para evitar que el modelo vea la salida esperada, el modelo usa una máscara de anticipación.

EPOCHS = 20
# The @tf.function trace-compiles train_step into a TF graph for faster
# execution. The function specializes to the precise shape of the argument
# tensors. To avoid re-tracing due to the variable sequence lengths or variable
# batch sizes (the last batch is smaller), use input_signature to specify
# more generic shapes.

train_step_signature = [
    tf.TensorSpec(shape=(None, None), dtype=tf.int64),
    tf.TensorSpec(shape=(None, None), dtype=tf.int64),
]


@tf.function(input_signature=train_step_signature)
def train_step(inp, tar):
  tar_inp = tar[:, :-1]
  tar_real = tar[:, 1:]

  enc_padding_mask, combined_mask, dec_padding_mask = create_masks(inp, tar_inp)

  with tf.GradientTape() as tape:
    predictions, _ = transformer(inp, tar_inp,
                                 True,
                                 enc_padding_mask,
                                 combined_mask,
                                 dec_padding_mask)
    loss = loss_function(tar_real, predictions)

  gradients = tape.gradient(loss, transformer.trainable_variables)
  optimizer.apply_gradients(zip(gradients, transformer.trainable_variables))

  train_loss(loss)
  train_accuracy(accuracy_function(tar_real, predictions))

El portugués se utiliza como idioma de entrada y el inglés es el idioma de destino.

for epoch in range(EPOCHS):
  start = time.time()

  train_loss.reset_states()
  train_accuracy.reset_states()

  # inp -> portuguese, tar -> english
  for (batch, (inp, tar)) in enumerate(train_batches):
    train_step(inp, tar)

    if batch % 50 == 0:
      print(f'Epoch {epoch + 1} Batch {batch} Loss {train_loss.result():.4f} Accuracy {train_accuracy.result():.4f}')

  if (epoch + 1) % 5 == 0:
    ckpt_save_path = ckpt_manager.save()
    print(f'Saving checkpoint for epoch {epoch+1} at {ckpt_save_path}')

  print(f'Epoch {epoch + 1} Loss {train_loss.result():.4f} Accuracy {train_accuracy.result():.4f}')

  print(f'Time taken for 1 epoch: {time.time() - start:.2f} secs\n')
Epoch 1 Batch 0 Loss 8.8707 Accuracy 0.0000
Epoch 1 Batch 50 Loss 8.8072 Accuracy 0.0017
Epoch 1 Batch 100 Loss 8.6996 Accuracy 0.0249
Epoch 1 Batch 150 Loss 8.5796 Accuracy 0.0330
Epoch 1 Batch 200 Loss 8.4349 Accuracy 0.0369
Epoch 1 Batch 250 Loss 8.2633 Accuracy 0.0401
Epoch 1 Batch 300 Loss 8.0702 Accuracy 0.0488
Epoch 1 Batch 350 Loss 7.8725 Accuracy 0.0566
Epoch 1 Batch 400 Loss 7.6858 Accuracy 0.0646
Epoch 1 Batch 450 Loss 7.5170 Accuracy 0.0724
Epoch 1 Batch 500 Loss 7.3676 Accuracy 0.0796
Epoch 1 Batch 550 Loss 7.2332 Accuracy 0.0864
Epoch 1 Batch 600 Loss 7.1098 Accuracy 0.0936
Epoch 1 Batch 650 Loss 6.9939 Accuracy 0.1007
Epoch 1 Batch 700 Loss 6.8868 Accuracy 0.1073
Epoch 1 Batch 750 Loss 6.7882 Accuracy 0.1131
Epoch 1 Batch 800 Loss 6.6971 Accuracy 0.1187
Epoch 1 Loss 6.6815 Accuracy 0.1196
Time taken for 1 epoch: 62.90 secs

Epoch 2 Batch 0 Loss 5.3044 Accuracy 0.1971
Epoch 2 Batch 50 Loss 5.2654 Accuracy 0.2064
Epoch 2 Batch 100 Loss 5.2081 Accuracy 0.2128
Epoch 2 Batch 150 Loss 5.1838 Accuracy 0.2158
Epoch 2 Batch 200 Loss 5.1464 Accuracy 0.2194
Epoch 2 Batch 250 Loss 5.1202 Accuracy 0.2223
Epoch 2 Batch 300 Loss 5.1010 Accuracy 0.2243
Epoch 2 Batch 350 Loss 5.0770 Accuracy 0.2268
Epoch 2 Batch 400 Loss 5.0576 Accuracy 0.2290
Epoch 2 Batch 450 Loss 5.0360 Accuracy 0.2310
Epoch 2 Batch 500 Loss 5.0142 Accuracy 0.2332
Epoch 2 Batch 550 Loss 4.9944 Accuracy 0.2353
Epoch 2 Batch 600 Loss 4.9748 Accuracy 0.2370
Epoch 2 Batch 650 Loss 4.9564 Accuracy 0.2387
Epoch 2 Batch 700 Loss 4.9403 Accuracy 0.2404
Epoch 2 Batch 750 Loss 4.9225 Accuracy 0.2421
Epoch 2 Batch 800 Loss 4.9034 Accuracy 0.2440
Epoch 2 Loss 4.8998 Accuracy 0.2444
Time taken for 1 epoch: 49.91 secs

