代码示例 / 生成式深度学习 / 使用 FNet 进行文本生成

使用 FNet 进行文本生成

作者: Darshan Deshpande
创建日期 2021/10/05
上次修改 2021/10/05

ⓘ 此示例使用 Keras 2

在 Colab 中查看 GitHub 源代码

描述:Keras 中用于文本生成的 FNet 变压器。


引言

原始的 Transformer 实现(Vaswani 等人,2017 年)是自然语言处理领域的一项重大突破,催生了 BERT 和 GPT 等重要的架构。但是,这些架构的缺点是它们使用的自注意力机制计算量很大。FNet 架构建议用一种更精简的机制来替代这种自注意力机制:一种基于傅里叶变换的输入 token 线性混合器。

FNet 模型能够在训练速度上比 BERT 快 80%(GPU)和近 70%(TPU),同时达到 92-97% 的 BERT 准确率。这种设计提供了一种高效且模型尺寸小的方案,从而缩短推理时间。

在本例中,我们将在此架构上实现和训练康奈尔电影对话语料库,以展示该模型在文本生成方面的适用性。


导入

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
import os

# Defining hyperparameters

VOCAB_SIZE = 8192
MAX_SAMPLES = 50000
BUFFER_SIZE = 20000
MAX_LENGTH = 40
EMBED_DIM = 256
LATENT_DIM = 512
NUM_HEADS = 8
BATCH_SIZE = 64

加载数据

我们将使用康奈尔对话语料库。我们将把电影对话解析成问题和答案集。

path_to_zip = keras.utils.get_file(
    "cornell_movie_dialogs.zip",
    origin="http://www.cs.cornell.edu/~cristian/data/cornell_movie_dialogs_corpus.zip",
    extract=True,
)

path_to_dataset = os.path.join(
    os.path.dirname(path_to_zip), "cornell movie-dialogs corpus"
)
path_to_movie_lines = os.path.join(path_to_dataset, "movie_lines.txt")
path_to_movie_conversations = os.path.join(path_to_dataset, "movie_conversations.txt")


def load_conversations():
    # Helper function for loading the conversation splits
    id2line = {}
    with open(path_to_movie_lines, errors="ignore") as file:
        lines = file.readlines()
    for line in lines:
        parts = line.replace("\n", "").split(" +++$+++ ")
        id2line[parts[0]] = parts[4]

    inputs, outputs = [], []
    with open(path_to_movie_conversations, "r") as file:
        lines = file.readlines()
    for line in lines:
        parts = line.replace("\n", "").split(" +++$+++ ")
        # get conversation in a list of line ID
        conversation = [line[1:-1] for line in parts[3][1:-1].split(", ")]
        for i in range(len(conversation) - 1):
            inputs.append(id2line[conversation[i]])
            outputs.append(id2line[conversation[i + 1]])
            if len(inputs) >= MAX_SAMPLES:
                return inputs, outputs
    return inputs, outputs


questions, answers = load_conversations()

# Splitting training and validation sets

train_dataset = tf.data.Dataset.from_tensor_slices((questions[:40000], answers[:40000]))
val_dataset = tf.data.Dataset.from_tensor_slices((questions[40000:], answers[40000:]))
Downloading data from http://www.cs.cornell.edu/~cristian/data/cornell_movie_dialogs_corpus.zip
9920512/9916637 [==============================] - 0s 0us/step
9928704/9916637 [==============================] - 0s 0us/step

预处理和标记化

def preprocess_text(sentence):
    sentence = tf.strings.lower(sentence)
    # Adding a space between the punctuation and the last word to allow better tokenization
    sentence = tf.strings.regex_replace(sentence, r"([?.!,])", r" \1 ")
    # Replacing multiple continuous spaces with a single space
    sentence = tf.strings.regex_replace(sentence, r"\s\s+", " ")
    # Replacing non english words with spaces
    sentence = tf.strings.regex_replace(sentence, r"[^a-z?.!,]+", " ")
    sentence = tf.strings.strip(sentence)
    sentence = tf.strings.join(["[start]", sentence, "[end]"], separator=" ")
    return sentence


vectorizer = layers.TextVectorization(
    VOCAB_SIZE,
    standardize=preprocess_text,
    output_mode="int",
    output_sequence_length=MAX_LENGTH,
)

# We will adapt the vectorizer to both the questions and answers
# This dataset is batched to parallelize and speed up the process
vectorizer.adapt(tf.data.Dataset.from_tensor_slices((questions + answers)).batch(128))

使用 TextVectorization 对句子进行标记化和填充。

def vectorize_text(inputs, outputs):
    inputs, outputs = vectorizer(inputs), vectorizer(outputs)
    # One extra padding token to the right to match the output shape
    outputs = tf.pad(outputs, [[0, 1]])
    return (
        {"encoder_inputs": inputs, "decoder_inputs": outputs[:-1]},
        {"outputs": outputs[1:]},
    )


train_dataset = train_dataset.map(vectorize_text, num_parallel_calls=tf.data.AUTOTUNE)
val_dataset = val_dataset.map(vectorize_text, num_parallel_calls=tf.data.AUTOTUNE)

train_dataset = (
    train_dataset.cache()
    .shuffle(BUFFER_SIZE)
    .batch(BATCH_SIZE)
    .prefetch(tf.data.AUTOTUNE)
)
val_dataset = val_dataset.cache().batch(BATCH_SIZE).prefetch(tf.data.AUTOTUNE)

创建 FNet 编码器

FNet 论文提出了一种替代 Transformer 架构(Vaswani 等人,2017 年)中使用的标准注意力机制的方法。

Architecture

FFT 层的输出是复数。为了避免处理复数层,只提取实部(幅值)。

跟随傅里叶变换的密集层充当应用于频域的卷积。

class FNetEncoder(layers.Layer):
    def __init__(self, embed_dim, dense_dim, **kwargs):
        super().__init__(**kwargs)
        self.embed_dim = embed_dim
        self.dense_dim = dense_dim
        self.dense_proj = keras.Sequential(
            [
                layers.Dense(dense_dim, activation="relu"),
                layers.Dense(embed_dim),
            ]
        )
        self.layernorm_1 = layers.LayerNormalization()
        self.layernorm_2 = layers.LayerNormalization()

    def call(self, inputs):
        # Casting the inputs to complex64
        inp_complex = tf.cast(inputs, tf.complex64)
        # Projecting the inputs to the frequency domain using FFT2D and
        # extracting the real part of the output
        fft = tf.math.real(tf.signal.fft2d(inp_complex))
        proj_input = self.layernorm_1(inputs + fft)
        proj_output = self.dense_proj(proj_input)
        return self.layernorm_2(proj_input + proj_output)

创建解码器

解码器架构与原始 Transformer 架构(Vaswani 等人,2017 年)中提出的架构相同,包括嵌入、位置编码、两个掩码多头注意力层,最后是密集输出层。以下架构摘自 《Python 深度学习(第二版)》,第 11 章

class PositionalEmbedding(layers.Layer):
    def __init__(self, sequence_length, vocab_size, embed_dim, **kwargs):
        super().__init__(**kwargs)
        self.token_embeddings = layers.Embedding(
            input_dim=vocab_size, output_dim=embed_dim
        )
        self.position_embeddings = layers.Embedding(
            input_dim=sequence_length, output_dim=embed_dim
        )
        self.sequence_length = sequence_length
        self.vocab_size = vocab_size
        self.embed_dim = embed_dim

    def call(self, inputs):
        length = tf.shape(inputs)[-1]
        positions = tf.range(start=0, limit=length, delta=1)
        embedded_tokens = self.token_embeddings(inputs)
        embedded_positions = self.position_embeddings(positions)
        return embedded_tokens + embedded_positions

    def compute_mask(self, inputs, mask=None):
        return tf.math.not_equal(inputs, 0)


class FNetDecoder(layers.Layer):
    def __init__(self, embed_dim, latent_dim, num_heads, **kwargs):
        super().__init__(**kwargs)
        self.embed_dim = embed_dim
        self.latent_dim = latent_dim
        self.num_heads = num_heads
        self.attention_1 = layers.MultiHeadAttention(
            num_heads=num_heads, key_dim=embed_dim
        )
        self.attention_2 = layers.MultiHeadAttention(
            num_heads=num_heads, key_dim=embed_dim
        )
        self.dense_proj = keras.Sequential(
            [
                layers.Dense(latent_dim, activation="relu"),
                layers.Dense(embed_dim),
            ]
        )
        self.layernorm_1 = layers.LayerNormalization()
        self.layernorm_2 = layers.LayerNormalization()
        self.layernorm_3 = layers.LayerNormalization()
        self.supports_masking = True

    def call(self, inputs, encoder_outputs, mask=None):
        causal_mask = self.get_causal_attention_mask(inputs)
        if mask is not None:
            padding_mask = tf.cast(mask[:, tf.newaxis, :], dtype="int32")
            padding_mask = tf.minimum(padding_mask, causal_mask)

        attention_output_1 = self.attention_1(
            query=inputs, value=inputs, key=inputs, attention_mask=causal_mask
        )
        out_1 = self.layernorm_1(inputs + attention_output_1)

        attention_output_2 = self.attention_2(
            query=out_1,
            value=encoder_outputs,
            key=encoder_outputs,
            attention_mask=padding_mask,
        )
        out_2 = self.layernorm_2(out_1 + attention_output_2)

        proj_output = self.dense_proj(out_2)
        return self.layernorm_3(out_2 + proj_output)

    def get_causal_attention_mask(self, inputs):
        input_shape = tf.shape(inputs)
        batch_size, sequence_length = input_shape[0], input_shape[1]
        i = tf.range(sequence_length)[:, tf.newaxis]
        j = tf.range(sequence_length)
        mask = tf.cast(i >= j, dtype="int32")
        mask = tf.reshape(mask, (1, input_shape[1], input_shape[1]))
        mult = tf.concat(
            [tf.expand_dims(batch_size, -1), tf.constant([1, 1], dtype=tf.int32)],
            axis=0,
        )
        return tf.tile(mask, mult)


def create_model():
    encoder_inputs = keras.Input(shape=(None,), dtype="int32", name="encoder_inputs")
    x = PositionalEmbedding(MAX_LENGTH, VOCAB_SIZE, EMBED_DIM)(encoder_inputs)
    encoder_outputs = FNetEncoder(EMBED_DIM, LATENT_DIM)(x)
    encoder = keras.Model(encoder_inputs, encoder_outputs)
    decoder_inputs = keras.Input(shape=(None,), dtype="int32", name="decoder_inputs")
    encoded_seq_inputs = keras.Input(
        shape=(None, EMBED_DIM), name="decoder_state_inputs"
    )
    x = PositionalEmbedding(MAX_LENGTH, VOCAB_SIZE, EMBED_DIM)(decoder_inputs)
    x = FNetDecoder(EMBED_DIM, LATENT_DIM, NUM_HEADS)(x, encoded_seq_inputs)
    x = layers.Dropout(0.5)(x)
    decoder_outputs = layers.Dense(VOCAB_SIZE, activation="softmax")(x)
    decoder = keras.Model(
        [decoder_inputs, encoded_seq_inputs], decoder_outputs, name="outputs"
    )
    decoder_outputs = decoder([decoder_inputs, encoder_outputs])
    fnet = keras.Model([encoder_inputs, decoder_inputs], decoder_outputs, name="fnet")
    return fnet

创建和训练模型

fnet = create_model()
fnet.compile("adam", loss="sparse_categorical_crossentropy", metrics=["accuracy"])

这里,epochs 参数设置为一个 epoch,但在实践中,模型需要大约 20-30 个 epoch 的训练才能开始输出通顺的句子。虽然准确率对于这项任务来说不是一个好的衡量标准,但我们将使用它来大致了解网络的改进情况。

fnet.fit(train_dataset, epochs=1, validation_data=val_dataset)
625/625 [==============================] - 96s 133ms/step - loss: 1.3036 - accuracy: 0.4354 - val_loss: 0.7964 - val_accuracy: 0.6374

<keras.callbacks.History at 0x7f0d8d214c90>

执行推理

VOCAB = vectorizer.get_vocabulary()


def decode_sentence(input_sentence):
    # Mapping the input sentence to tokens and adding start and end tokens
    tokenized_input_sentence = vectorizer(
        tf.constant("[start] " + preprocess_text(input_sentence) + " [end]")
    )
    # Initializing the initial sentence consisting of only the start token.
    tokenized_target_sentence = tf.expand_dims(VOCAB.index("[start]"), 0)
    decoded_sentence = ""

    for i in range(MAX_LENGTH):
        # Get the predictions
        predictions = fnet.predict(
            {
                "encoder_inputs": tf.expand_dims(tokenized_input_sentence, 0),
                "decoder_inputs": tf.expand_dims(
                    tf.pad(
                        tokenized_target_sentence,
                        [[0, MAX_LENGTH - tf.shape(tokenized_target_sentence)[0]]],
                    ),
                    0,
                ),
            }
        )
        # Calculating the token with maximum probability and getting the corresponding word
        sampled_token_index = tf.argmax(predictions[0, i, :])
        sampled_token = VOCAB[sampled_token_index.numpy()]
        # If sampled token is the end token then stop generating and return the sentence
        if tf.equal(sampled_token_index, VOCAB.index("[end]")):
            break
        decoded_sentence += sampled_token + " "
        tokenized_target_sentence = tf.concat(
            [tokenized_target_sentence, [sampled_token_index]], 0
        )

    return decoded_sentence


decode_sentence("Where have you been all this time?")
'i m sorry .'

结论

此示例展示了如何使用 FNet 模型进行训练和执行推理。要深入了解架构或进一步阅读,您可以参考

  1. FNet:使用傅里叶变换混合 Token(Lee-Thorp 等人,2021 年)
  2. 注意力即一切(Vaswani 等人,2017 年)

感谢 François Chollet 提供他在 使用序列到序列 Transformer 进行英西翻译 中的 Keras 示例,解码器实现是从该示例中提取的。