Attention Is All You Need-论文解读（不含实验）_attention is all you need论文下载-CSDN博客

in particular, have been firmly established as state of the art approaches in sequence modeling and transduction problems such as language modeling and machine translation. Numerous efforts have since continued to push the boundaries of recurrent language models and encoder-decoder architectures.

翻译:递归神经网络、长短期记忆和门控递归神经网络特别是在诸如语言建模和机器翻译的序列建模和转换问题中已经被牢固地确立为最先进的方法。自那以后，许多努力继续推动循环语言模型和编码器-解码器架构的边界。

原文:Recurrent models typically factor computation along the symbol positions of the input and output sequences. Aligning the positions to steps in computation time, they generate a sequence of hidden states h t, as a function of the previous hidden state h t −1 and the input for position t. This inherently sequential nature precludes parallelization within training examples, which becomes critical at longer sequence lengths, as memory constraints limit batching across examples. Recent work has achieved significant improvements in computational efficiency through factorization tricks and conditional computation , while also improving model performance in case of the latter. The fundamental constraint of sequential computation, however, remains.

翻译:递归模型通常沿着输入和输出序列的符号位置来考虑计算。将位置与计算时间中的步骤对齐，它们生成一个隐藏状态序列ht，作为先前隐藏状态ht−1和位置t的输入的函数。这种固有的顺序性排除了训练示例中的并行化，这在较长的序列长度下变得至关重要，因为内存约束限制了示例之间的批处理。最近的工作通过因子分解技巧和条件计算显著提高了计算效率，同时也提高了后者的模型性能。然而，顺序计算的约束仍然存在。

解读:当前存在的递归模型，对于生成ht时，需要先前的ht-1和当前位置的输入，导致ht的生成是依赖于前面的计算过程的。缺点:(1)由于是时序（一步步计算的过程），从而使无法做到计算并行（在计算第t位置的时候，必须要保证t-1之前的输入完成）.(2)由于是一步一步向后传递的，会可能导致前面的数据在后面的时候会被丢掉.如果想要保证不丢失，可能要使ht较大并使其在每一个时间中都要保存，内存开销会很大.

原文:Attention mechanisms have become an integral part of compelling sequence modeling and transduction models in various tasks, allowing modeling of dependencies without regard to their distance in the input or output sequences. In all but a few cases, however, such attention mechanisms are used in conjunction with a recurrent network

翻译:注意力机制已经成为各种任务中令人信服的序列建模和转导模型的一个组成部分，允许对依赖性进行建模，而不考虑它们在输入或输出序列中的距离。然而，在除少数情况外的所有情况下，这种注意力机制都与递归网络结合使用

解读:目前attention已经成功地被应用在编码器和解码器里了，主要应用在怎么样有效的把编码器的东西传递给解码器里.

原文:In this work we propose the Transformer, a model architecture eschewing recurrence and instead relying entirely on an attention mechanism to draw global dependencies between input and output. The Transformer allows for significantly more parallelization and can reach a new state of the art in translation quality after being trained for as little as twelve hours on eight P100 GPUs.

翻译:在这项工作中，我们提出了Transformer，这是一种避免重复的模型架构，而是完全依赖于注意力机制来绘制输入和输出之间的全局依赖关系。Transformer允许显著更多的并行化，并且在八个P100 GPU上训练了12个小时后，可以在翻译质量方面达到最新的水平。

三.相关工作

原文:The goal of reducing sequential computation also forms the foundation of the Extended Neural GPU,ByteNet and ConvS2S , all of which use convolutional neural networks as basic building block, computing hidden representations in parallel for all input and output positions. In these models, the number of operations required to relate signals from two arbitrary input or output positions grows in the distance between positions, linearly for ConvS2S and logarithmically for ByteNet. This makes it more difficult to learn dependencies between distant positions . In the Transformer this is reduced to a constant number of operations, albeit at the cost of reduced effective resolution due to averaging attention-weighted positions, an effect we counteract with Multi-Head Attention as described in section 3.2.

翻译:减少顺序计算的目标也构成了扩展神经GPU、ByteNet和ConvS2S的基础，所有这些都使用卷积神经网络作为基本构建块，并行计算所有输入和输出位置的隐藏表示。在这些模型中，关联来自两个任意输入或输出位置的信号所需的操作数量随着位置之间的距离而增加，对ConvS2S是线性的，对于ByteNet是对数的。这使得学习远距离位置之间的依赖关系变得更加困难。在Transformer中，这被减少到恒定数量的操作，尽管代价是由于平均注意力加权位置而降低了有效分辨率，如第3.2节所述，我们用多头注意力来抵消这种影响。

解读:在卷积神经网络中，卷积在面对长序列时，会产生累积的效果，当累计到一定程度才能到达序列的结尾，此外，卷积还具有多通道输出的效果。因此，transformer的注意力机制在处理长序列时，可以每一次看到所有并学习卷积的多通道输出（多头注意力）.

原文:Self-attention, sometimes called intra-attention is an attention mechanism relating different positions of a single sequence in order to compute a representation of the sequence. Self-attention has been used successfully in a variety of tasks including reading comprehension, abstractive summarization, textual entailment and learning task-independent sentence representations.

End-to-end memory networks are based on a recurrent attention mechanism instead of sequence aligned recurrence and have been shown to perform well on simple-language question answering and language modeling tasks.

To the best of our knowledge, however, the Transformer is the first transduction model relying
entirely on self-attention to compute representations of its input and output without using sequence aligned RNNs or convolution. In the following sections, we will describe the Transformer, motivate self-attention and discuss its advantages over models such as and.

翻译:自我注意，有时称为内部注意，是一种将单个序列的不同位置联系起来以计算序列的表示的注意机制。自注意已经成功地应用于各种任务中，包括阅读理解、抽象概括、语篇隐含和学习任务独立的句子表征。

端到端记忆网络基于递归注意力机制，而不是序列对齐的递归机制，并且已被证明在简单的语言问答和语言建模任务中表现良好。

然而，据我们所知，Transformer是第一个依赖于完全依靠自我关注来计算其输入和输出的表示，而不使用序列对齐的RNN或卷积。在以下部分中，我们将描述变压器，激励自我关注，并讨论其相对于和等模型的优势。

四.模型体系结构

原文:Most competitive neural sequence transduction models have an encoder-decoder structure .Here, the encoder maps an input sequence of symbol representations (x1, ..., xn) to a sequence of continuous representations z = (z1, ..., zn). Given z, the decoder then generates an output sequence (y1, ..., ym) of symbols one element at a time. At each step the model is auto regressive, consuming the previously generated symbols as additional input when generating the next. The Transformer follows this overall architecture using stacked self-attention and point-wise, fully connected layers for both the encoder and decoder, shown in the left and right halves of Figure 1, respectively.

翻译:大多数竞争性神经序列转导模型都具有编码器-解码器结构。这里，编码器将符号表示的输入序列（x1，…，xn）映射到连续表示的序列z＝（z1，…，zn）。给定z，解码器然后一次一个元素地生成符号的输出序列（y1，…，ym）。在每一步，模型都是自回归的，在生成下一步时，将先前生成的符号作为额外输入。Transformer遵循这一总体架构，使用堆叠的自关注和逐点、完全连接的编码器和解码器层，分别如图1的左半部分和右半部分所示。

解读:编码器的作用:输入的序列是一个句子，如“这是一个句子”，x1~xn对应的就是这个句子中的每一个字。x1对应“这”，x2对应“是”等等。那么编码器是将每一个x，也就是字变成由向量进行表示，因此z1就是x1的向量表示（变成机器学习可以理解的向量）。解码器是负责一个接一个的对z进行处理，生成输出序列（注意:n和m可能不一样长）。模型的自回归过程是指解码器通过z1生成y1，y2 的生成需要y1的输入，以此类推。如下图:左边为编码器，右边为解码器。

1.编码器和解码器堆栈

原文:Encoder: The encoder is composed of a stack of N = 6 identical layers. Each layer has two sub-layers. The first is a multi-head self-attention mechanism, and the second is a simple, position-wise fully connected feed-forward network. We employ a residual connection [10] around each of the two sub-layers, followed by layer normalization. That is, the output of each sub-layer is LayerNorm(x + Sublayer(x)), where Sublayer(x) is the function implemented by the sub-layer itself. To facilitate these residual connections, all sub-layers in the model, as well as the embedding layers, produce outputs of dimension dmodel = 512.

翻译:编码器：编码器由N=6个相同层的堆栈组成。每层有两个子层。第一种是多头自注意机制，第二种是简单的位置全连接前馈网络。我们在两个子层中的每一个子层周围使用残差连接[10]，然后进行层归一化。也就是说，每个子层的输出是LayerNorm（x+Sublayer（x）），其中Sublayer（x）是由子层本身实现的函数。为了促进这些残余连接，模型中的所有子层以及嵌入层产生维度dmodel=512的输出。

解读:左侧的编码器，如图所示，这个是一个层，在编码器中有六个这种结构。那么在这种结构中也叫层，每层中都存在两个子层，分别为多头自注意力机制和位置全连接前馈网络（mlp）。

原文:Decoder: The decoder is also composed of a stack of N = 6 identical layers. In addition to the two sub-layers in each encoder layer, the decoder inserts a third sub-layer, which performs multi-head attention over the output of the encoder stack. Similar to the encoder, we employ residual connections around each of the sub-layers, followed by layer normalization. We also modify the self-attention sub-layer in the decoder stack to prevent positions from attending to subsequent positions. This masking, combined with fact that the output embeddings are offset by one position, ensures that the predictions for position i can depend only on the known outputs at positions less than.

翻译:解码器：解码器也是由N=6个相同层的堆栈组成的。除了每个编码器层中的两个子层之外，解码器还插入第三个子层，该第三子层对编码器堆栈的输出执行多头关注。与编码器类似，我们在每个子层周围使用残差连接，然后进行层归一化。我们还修改了解码器堆栈中的自注意子层，以防止位置关注后续位置。这种掩蔽，再加上输出嵌入偏移一个位置的事实，确保了对位置i的预测只能取决于小于的位置处的已知输出。

解读:在下面的解码器中，我们要注意红色框中的内容，它实现了在t时间的时候不会看到t时间之后的输入，以保证训练和预测的时候行为一致。

2.注意力

原文: An attention function can be described as mapping a query and a set of key-value pairs to an output, where the query, keys, values, and output are all vectors. The output is computed as a weighted sum of the values, where the weight assigned to each value is computed by a compatibility function of the query with the corresponding key.

翻译:注意力函数可以描述为将查询和一组键值对映射到输出，其中查询、键、值和输出都是向量。输出被计算为值的加权和，其中分配给每个值的权重由查询与相应关键字的兼容性函数计算。

2.1缩放点产品注意力

原文:We call our particular attention "Scaled Dot-Product Attention" (Figure 2). The input consists of queries and keys of dimension dk, and values of dimension dv. We compute the dot products of the Figure 2: (left) Scaled Dot-Product Attention. (right) Multi-Head Attention consists of several attention layers running in parallel. query with all keys, divide each by √ dk, and apply a softmax function to obtain the weights on the values. In practice, we compute the attention function on a set of queries simultaneously, packed together into a matrix Q. The keys and values are also packed together into matrices K and V . We compute the matrix of outputs as:

$eq?Attention%28Q%2CK%2CV%29%20%3D%20softmax%28%5Cfrac%7BQK%5E%7BT%7D%7D%7B%5Csqrt%7Bd_%7Bk%7D%7D%7D%29V$

The two most commonly used attention functions are additive attention, and dot-product (multiplicative) attention. Dot-product attention is identical to our algorithm, except for the scaling factor of $eq?%5Cfrac%7B1%7D%7B%5Csqrt%7Bd_%7Bk%7D%7D%7D$ .Additive attention computes the compatibility function using a feed-forward network with a single hidden layer. While the two are similar in theoretical complexity, dot-product attention is much faster and more space-efficient in practice, since it can be implemented using highly optimized matrix multiplication code. While for small values of dk the two mechanisms perform similarly, additive attention outperforms dot product attention without scaling for larger values of d k . We suspect that for large values of d k, the dot products grow large in magnitude, pushing the softmax function into regions where it has extremely small gradients. To counteract this effect, we scale the dot products by $eq?%5Cfrac%7B1%7D%7B%5Csqrt%7Bd_%7Bk%7D%7D%7D$

翻译:我们将我们的特别关注称为“标度点产品关注”（图2）。输入包括维度dk的查询和键以及维度dv的值。我们计算图2的点积：（左）缩放点积注意力。（右）多头注意力由几个平行运行的注意力层组成。使用所有键进行查询，将每个键除以√dk，然后应用softmax函数来获得值的权重。在实践中，我们同时计算一组查询的注意力函数，并将其打包成矩阵Q。键和值也打包成矩阵K和V。我们将输出矩阵计算为：

$eq?Attention%28Q%2CK%2CV%29%20%3D%20softmax%28%5Cfrac%7BQK%5E%7BT%7D%7D%7B%5Csqrt%7Bd_%7Bk%7D%7D%7D%29V$

两个最常用的注意力函数是加性注意力和点积（乘法）注意力。点积注意力与我们的算法相同，只是比例因子为。加性注意力使用具有单个隐藏层的前馈网络来计算兼容性函数。虽然两者在理论复杂性上相似，但点积注意力在实践中要快得多，空间效率更高，因为它可以使用高度优化的矩阵乘法代码来实现。虽然对于较小的dk值，这两种机制的表现相似，但在不缩放较大的dk的情况下，相加注意力优于点积注意力。我们怀疑，对于较大的dk值，点积的大小会变大，从而将softmax函数推向具有极小梯度的区域。为了抵消这种影响，我们按比例缩放点积.

解读:在下图中，由于attention在处理ht时，可以获取到ht之前的内容和之后的内容，那么为了防止ht之后的内容对结果的影响，通过下方的mask使ht之后的权重很小，在经过softmax后就会等于0.

2.2多头注意力

原文:Instead of performing a single attention function with dmodel-dimensional keys, values and queries,we found it beneficial to linearly project the queries, keys and values h times with different, learned linear projections to dk, dk and dv dimensions, respectively. On each of these projected versions of queries, keys and values we then perform the attention function in parallel, yielding dv-dimensional output values. These are concatenated and once again projected, resulting in the final values, as depicted in Figure 2.Multi-head attention allows the model to jointly attend to information from different representation subspaces at different positions. With a single attention head, averaging inhibits this.

$eq?MultiHead%28Q%2C%20K%2C%20V%20%29%20%3D%20Concat%28head_%7B1%7D%2C%20...%2C%20head_%7Bh%7D%29W%5E%7BO%7D$

$eq?where%20head_%7Bi%7D%20%3D%20Attention%28QW_%7Bi%7D%5E%7BQ%7D%2C%20KW_%7Bi%7D%5E%7BK%7D%2C%20V%20W_%7Bi%7D%5E%7BV%7D%29$

Where the projections are parameter matrices W i Q ∈ R dmodel × d k , W i K ∈ R dmodel × d k , W i V ∈ R dmodel × d v and W O ∈ R hd v × dmodel .In this work we employ h = 8 parallel attention layers, or heads. For each of these we use d k = d v = dmodel /h = 64. Due to the reduced dimension of each head, the total computational cost is similar to that of single-head attention with full dimensionality.

翻译:我们发现，与使用dmodel维度的键、值和查询执行单个注意力函数不同，将查询、键和值分别线性投影到dk、dk和dv维度的不同学习线性投影h次是有益的。然后，在这些查询、键和值的投影版本中的每一个上，我们并行执行注意力函数，生成dv维输出值。这些被连接起来并再次投影，产生最终值，如图2所示。多头注意力允许模型联合关注来自不同位置的不同表示子空间的信息。对于一个注意力集中的头部，平均值会抑制这种情况。

$eq?MultiHead%28Q%2C%20K%2C%20V%20%29%20%3D%20Concat%28head_%7B1%7D%2C%20...%2C%20head_%7Bh%7D%29W%5E%7BO%7D$

$eq?where%20head_%7Bi%7D%20%3D%20Attention%28QW_%7Bi%7D%5E%7BQ%7D%2C%20KW_%7Bi%7D%5E%7BK%7D%2C%20V%20W_%7Bi%7D%5E%7BV%7D%29$

其中投影是参数矩阵Wi-Q∈Rdmodel×dk，Wi-K∈R dmodel x dk，Wi-Fi∈R dm odel×dv和WO∈R hdv×dmodel。在这项工作中，我们使用了h=8个平行的注意力层或头部。对于其中的每一个，我们使用dk=dv=dmodel/h=64。由于每个头部的维数降低，总计算成本与全维度的单头注意力相似。

2.3注意点在我们的模型中的应用

原文:The Transformer uses multi-head attention in three different ways:

In "encoder-decoder attention" layers, the queries come from the previous decoder layer,

and the memory keys and values come from the output of the encoder. This allows every

position in the decoder to attend over all positions in the input sequence. This mimics the

typical encoder-decoder attention mechanisms in sequence-to-sequence models such as
The encoder contains self-attention layers. In a self-attention layer all of the keys, values

and queries come from the same place, in this case, the output of the previous layer in the

encoder. Each position in the encoder can attend to all positions in the previous layer of the

encoder.
Similarly, self-attention layers in the decoder allow each position in the decoder to attend to

all positions in the decoder up to and including that position. We need to prevent leftward

information flow in the decoder to preserve the auto-regressive property. We implement this

inside of scaled dot-product attention by masking out (setting to −∞) all values in the input

of the softmax which correspond to illegal connections. See Figure 2.

翻译: Transformer以三种不同的方式使用多头注意力：

在“编码器-解码器注意力”层中，查询来自前一个解码器层，并且存储器密钥和值来自编码器的输出。这允许解码器中的位置，以关注输入序列中的所有位置。这模仿了序列到序列模型中的典型编码器-解码器注意力机制，例如
编码器包含自关注层。在自我关注层中，所有的键、值和查询来自同一个位置，在这种情况下，是中上一层的输出编码器。编码器中的每个位置都可以处理的前一层中的所有位置编码器。
类似地，解码器中的自关注层允许解码器中的每个位置关注解码器中直到并包括该位置的所有位置。我们需要防止向左解码器中的信息流以保持自回归特性。我们实施这一点通过屏蔽（设置为-∞）输入中的所有值来引起缩放点积内部的注意对应于非法连接的softmax的。见图2。

解读: 在下方的编码器结构中，我们的输入复制成为了三份，分别作为key，value，query，本质上输出就是与输入的不同相似度。解码器输入也是，masked前面也提到过，不让他看到后面的东西。但解码器中的Muti-Head Attention，输入的key和value来自编码器，query来自解码器，

3.位置级的前馈网络

原文:In addition to attention sub-layers, each of the layers in our encoder and decoder contains a fully connected feed-forward network, which is applied to each position separately and identically. This consists of two linear transformations with a ReLU activation in between

$eq?FFN%28x%29%20%3D%20max%280%2C%20xW_%7B1%7D%20+%20b_%7B1%7D%29%20W_%7B2%7D%20+%20b_%7B2%7D$

While the linear transformations are the same across different positions, they use different parameters from layer to layer. Another way of describing this is as two convolutions with kernel size 1. The dimensionality of input and output is dmodel = 512, and the inner-layer has dimensionality d f f = 2048.

翻译:除了注意力子层之外，我们的编码器和解码器中的每个层都包含一个完全连接的前馈网络，该网络分别且相同地应用于每个位置。这包括两个线性变换，其间有ReLU激活

$eq?FFN%28x%29%20%3D%20max%280%2C%20xW_%7B1%7D%20+%20b_%7B1%7D%29%20W_%7B2%7D%20+%20b_%7B2%7D$

虽然线性变换在不同的位置上是相同的，但它们在不同的层之间使用不同的参数。另一种描述方式是将其描述为核大小为1的两个卷积。输入和输出的维度为dmodel=512，内层的维度dff=2048。

4.嵌入和softmax

原文:Similarly to other sequence transduction models, we use learned embeddings to convert the input tokens and output tokens to vectors of dimension dmodel. We also use the usual learned linear transformation and softmax function to convert the decoder output to predicted next-token probabilities. In our model, we share the same weight matrix between the two embedding layers and the pre-softmax linear transformation, similar to. In the embedding layers, we multiply those weights by √ dmodel.

翻译:与其他序列转导模型类似，我们使用学习嵌入将输入令牌和输出令牌转换为维度dmodel的向量。我们还使用通常学习的线性变换和softmax函数将解码器输出转换为预测的下一个令牌概率。在我们的模型中，我们在两个嵌入层和预softmax线性变换之间共享相同的权重矩阵，类似于。在嵌入层中，我们将这些权重乘以√dmodel。

5.位置编码

原文:Since our model contains no recurrence and no convolution, in order for the model to make use of the order of the sequence, we must inject some information about the relative or absolute position of the tokens in the sequence. To this end, we add "positional encodings" to the input embeddings at the Table 1: Maximum path lengths, per-layer complexity and minimum number of sequential operations for different layer types. n is the sequence length, d is the representation dimension, k is the kernel size of convolutions and r the size of the neighborhood in restricted self-attention.bottoms of the encoder and decoder stacks. The positional encodings have the same dimension dmodel as the embeddings, so that the two can be summed. There are many choices of positional encodings, learned and fixed. In this work, we use sine and cosine functions of different frequencies:

where pos is the position and i is the dimension. That is, each dimension of the positional encoding corresponds to a sinusoid. The wavelengths form a geometric progression from 2π to 10000 · 2π. We chose this function because we hypothesized it would allow the model to easily learn to attend by relative positions, since for any fixed offset k, P Epos+k can be represented as a linear function of P Epos. We also experimented with using learned positional embeddings [8] instead, and found that the two versions produced nearly identical results (see Table 3 row (E)). We chose the sinusoidal version because it may allow the model to extrapolate to sequence lengths longer than the ones encountered during training.

翻译:由于我们的模型不包含递归和卷积，为了让模型利用序列的顺序，我们必须注入一些关于令牌在序列中的相对或绝对位置的信息。为此，我们在表1的输入嵌入中添加了“位置编码”：不同层类型的最大路径长度、每层复杂性和最小顺序操作数。n是序列长度，d是表示维数，k是卷积的核大小，r是编码器和解码器堆栈的限制自关注基底中的邻域的大小。位置编码与嵌入具有相同的维度dmodel，因此可以将两者相加。有许多位置编码的选择，有学习的和固定的。在这项工作中，我们使用不同频率的正弦和余弦函数：

其中pos是位置，i是尺寸。也就是说，位置编码的每个维度对应于正弦曲线。波长形成从2π到10000·2π的几何级数。我们选择这个函数是因为我们假设它可以让模型很容易地学会通过相对位置来关注，因为对于任何固定的偏移k，P Epos+k可以表示为P Epos的线性函数。我们还使用学习的位置嵌入[8]进行了实验，发现这两个版本产生了几乎相同的结果（见表3第（E）行）。我们选择正弦版本是因为它可以允许模型外推到比训练期间遇到的序列长度更长的序列长度。