Epoch 3 Batch 0 Loss 4.7305 Accuracy 0.2408
Epoch 3 Batch 50 Loss 4.5827 Accuracy 0.2703
Epoch 3 Batch 100 Loss 4.5608 Accuracy 0.2727
Epoch 3 Batch 150 Loss 4.5528 Accuracy 0.2748
Epoch 3 Batch 200 Loss 4.5467 Accuracy 0.2765
Epoch 3 Batch 250 Loss 4.5297 Accuracy 0.2784
Epoch 3 Batch 300 Loss 4.5212 Accuracy 0.2795
Epoch 3 Batch 350 Loss 4.5021 Accuracy 0.2816
Epoch 3 Batch 400 Loss 4.4806 Accuracy 0.2840
Epoch 3 Batch 450 Loss 4.4646 Accuracy 0.2861
Epoch 3 Batch 500 Loss 4.4517 Accuracy 0.2876
Epoch 3 Batch 550 Loss 4.4381 Accuracy 0.2891
Epoch 3 Batch 600 Loss 4.4239 Accuracy 0.2908
Epoch 3 Batch 650 Loss 4.4090 Accuracy 0.2927
Epoch 3 Batch 700 Loss 4.3931 Accuracy 0.2944
Epoch 3 Batch 750 Loss 4.3753 Accuracy 0.2965
Epoch 3 Batch 800 Loss 4.3579 Accuracy 0.2987
Epoch 3 Loss 4.3555 Accuracy 0.2990
Time taken for 1 epoch: 50.01 secs

Epoch 4 Batch 0 Loss 4.0721 Accuracy 0.3377
Epoch 4 Batch 50 Loss 3.9834 Accuracy 0.3413
Epoch 4 Batch 100 Loss 3.9864 Accuracy 0.3406
Epoch 4 Batch 150 Loss 3.9786 Accuracy 0.3416
Epoch 4 Batch 200 Loss 3.9657 Accuracy 0.3439
Epoch 4 Batch 250 Loss 3.9558 Accuracy 0.3455
Epoch 4 Batch 300 Loss 3.9431 Accuracy 0.3475
Epoch 4 Batch 350 Loss 3.9333 Accuracy 0.3486
Epoch 4 Batch 400 Loss 3.9178 Accuracy 0.3504
Epoch 4 Batch 450 Loss 3.9007 Accuracy 0.3528
Epoch 4 Batch 500 Loss 3.8905 Accuracy 0.3541
Epoch 4 Batch 550 Loss 3.8733 Accuracy 0.3562
Epoch 4 Batch 600 Loss 3.8643 Accuracy 0.3571
Epoch 4 Batch 650 Loss 3.8486 Accuracy 0.3593
Epoch 4 Batch 700 Loss 3.8361 Accuracy 0.3609
Epoch 4 Batch 750 Loss 3.8247 Accuracy 0.3624
Epoch 4 Batch 800 Loss 3.8103 Accuracy 0.3644
Epoch 4 Loss 3.8085 Accuracy 0.3646
Time taken for 1 epoch: 49.71 secs

Epoch 5 Batch 0 Loss 3.7110 Accuracy 0.3674
Epoch 5 Batch 50 Loss 3.4971 Accuracy 0.3997
Epoch 5 Batch 100 Loss 3.4753 Accuracy 0.4034
Epoch 5 Batch 150 Loss 3.4749 Accuracy 0.4028
Epoch 5 Batch 200 Loss 3.4689 Accuracy 0.4033
Epoch 5 Batch 250 Loss 3.4604 Accuracy 0.4048
Epoch 5 Batch 300 Loss 3.4442 Accuracy 0.4071
Epoch 5 Batch 350 Loss 3.4375 Accuracy 0.4080
Epoch 5 Batch 400 Loss 3.4288 Accuracy 0.4094
Epoch 5 Batch 450 Loss 3.4218 Accuracy 0.4104
Epoch 5 Batch 500 Loss 3.4095 Accuracy 0.4120
Epoch 5 Batch 550 Loss 3.4026 Accuracy 0.4128
Epoch 5 Batch 600 Loss 3.3927 Accuracy 0.4140
Epoch 5 Batch 650 Loss 3.3852 Accuracy 0.4149
Epoch 5 Batch 700 Loss 3.3768 Accuracy 0.4163
Epoch 5 Batch 750 Loss 3.3687 Accuracy 0.4174
Epoch 5 Batch 800 Loss 3.3597 Accuracy 0.4187
Saving checkpoint for epoch 5 at ./checkpoints/train/ckpt-1
Epoch 5 Loss 3.3582 Accuracy 0.4189
Time taken for 1 epoch: 49.90 secs

Epoch 6 Batch 0 Loss 3.4475 Accuracy 0.3957
Epoch 6 Batch 50 Loss 3.0962 Accuracy 0.4477
Epoch 6 Batch 100 Loss 3.0798 Accuracy 0.4508
Epoch 6 Batch 150 Loss 3.0787 Accuracy 0.4510
Epoch 6 Batch 200 Loss 3.0601 Accuracy 0.4530
Epoch 6 Batch 250 Loss 3.0577 Accuracy 0.4536
Epoch 6 Batch 300 Loss 3.0457 Accuracy 0.4554
Epoch 6 Batch 350 Loss 3.0400 Accuracy 0.4561
Epoch 6 Batch 400 Loss 3.0303 Accuracy 0.4576
Epoch 6 Batch 450 Loss 3.0230 Accuracy 0.4587
Epoch 6 Batch 500 Loss 3.0081 Accuracy 0.4610
Epoch 6 Batch 550 Loss 3.0013 Accuracy 0.4621
Epoch 6 Batch 600 Loss 2.9921 Accuracy 0.4635
Epoch 6 Batch 650 Loss 2.9814 Accuracy 0.4653
Epoch 6 Batch 700 Loss 2.9738 Accuracy 0.4663
Epoch 6 Batch 750 Loss 2.9645 Accuracy 0.4676
Epoch 6 Batch 800 Loss 2.9575 Accuracy 0.4687
Epoch 6 Loss 2.9559 Accuracy 0.4690
Time taken for 1 epoch: 49.51 secs

Epoch 7 Batch 0 Loss 2.9116 Accuracy 0.4652
Epoch 7 Batch 50 Loss 2.6920 Accuracy 0.5008
Epoch 7 Batch 100 Loss 2.6726 Accuracy 0.5045
Epoch 7 Batch 150 Loss 2.6776 Accuracy 0.5042
Epoch 7 Batch 200 Loss 2.6715 Accuracy 0.5049
Epoch 7 Batch 250 Loss 2.6665 Accuracy 0.5056
Epoch 7 Batch 300 Loss 2.6618 Accuracy 0.5067
Epoch 7 Batch 350 Loss 2.6560 Accuracy 0.5075
Epoch 7 Batch 400 Loss 2.6515 Accuracy 0.5084
Epoch 7 Batch 450 Loss 2.6440 Accuracy 0.5092
Epoch 7 Batch 500 Loss 2.6372 Accuracy 0.5101
Epoch 7 Batch 550 Loss 2.6334 Accuracy 0.5108
Epoch 7 Batch 600 Loss 2.6284 Accuracy 0.5115
Epoch 7 Batch 650 Loss 2.6249 Accuracy 0.5122
Epoch 7 Batch 700 Loss 2.6191 Accuracy 0.5131
Epoch 7 Batch 750 Loss 2.6154 Accuracy 0.5137
Epoch 7 Batch 800 Loss 2.6133 Accuracy 0.5140
Epoch 7 Loss 2.6122 Accuracy 0.5142
Time taken for 1 epoch: 50.79 secs

Epoch 8 Batch 0 Loss 2.2190 Accuracy 0.5521
Epoch 8 Batch 50 Loss 2.3789 Accuracy 0.5434
Epoch 8 Batch 100 Loss 2.4058 Accuracy 0.5404
Epoch 8 Batch 150 Loss 2.3813 Accuracy 0.5438
Epoch 8 Batch 200 Loss 2.3857 Accuracy 0.5428
Epoch 8 Batch 250 Loss 2.3861 Accuracy 0.5430
Epoch 8 Batch 300 Loss 2.3847 Accuracy 0.5432
Epoch 8 Batch 350 Loss 2.3804 Accuracy 0.5441
Epoch 8 Batch 400 Loss 2.3820 Accuracy 0.5437
Epoch 8 Batch 450 Loss 2.3807 Accuracy 0.5442
Epoch 8 Batch 500 Loss 2.3789 Accuracy 0.5446
Epoch 8 Batch 550 Loss 2.3779 Accuracy 0.5448
Epoch 8 Batch 600 Loss 2.3762 Accuracy 0.5452
Epoch 8 Batch 650 Loss 2.3746 Accuracy 0.5456
Epoch 8 Batch 700 Loss 2.3757 Accuracy 0.5455
Epoch 8 Batch 750 Loss 2.3760 Accuracy 0.5456
Epoch 8 Batch 800 Loss 2.3746 Accuracy 0.5459
Epoch 8 Loss 2.3733 Accuracy 0.5461
Time taken for 1 epoch: 49.50 secs

Epoch 9 Batch 0 Loss 2.0450 Accuracy 0.5789
Epoch 9 Batch 50 Loss 2.1817 Accuracy 0.5704
Epoch 9 Batch 100 Loss 2.1819 Accuracy 0.5716
Epoch 9 Batch 150 Loss 2.1817 Accuracy 0.5717
Epoch 9 Batch 200 Loss 2.1886 Accuracy 0.5702
Epoch 9 Batch 250 Loss 2.1971 Accuracy 0.5694
Epoch 9 Batch 300 Loss 2.1957 Accuracy 0.5695
Epoch 9 Batch 350 Loss 2.1913 Accuracy 0.5704
Epoch 9 Batch 400 Loss 2.1918 Accuracy 0.5704
Epoch 9 Batch 450 Loss 2.1886 Accuracy 0.5709
Epoch 9 Batch 500 Loss 2.1903 Accuracy 0.5707
Epoch 9 Batch 550 Loss 2.1898 Accuracy 0.5708
Epoch 9 Batch 600 Loss 2.1901 Accuracy 0.5707
Epoch 9 Batch 650 Loss 2.1903 Accuracy 0.5707
Epoch 9 Batch 700 Loss 2.1919 Accuracy 0.5706
Epoch 9 Batch 750 Loss 2.1929 Accuracy 0.5708
Epoch 9 Batch 800 Loss 2.1940 Accuracy 0.5707
Epoch 9 Loss 2.1945 Accuracy 0.5706
Time taken for 1 epoch: 49.36 secs

Epoch 10 Batch 0 Loss 2.2947 Accuracy 0.5558
Epoch 10 Batch 50 Loss 2.0508 Accuracy 0.5876
Epoch 10 Batch 100 Loss 2.0412 Accuracy 0.5909
Epoch 10 Batch 150 Loss 2.0439 Accuracy 0.5906
Epoch 10 Batch 200 Loss 2.0526 Accuracy 0.5896
Epoch 10 Batch 250 Loss 2.0444 Accuracy 0.5909
Epoch 10 Batch 300 Loss 2.0418 Accuracy 0.5914
Epoch 10 Batch 350 Loss 2.0433 Accuracy 0.5910
Epoch 10 Batch 400 Loss 2.0449 Accuracy 0.5906
Epoch 10 Batch 450 Loss 2.0450 Accuracy 0.5907
Epoch 10 Batch 500 Loss 2.0477 Accuracy 0.5905
Epoch 10 Batch 550 Loss 2.0502 Accuracy 0.5902
Epoch 10 Batch 600 Loss 2.0516 Accuracy 0.5901
Epoch 10 Batch 650 Loss 2.0531 Accuracy 0.5901
Epoch 10 Batch 700 Loss 2.0548 Accuracy 0.5900
Epoch 10 Batch 750 Loss 2.0560 Accuracy 0.5899
Epoch 10 Batch 800 Loss 2.0573 Accuracy 0.5900
Saving checkpoint for epoch 10 at ./checkpoints/train/ckpt-2
Epoch 10 Loss 2.0580 Accuracy 0.5899
Time taken for 1 epoch: 50.74 secs

Epoch 11 Batch 0 Loss 1.9999 Accuracy 0.5880
Epoch 11 Batch 50 Loss 1.9237 Accuracy 0.6101
Epoch 11 Batch 100 Loss 1.9318 Accuracy 0.6070
Epoch 11 Batch 150 Loss 1.9259 Accuracy 0.6077
Epoch 11 Batch 200 Loss 1.9252 Accuracy 0.6076
Epoch 11 Batch 250 Loss 1.9257 Accuracy 0.6081
Epoch 11 Batch 300 Loss 1.9273 Accuracy 0.6079
Epoch 11 Batch 350 Loss 1.9323 Accuracy 0.6073
Epoch 11 Batch 400 Loss 1.9311 Accuracy 0.6079
Epoch 11 Batch 450 Loss 1.9344 Accuracy 0.6073
Epoch 11 Batch 500 Loss 1.9359 Accuracy 0.6070
Epoch 11 Batch 550 Loss 1.9357 Accuracy 0.6072
Epoch 11 Batch 600 Loss 1.9351 Accuracy 0.6074
Epoch 11 Batch 650 Loss 1.9375 Accuracy 0.6072
Epoch 11 Batch 700 Loss 1.9402 Accuracy 0.6069
Epoch 11 Batch 750 Loss 1.9411 Accuracy 0.6068
Epoch 11 Batch 800 Loss 1.9437 Accuracy 0.6065
Epoch 11 Loss 1.9444 Accuracy 0.6063
Time taken for 1 epoch: 49.69 secs

Epoch 12 Batch 0 Loss 1.8838 Accuracy 0.6104
Epoch 12 Batch 50 Loss 1.8113 Accuracy 0.6233
Epoch 12 Batch 100 Loss 1.8283 Accuracy 0.6208
Epoch 12 Batch 150 Loss 1.8255 Accuracy 0.6225
Epoch 12 Batch 200 Loss 1.8361 Accuracy 0.6209
Epoch 12 Batch 250 Loss 1.8375 Accuracy 0.6207
Epoch 12 Batch 300 Loss 1.8418 Accuracy 0.6201
Epoch 12 Batch 350 Loss 1.8439 Accuracy 0.6198
Epoch 12 Batch 400 Loss 1.8457 Accuracy 0.6198
Epoch 12 Batch 450 Loss 1.8468 Accuracy 0.6196
Epoch 12 Batch 500 Loss 1.8463 Accuracy 0.6198
Epoch 12 Batch 550 Loss 1.8468 Accuracy 0.6199
Epoch 12 Batch 600 Loss 1.8486 Accuracy 0.6197
Epoch 12 Batch 650 Loss 1.8492 Accuracy 0.6198
Epoch 12 Batch 700 Loss 1.8494 Accuracy 0.6199
Epoch 12 Batch 750 Loss 1.8515 Accuracy 0.6195
Epoch 12 Batch 800 Loss 1.8527 Accuracy 0.6194
Epoch 12 Loss 1.8535 Accuracy 0.6193
Time taken for 1 epoch: 49.50 secs

