# Attention Scoring Functions :label:`sec_attention-scoring-functions` In :numref:`sec_nadaraya-waston`, we used a Gaussian kernel to model interactions between queries and keys. Treating the exponent of the Gaussian kernel in :eqref:`eq_nadaraya-waston-gaussian` as an *attention scoring function* (or *scoring function* for short), the results of this function were essentially fed into a softmax operation. As a result, we obtained a probability distribution (attention weights) over values that are paired with keys. In the end, the output of the attention pooling is simply a weighted sum of the values based on these attention weights. At a high level, we can use the above algorithm to instantiate the framework of attention mechanisms in :numref:`fig_qkv`. Denoting an attention scoring function by $a$, :numref:`fig_attention_output` illustrates how the output of attention pooling can be computed as a weighted sum of values. Since attention weights are a probability distribution, the weighted sum is essentially a weighted average. ![Computing the output of attention pooling as a weighted average of values.](../img/attention-output.svg) :label:`fig_attention_output` Mathematically, suppose that we have a query $\mathbf{q} \in \mathbb{R}^q$ and $m$ key-value pairs $(\mathbf{k}_1, \mathbf{v}_1), \ldots, (\mathbf{k}_m, \mathbf{v}_m)$, where any $\mathbf{k}_i \in \mathbb{R}^k$ and any $\mathbf{v}_i \in \mathbb{R}^v$. The attention pooling $f$ is instantiated as a weighted sum of the values: $$f(\mathbf{q}, (\mathbf{k}_1, \mathbf{v}_1), \ldots, (\mathbf{k}_m, \mathbf{v}_m)) = \sum_{i=1}^m \alpha(\mathbf{q}, \mathbf{k}_i) \mathbf{v}_i \in \mathbb{R}^v,$$ :eqlabel:`eq_attn-pooling` where the attention weight (scalar) for the query $\mathbf{q}$ and key $\mathbf{k}_i$ is computed by the softmax operation of an attention scoring function $a$ that maps two vectors to a scalar: $$\alpha(\mathbf{q}, \mathbf{k}_i) = \mathrm{softmax}(a(\mathbf{q}, \mathbf{k}_i)) = \frac{\exp(a(\mathbf{q}, \mathbf{k}_i))}{\sum_{j=1}^m \exp(a(\mathbf{q}, \mathbf{k}_j))} \in \mathbb{R}.$$ :eqlabel:`eq_attn-scoring-alpha` As we can see, different choices of the attention scoring function $a$ lead to different behaviors of attention pooling. In this section, we introduce two popular scoring functions that we will use to develop more sophisticated attention mechanisms later. ```{.python .input} import math from d2l import mxnet as d2l from mxnet import np, npx from mxnet.gluon import nn npx.set_np() ``` ```{.python .input} #@tab pytorch from d2l import torch as d2l import math import torch from torch import nn ``` ## Masked Softmax Operation As we just mentioned, a softmax operation is used to output a probability distribution as attention weights. In some cases, not all the values should be fed into attention pooling. For instance, for efficient minibatch processing in :numref:`sec_machine_translation`, some text sequences are padded with special tokens that do not carry meaning. To get an attention pooling over only meaningful tokens as values, we can specify a valid sequence length (in number of tokens) to filter out those beyond this specified range when computing softmax. In this way, we can implement such a *masked softmax operation* in the following `masked_softmax` function, where any value beyond the valid length is masked as zero. ```{.python .input} #@save def masked_softmax(X, valid_lens): """Perform softmax operation by masking elements on the last axis.""" # `X`: 3D tensor, `valid_lens`: 1D or 2D tensor if valid_lens is None: return npx.softmax(X) else: shape = X.shape if valid_lens.ndim == 1: valid_lens = valid_lens.repeat(shape[1]) else: valid_lens = valid_lens.reshape(-1) # On the last axis, replace masked elements with a very large negative # value, whose exponentiation outputs 0 X = npx.sequence_mask(X.reshape(-1, shape[-1]), valid_lens, True, value=-1e6, axis=1) return npx.softmax(X).reshape(shape) ``` ```{.python .input} #@tab pytorch #@save def masked_softmax(X, valid_lens): """Perform softmax operation by masking elements on the last axis.""" # `X`: 3D tensor, `valid_lens`: 1D or 2D tensor if valid_lens is None: return nn.functional.softmax(X, dim=-1) else: shape = X.shape if valid_lens.dim() == 1: valid_lens = torch.repeat_interleave(valid_lens, shape[1]) else: valid_lens = valid_lens.reshape(-1) # On the last axis, replace masked elements with a very large negative # value, whose exponentiation outputs 0 X = d2l.sequence_mask(X.reshape(-1, shape[-1]), valid_lens, value=-1e6) return nn.functional.softmax(X.reshape(shape), dim=-1) ``` To demonstrate how this function works, consider a minibatch of two $2 \times 4$ matrix examples, where the valid lengths for these two examples are two and three, respectively. As a result of the masked softmax operation, values beyond the valid lengths are all masked as zero. ```{.python .input} masked_softmax(np.random.uniform(size=(2, 2, 4)), d2l.tensor([2, 3])) ``` ```{.python .input} #@tab pytorch masked_softmax(torch.rand(2, 2, 4), torch.tensor([2, 3])) ``` Similarly, we can also use a two-dimensional tensor to specify valid lengths for every row in each matrix example. ```{.python .input} masked_softmax(np.random.uniform(size=(2, 2, 4)), d2l.tensor([[1, 3], [2, 4]])) ``` ```{.python .input} #@tab pytorch masked_softmax(torch.rand(2, 2, 4), d2l.tensor([[1, 3], [2, 4]])) ``` ## Additive Attention :label:`subsec_additive-attention` In general, when queries and keys are vectors of different lengths, we can use additive attention as the scoring function. Given a query $\mathbf{q} \in \mathbb{R}^q$ and a key $\mathbf{k} \in \mathbb{R}^k$, the *additive attention* scoring function $$a(\mathbf q, \mathbf k) = \mathbf w_v^\top \text{tanh}(\mathbf W_q\mathbf q + \mathbf W_k \mathbf k) \in \mathbb{R},$$ :eqlabel:`eq_additive-attn` where learnable parameters $\mathbf W_q\in\mathbb R^{h\times q}$, $\mathbf W_k\in\mathbb R^{h\times k}$, and $\mathbf w_v\in\mathbb R^{h}$. Equivalent to :eqref:`eq_additive-attn`, the query and the key are concatenated and fed into an MLP with a single hidden layer whose number of hidden units is $h$, a hyperparameter. By using $\tanh$ as the activation function and disabling bias terms, we implement additive attention in the following. ```{.python .input} #@save class AdditiveAttention(nn.Block): """Additive attention.""" def __init__(self, num_hiddens, dropout, **kwargs): super(AdditiveAttention, self).__init__(**kwargs) # Use `flatten=False` to only transform the last axis so that the # shapes for the other axes are kept the same self.W_k = nn.Dense(num_hiddens, use_bias=False, flatten=False) self.W_q = nn.Dense(num_hiddens, use_bias=False, flatten=False) self.w_v = nn.Dense(1, use_bias=False, flatten=False) self.dropout = nn.Dropout(dropout) def forward(self, queries, keys, values, valid_lens): queries, keys = self.W_q(queries), self.W_k(keys) # After dimension expansion, shape of `queries`: (`batch_size`, no. of # queries, 1, `num_hiddens`) and shape of `keys`: (`batch_size`, 1, # no. of key-value pairs, `num_hiddens`). Sum them up with # broadcasting features = np.expand_dims(queries, axis=2) + np.expand_dims( keys, axis=1) features = np.tanh(features) # There is only one output of `self.w_v`, so we remove the last # one-dimensional entry from the shape. Shape of `scores`: # (`batch_size`, no. of queries, no. of key-value pairs) scores = np.squeeze(self.w_v(features), axis=-1) self.attention_weights = masked_softmax(scores, valid_lens) # Shape of `values`: (`batch_size`, no. of key-value pairs, value # dimension) return npx.batch_dot(self.dropout(self.attention_weights), values) ``` ```{.python .input} #@tab pytorch #@save class AdditiveAttention(nn.Module): def __init__(self, key_size, query_size, num_hiddens, dropout, **kwargs): super(AdditiveAttention, self).