Epoch 13 Batch 0 Loss 1.6777 Accuracy 0.6374
Epoch 13 Batch 50 Loss 1.7137 Accuracy 0.6387
Epoch 13 Batch 100 Loss 1.7346 Accuracy 0.6352
Epoch 13 Batch 150 Loss 1.7383 Accuracy 0.6348
Epoch 13 Batch 200 Loss 1.7402 Accuracy 0.6349
Epoch 13 Batch 250 Loss 1.7442 Accuracy 0.6345
Epoch 13 Batch 300 Loss 1.7496 Accuracy 0.6340
Epoch 13 Batch 350 Loss 1.7529 Accuracy 0.6332
Epoch 13 Batch 400 Loss 1.7565 Accuracy 0.6324
Epoch 13 Batch 450 Loss 1.7574 Accuracy 0.6324
Epoch 13 Batch 500 Loss 1.7596 Accuracy 0.6321
Epoch 13 Batch 550 Loss 1.7597 Accuracy 0.6322
Epoch 13 Batch 600 Loss 1.7586 Accuracy 0.6326
Epoch 13 Batch 650 Loss 1.7634 Accuracy 0.6320
Epoch 13 Batch 700 Loss 1.7669 Accuracy 0.6315
Epoch 13 Batch 750 Loss 1.7693 Accuracy 0.6313
Epoch 13 Batch 800 Loss 1.7714 Accuracy 0.6312
Epoch 13 Loss 1.7726 Accuracy 0.6310
Time taken for 1 epoch: 49.98 secs

Epoch 14 Batch 0 Loss 1.6842 Accuracy 0.6471
Epoch 14 Batch 50 Loss 1.6872 Accuracy 0.6433
Epoch 14 Batch 100 Loss 1.6708 Accuracy 0.6463
Epoch 14 Batch 150 Loss 1.6705 Accuracy 0.6459
Epoch 14 Batch 200 Loss 1.6734 Accuracy 0.6454
Epoch 14 Batch 250 Loss 1.6748 Accuracy 0.6455
Epoch 14 Batch 300 Loss 1.6773 Accuracy 0.6451
Epoch 14 Batch 350 Loss 1.6787 Accuracy 0.6450
Epoch 14 Batch 400 Loss 1.6830 Accuracy 0.6446
Epoch 14 Batch 450 Loss 1.6830 Accuracy 0.6446
Epoch 14 Batch 500 Loss 1.6862 Accuracy 0.6440
Epoch 14 Batch 550 Loss 1.6862 Accuracy 0.6440
Epoch 14 Batch 600 Loss 1.6871 Accuracy 0.6441
Epoch 14 Batch 650 Loss 1.6900 Accuracy 0.6436
Epoch 14 Batch 700 Loss 1.6950 Accuracy 0.6429
Epoch 14 Batch 750 Loss 1.6996 Accuracy 0.6423
Epoch 14 Batch 800 Loss 1.7041 Accuracy 0.6417
Epoch 14 Loss 1.7045 Accuracy 0.6416
Time taken for 1 epoch: 50.59 secs

Epoch 15 Batch 0 Loss 1.6510 Accuracy 0.6451
Epoch 15 Batch 50 Loss 1.6027 Accuracy 0.6561
Epoch 15 Batch 100 Loss 1.6035 Accuracy 0.6564
Epoch 15 Batch 150 Loss 1.6128 Accuracy 0.6549
Epoch 15 Batch 200 Loss 1.6139 Accuracy 0.6543
Epoch 15 Batch 250 Loss 1.6123 Accuracy 0.6548
Epoch 15 Batch 300 Loss 1.6125 Accuracy 0.6550
Epoch 15 Batch 350 Loss 1.6196 Accuracy 0.6537
Epoch 15 Batch 400 Loss 1.6209 Accuracy 0.6536
Epoch 15 Batch 450 Loss 1.6243 Accuracy 0.6531
Epoch 15 Batch 500 Loss 1.6269 Accuracy 0.6526
Epoch 15 Batch 550 Loss 1.6300 Accuracy 0.6522
Epoch 15 Batch 600 Loss 1.6316 Accuracy 0.6519
Epoch 15 Batch 650 Loss 1.6329 Accuracy 0.6519
Epoch 15 Batch 700 Loss 1.6354 Accuracy 0.6517
Epoch 15 Batch 750 Loss 1.6378 Accuracy 0.6513
Epoch 15 Batch 800 Loss 1.6422 Accuracy 0.6507
Saving checkpoint for epoch 15 at ./checkpoints/train/ckpt-3
Epoch 15 Loss 1.6425 Accuracy 0.6506
Time taken for 1 epoch: 51.91 secs

Epoch 16 Batch 0 Loss 1.2860 Accuracy 0.7074
Epoch 16 Batch 50 Loss 1.5397 Accuracy 0.6670
Epoch 16 Batch 100 Loss 1.5438 Accuracy 0.6663
Epoch 16 Batch 150 Loss 1.5483 Accuracy 0.6653
Epoch 16 Batch 200 Loss 1.5520 Accuracy 0.6650
Epoch 16 Batch 250 Loss 1.5540 Accuracy 0.6643
Epoch 16 Batch 300 Loss 1.5614 Accuracy 0.6628
Epoch 16 Batch 350 Loss 1.5640 Accuracy 0.6621
Epoch 16 Batch 400 Loss 1.5692 Accuracy 0.6611
Epoch 16 Batch 450 Loss 1.5721 Accuracy 0.6607
Epoch 16 Batch 500 Loss 1.5720 Accuracy 0.6607
Epoch 16 Batch 550 Loss 1.5754 Accuracy 0.6603
Epoch 16 Batch 600 Loss 1.5768 Accuracy 0.6600
Epoch 16 Batch 650 Loss 1.5785 Accuracy 0.6600
Epoch 16 Batch 700 Loss 1.5822 Accuracy 0.6596
Epoch 16 Batch 750 Loss 1.5858 Accuracy 0.6589
Epoch 16 Batch 800 Loss 1.5890 Accuracy 0.6585
Epoch 16 Loss 1.5895 Accuracy 0.6585
Time taken for 1 epoch: 51.78 secs

Epoch 17 Batch 0 Loss 1.5084 Accuracy 0.6485
Epoch 17 Batch 50 Loss 1.4844 Accuracy 0.6755
Epoch 17 Batch 100 Loss 1.4821 Accuracy 0.6752
Epoch 17 Batch 150 Loss 1.4926 Accuracy 0.6736
Epoch 17 Batch 200 Loss 1.5024 Accuracy 0.6724
Epoch 17 Batch 250 Loss 1.5022 Accuracy 0.6726
Epoch 17 Batch 300 Loss 1.5114 Accuracy 0.6711
Epoch 17 Batch 350 Loss 1.5145 Accuracy 0.6705
Epoch 17 Batch 400 Loss 1.5176 Accuracy 0.6701
Epoch 17 Batch 450 Loss 1.5205 Accuracy 0.6696
Epoch 17 Batch 500 Loss 1.5215 Accuracy 0.6696
Epoch 17 Batch 550 Loss 1.5245 Accuracy 0.6690
Epoch 17 Batch 600 Loss 1.5274 Accuracy 0.6684
Epoch 17 Batch 650 Loss 1.5304 Accuracy 0.6680
Epoch 17 Batch 700 Loss 1.5326 Accuracy 0.6677
Epoch 17 Batch 750 Loss 1.5372 Accuracy 0.6670
Epoch 17 Batch 800 Loss 1.5418 Accuracy 0.6661
Epoch 17 Loss 1.5420 Accuracy 0.6662
Time taken for 1 epoch: 51.36 secs

Epoch 18 Batch 0 Loss 1.2714 Accuracy 0.7098
Epoch 18 Batch 50 Loss 1.4713 Accuracy 0.6796
Epoch 18 Batch 100 Loss 1.4666 Accuracy 0.6791
Epoch 18 Batch 150 Loss 1.4654 Accuracy 0.6788
Epoch 18 Batch 200 Loss 1.4648 Accuracy 0.6787
Epoch 18 Batch 250 Loss 1.4658 Accuracy 0.6783
Epoch 18 Batch 300 Loss 1.4673 Accuracy 0.6782
Epoch 18 Batch 350 Loss 1.4720 Accuracy 0.6772
Epoch 18 Batch 400 Loss 1.4728 Accuracy 0.6772
Epoch 18 Batch 450 Loss 1.4752 Accuracy 0.6768
Epoch 18 Batch 500 Loss 1.4774 Accuracy 0.6767
Epoch 18 Batch 550 Loss 1.4805 Accuracy 0.6761
Epoch 18 Batch 600 Loss 1.4834 Accuracy 0.6756
Epoch 18 Batch 650 Loss 1.4881 Accuracy 0.6750
Epoch 18 Batch 700 Loss 1.4922 Accuracy 0.6743
Epoch 18 Batch 750 Loss 1.4945 Accuracy 0.6740
Epoch 18 Batch 800 Loss 1.4965 Accuracy 0.6736
Epoch 18 Loss 1.4969 Accuracy 0.6735
Time taken for 1 epoch: 51.68 secs

Epoch 19 Batch 0 Loss 1.3652 Accuracy 0.7007
Epoch 19 Batch 50 Loss 1.3857 Accuracy 0.6900
Epoch 19 Batch 100 Loss 1.4049 Accuracy 0.6869
Epoch 19 Batch 150 Loss 1.4164 Accuracy 0.6854
Epoch 19 Batch 200 Loss 1.4223 Accuracy 0.6842
Epoch 19 Batch 250 Loss 1.4252 Accuracy 0.6836
Epoch 19 Batch 300 Loss 1.4288 Accuracy 0.6832
Epoch 19 Batch 350 Loss 1.4327 Accuracy 0.6829
Epoch 19 Batch 400 Loss 1.4351 Accuracy 0.6823
Epoch 19 Batch 450 Loss 1.4395 Accuracy 0.6815
Epoch 19 Batch 500 Loss 1.4425 Accuracy 0.6812
Epoch 19 Batch 550 Loss 1.4472 Accuracy 0.6805
Epoch 19 Batch 600 Loss 1.4476 Accuracy 0.6806
Epoch 19 Batch 650 Loss 1.4503 Accuracy 0.6802
Epoch 19 Batch 700 Loss 1.4532 Accuracy 0.6797
Epoch 19 Batch 750 Loss 1.4549 Accuracy 0.6796
Epoch 19 Batch 800 Loss 1.4587 Accuracy 0.6792
Epoch 19 Loss 1.4595 Accuracy 0.6791
Time taken for 1 epoch: 51.77 secs

Epoch 20 Batch 0 Loss 1.2816 Accuracy 0.7223
Epoch 20 Batch 50 Loss 1.3843 Accuracy 0.6913
Epoch 20 Batch 100 Loss 1.3764 Accuracy 0.6929
Epoch 20 Batch 150 Loss 1.3833 Accuracy 0.6913
Epoch 20 Batch 200 Loss 1.3916 Accuracy 0.6900
Epoch 20 Batch 250 Loss 1.3963 Accuracy 0.6893
Epoch 20 Batch 300 Loss 1.4006 Accuracy 0.6886
Epoch 20 Batch 350 Loss 1.4011 Accuracy 0.6884
Epoch 20 Batch 400 Loss 1.4031 Accuracy 0.6881
Epoch 20 Batch 450 Loss 1.4064 Accuracy 0.6873
Epoch 20 Batch 500 Loss 1.4080 Accuracy 0.6872
Epoch 20 Batch 550 Loss 1.4076 Accuracy 0.6873
Epoch 20 Batch 600 Loss 1.4094 Accuracy 0.6872
Epoch 20 Batch 650 Loss 1.4153 Accuracy 0.6862
Epoch 20 Batch 700 Loss 1.4170 Accuracy 0.6859
Epoch 20 Batch 750 Loss 1.4208 Accuracy 0.6854
Epoch 20 Batch 800 Loss 1.4242 Accuracy 0.6848
Saving checkpoint for epoch 20 at ./checkpoints/train/ckpt-4
Epoch 20 Loss 1.4252 Accuracy 0.6846
Time taken for 1 epoch: 52.57 secs

Evaluar

Los siguientes pasos se utilizan para la evaluación:

  • Codificar la frase de entrada utilizando el tokenizer portuguesa ( tokenizers.pt ). Esta es la entrada del codificador.
  • La entrada del decodificador se inicializa a la [START] token.
  • Calcule las máscaras de relleno y las máscaras de anticipación.
  • El decoder continuación da salida a las predicciones observando la encoder output y su propia salida (auto-atención).
  • El modelo hace predicciones de la siguiente palabra para cada palabra en la salida. La mayoría de estos son redundantes. Utilice las predicciones de la última palabra.
  • Concatenar la palabra predicha a la entrada del decodificador y pasarla al decodificador.
  • En este enfoque, el decodificador predice la siguiente palabra basándose en las palabras anteriores que predijo.
def evaluate(sentence, max_length=40):
  # inp sentence is portuguese, hence adding the start and end token
  sentence = tf.convert_to_tensor([sentence])
  sentence = tokenizers.pt.tokenize(sentence).to_tensor()

  encoder_input = sentence

  # as the target is english, the first word to the transformer should be the
  # english start token.
  start, end = tokenizers.en.tokenize([''])[0]
  output = tf.convert_to_tensor([start])
  output = tf.expand_dims(output, 0)

  for i in range(max_length):
    enc_padding_mask, combined_mask, dec_padding_mask = create_masks(
        encoder_input, output)

    # predictions.shape == (batch_size, seq_len, vocab_size)
    predictions, attention_weights = transformer(encoder_input,
                                                 output,
                                                 False,
                                                 enc_padding_mask,
                                                 combined_mask,
                                                 dec_padding_mask)

    # select the last word from the seq_len dimension
    predictions = predictions[:, -1:, :]  # (batch_size, 1, vocab_size)

    predicted_id = tf.argmax(predictions, axis=-1)

    # concatentate the predicted_id to the output which is given to the decoder
    # as its input.
    output = tf.concat([output, predicted_id], axis=-1)

    # return the result if the predicted_id is equal to the end token
    if predicted_id == end:
      break

  # output.shape (1, tokens)
  text = tokenizers.en.detokenize(output)[0]  # shape: ()

  tokens = tokenizers.en.lookup(output)[0]

  return text, tokens, attention_weights
def print_translation(sentence, tokens, ground_truth):
  print(f'{"Input:":15s}: {sentence}')
  print(f'{"Prediction":15s}: {tokens.numpy().decode("utf-8")}')
  print(f'{"Ground truth":15s}: {ground_truth}')
sentence = "este \xe9 um problema que temos que resolver."
ground_truth = "this is a problem we have to solve ."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)
Input:         : este \xe9 um problema que temos que resolver.
Prediction     : this is a problem that we have to solve .
Ground truth   : this is a problem we have to solve .
sentence = "os meus vizinhos ouviram sobre esta ideia."
ground_truth = "and my neighboring homes heard about this idea ."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)
Input:         : os meus vizinhos ouviram sobre esta ideia.
Prediction     : my neighbors heard about this idea .
Ground truth   : and my neighboring homes heard about this idea .
sentence = "vou ent\xe3o muito rapidamente partilhar convosco algumas hist\xf3rias de algumas coisas m\xe1gicas que aconteceram."
ground_truth = "so i \'ll just share with you some stories very quickly of some magical things that have happened ."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)
Input:         : vou ent\xe3o muito rapidamente partilhar convosco algumas hist\xf3rias de algumas coisas m\xe1gicas que aconteceram.
Prediction     : so i ' ll be very quickly share with you some stories of some magic stuff that will happen .
Ground truth   : so i 'll just share with you some stories very quickly of some magical things that have happened .

Puede pasar diferentes capas y bloques atención del decodificador a la plot de parámetros.

Parcelas de atención

El evaluate la función también devuelve un diccionario de atención mapas se puede utilizar para visualizar el funcionamiento interno de la modelo:

sentence = "este \xe9 o primeiro livro que eu fiz."
ground_truth = "this is the first book i've ever done."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)
Input:         : este \xe9 o primeiro livro que eu fiz.
Prediction     : this is the first book i did .
Ground truth   : this is the first book i've ever done.
def plot_attention_head(in_tokens, translated_tokens, attention):
  # The plot is of the attention when a token was generated.
  # The model didn't generate `<START>` in the output. Skip it.
  translated_tokens = translated_tokens[1:]

  ax = plt.gca()
  ax.matshow(attention)
  ax.set_xticks(range(len(in_tokens)))
  ax.set_yticks(range(len(translated_tokens)))

  labels = [label.decode('utf-8') for label in in_tokens.numpy()]
  ax.set_xticklabels(
      labels, rotation=90)

  labels = [label.decode('utf-8') for label in translated_tokens.numpy()]
  ax.set_yticklabels(labels)
head = 0
# shape: (batch=1, num_heads, seq_len_q, seq_len_k)
attention_heads = tf.squeeze(
  attention_weights['decoder_layer4_block2'], 0)
attention = attention_heads[head]
attention.shape
TensorShape([9, 11])
in_tokens = tf.convert_to_tensor([sentence])
in_tokens = tokenizers.pt.tokenize(in_tokens).to_tensor()
in_tokens = tokenizers.pt.lookup(in_tokens)[0]
in_tokens
<tf.Tensor: shape=(11,), dtype=string, numpy=
array([b'[START]', b'este', b'e', b'o', b'primeiro', b'livro', b'que',
       b'eu', b'fiz', b'.', b'[END]'], dtype=object)>
translated_tokens
<tf.Tensor: shape=(10,), dtype=string, numpy=
array([b'[START]', b'this', b'is', b'the', b'first', b'book', b'i',
       b'did', b'.', b'[END]'], dtype=object)>
plot_attention_head(in_tokens, translated_tokens, attention)

png

def plot_attention_weights(sentence, translated_tokens, attention_heads):
  in_tokens = tf.convert_to_tensor([sentence])
  in_tokens = tokenizers.pt.tokenize(in_tokens).to_tensor()
  in_tokens = tokenizers.pt.lookup(in_tokens)[0]
  in_tokens

  fig = plt.figure(figsize=(16, 8))

  for h, head in enumerate(attention_heads):
    ax = fig.add_subplot(2, 4, h+1)

    plot_attention_head(in_tokens, translated_tokens, head)

    ax.set_xlabel(f'Head {h+1}')

  plt.tight_layout()
  plt.show()
plot_attention_weights(sentence, translated_tokens,
                       attention_weights['decoder_layer4_block2'][0])

png

El modelo funciona bien con palabras desconocidas. Ni "triceratops" ni "enciclopedia" están en el conjunto de datos de entrada y el modelo casi aprende a transliterarlos, incluso sin un vocabulario compartido:

sentence = "Eu li sobre triceratops na enciclop\xe9dia."
ground_truth = "I read about triceratops in the encyclopedia."

translated_text, translated_tokens, attention_weights = evaluate(sentence)
print_translation(sentence, translated_text, ground_truth)

plot_attention_weights(sentence, translated_tokens,
                       attention_weights['decoder_layer4_block2'][0])
Input:         : Eu li sobre triceratops na enciclop\xe9dia.
Prediction     : i read about tattoosees in encyclopedia .
Ground truth   : I read about triceratops in the encyclopedia.

png

Resumen

En este tutorial, aprendió sobre la codificación posicional, la atención de múltiples cabezales, la importancia del enmascaramiento y cómo crear un transformador.

Intente usar un conjunto de datos diferente para entrenar el transformador. También puede crear el transformador base o el transformador XL cambiando los hiperparámetros anteriores. También puede utilizar las capas definidas aquí para crear BERT y trenes estado de los modelos de arte. Además, puede implementar la búsqueda de haces para obtener mejores predicciones.