__init__(**kwargs) self.W_k = nn.Linear(key_size, num_hiddens, bias=False) self.W_q = nn.Linear(query_size, num_hiddens, bias=False) self.w_v = nn.Linear(num_hiddens, 1, bias=False) self.dropout = nn.Dropout(dropout) def forward(self, queries, keys, values, valid_lens): queries, keys = self.W_q(queries), self.W_k(keys) # After dimension expansion, shape of `queries`: (`batch_size`, no. of # queries, 1, `num_hiddens`) and shape of `keys`: (`batch_size`, 1, # no. of key-value pairs, `num_hiddens`). Sum them up with # broadcasting features = queries.unsqueeze(2) + keys.unsqueeze(1) features = torch.tanh(features) # There is only one output of `self.w_v`, so we remove the last # one-dimensional entry from the shape. Shape of `scores`: # (`batch_size`, no. of queries, no. of key-value pairs) scores = self.w_v(features).squeeze(-1) self.attention_weights = masked_softmax(scores, valid_lens) # Shape of `values`: (`batch_size`, no. of key-value pairs, value # dimension) return torch.bmm(self.dropout(self.attention_weights), values) ``` Let us demonstrate the above `AdditiveAttention` class with a toy example, where shapes (batch size, number of steps or sequence length in tokens, feature size) of queries, keys, and values are ($2$, $1$, $20$), ($2$, $10$, $2$), and ($2$, $10$, $4$), respectively. The attention pooling output has a shape of (batch size, number of steps for queries, feature size for values). ```{.python .input} queries, keys = d2l.normal(0, 1, (2, 1, 20)), d2l.ones((2, 10, 2)) # The two value matrices in the `values` minibatch are identical values = np.arange(40).reshape(1, 10, 4).repeat(2, axis=0) valid_lens = d2l.tensor([2, 6]) attention = AdditiveAttention(num_hiddens=8, dropout=0.1) attention.initialize() attention(queries, keys, values, valid_lens) ``` ```{.python .input} #@tab pytorch queries, keys = d2l.normal(0, 1, (2, 1, 20)), d2l.ones((2, 10, 2)) # The two value matrices in the `values` minibatch are identical values = torch.arange(40, dtype=torch.float32).reshape(1, 10, 4).repeat( 2, 1, 1) valid_lens = d2l.tensor([2, 6]) attention = AdditiveAttention(key_size=2, query_size=20, num_hiddens=8, dropout=0.1) attention.eval() attention(queries, keys, values, valid_lens) ``` Although additive attention contains learnable parameters, since every key is the same in this example, the attention weights are uniform, determined by the specified valid lengths. ```{.python .input} #@tab all d2l.show_heatmaps(d2l.reshape(attention.attention_weights, (1, 1, 2, 10)), xlabel='Keys', ylabel='Queries') ``` ## Scaled Dot-Product Attention A more computationally efficient design for the scoring function can be simply dot product. However, the dot product operation requires that both the query and the key have the same vector length, say $d$. Assume that all the elements of the query and the key are independent random variables with zero mean and unit variance. The dot product of both vectors has zero mean and a variance of $d$. To ensure that the variance of the dot product still remains one regardless of vector length, the *scaled dot-product attention* scoring function $$a(\mathbf q, \mathbf k) = \mathbf{q}^\top \mathbf{k} /\sqrt{d}$$ divides the dot product by $\sqrt{d}$. In practice, we often think in minibatches for efficiency, such as computing attention for $n$ queries and $m$ key-value pairs, where queries and keys are of length $d$ and values are of length $v$. The scaled dot-product attention of queries $\mathbf Q\in\mathbb R^{n\times d}$, keys $\mathbf K\in\mathbb R^{m\times d}$, and values $\mathbf V\in\mathbb R^{m\times v}$ is $$ \mathrm{softmax}\left(\frac{\mathbf Q \mathbf K^\top }{\sqrt{d}}\right) \mathbf V \in \mathbb{R}^{n\times v}.$$ :eqlabel:`eq_softmax_QK_V` In the following implementation of the scaled dot product attention, we use dropout for model regularization. ```{.python .input} #@save class DotProductAttention(nn.Block): """Scaled dot product attention.""" def __init__(self, dropout, **kwargs): super(DotProductAttention, self).__init__(**kwargs) self.dropout = nn.Dropout(dropout) # Shape of `queries`: (`batch_size`, no. of queries, `d`) # Shape of `keys`: (`batch_size`, no. of key-value pairs, `d`) # Shape of `values`: (`batch_size`, no. of key-value pairs, value # dimension) # Shape of `valid_lens`: (`batch_size`,) or (`batch_size`, no. of queries) def forward(self, queries, keys, values, valid_lens=None): d = queries.shape[-1] # Set `transpose_b=True` to swap the last two dimensions of `keys` scores = npx.batch_dot(queries, keys, transpose_b=True) / math.sqrt(d) self.attention_weights = masked_softmax(scores, valid_lens) return npx.batch_dot(self.dropout(self.attention_weights), values) ``` ```{.python .input} #@tab pytorch #@save class DotProductAttention(nn.Module): """Scaled dot product attention.""" def __init__(self, dropout, **kwargs): super(DotProductAttention, self).__init__(**kwargs) self.dropout = nn.Dropout(dropout) # Shape of `queries`: (`batch_size`, no. of queries, `d`) # Shape of `keys`: (`batch_size`, no. of key-value pairs, `d`) # Shape of `values`: (`batch_size`, no. of key-value pairs, value # dimension) # Shape of `valid_lens`: (`batch_size`,) or (`batch_size`, no. of queries) def forward(self, queries, keys, values, valid_lens=None): d = queries.shape[-1] # Set `transpose_b=True` to swap the last two dimensions of `keys` scores = torch.bmm(queries, keys.transpose(1,2)) / math.sqrt(d) self.attention_weights = masked_softmax(scores, valid_lens) return torch.bmm(self.dropout(self.attention_weights), values) ``` To demonstrate the above `DotProductAttention` class, we use the same keys, values, and valid lengths from the earlier toy example for additive attention. For the dot product operation, we make the feature size of queries the same as that of keys. ```{.python .input} queries = d2l.normal(0, 1, (2, 1, 2)) attention = DotProductAttention(dropout=0.5) attention.initialize() attention(queries, keys, values, valid_lens) ``` ```{.python .input} #@tab pytorch queries = d2l.normal(0, 1, (2, 1, 2)) attention = DotProductAttention(dropout=0.5) attention.eval() attention(queries, keys, values, valid_lens) ``` Same as in the additive attention demonstration, since `keys` contains the same element that cannot be differentiated by any query, uniform attention weights are obtained. ```{.python .input} #@tab all d2l.show_heatmaps(d2l.reshape(attention.attention_weights, (1, 1, 2, 10)), xlabel='Keys', ylabel='Queries') ``` ## Summary * We can compute the output of attention pooling as a weighted average of values, where different choices of the attention scoring function lead to different behaviors of attention pooling. * When queries and keys are vectors of different lengths, we can use the additive attention scoring function. When they are the same, the scaled dot-product attention scoring function is more computationally efficient. ## Exercises 1. Modify keys in the toy example and visualize attention weights. Do additive attention and scaled dot-product attention still output the same attention weights? Why or why not? 1. Using matrix multiplications only, can you design a new scoring function for queries and keys with different vector lengths? 1. When queries and keys have the same vector length, is vector summation a better design than dot product for the scoring function? Why or why not? :begin_tab:`mxnet` [Discussions](https://discuss.d2l.ai/t/346) :end_tab: :begin_tab:`pytorch` [Discussions](https://discuss.d2l.ai/t/1064) :end_tab: