import numpy as np import torch import torch.nn as nn import torch.nn.functional as F import math, copy from torch.autograd import Variable # ---------------------Encoder and Decoder Stacks-------------------- class EncoderDecoder(nn.Module): """ A standard Encoder-Decoder architecture. Base for this and many other models. """ def __init__(self, encoder, decoder, src_embed, tgt_embed, generator): super(EncoderDecoder, self).__init__() self.encoder = encoder self.decoder = decoder self.src_embed = src_embed self.tgt_embed = tgt_embed self.generator = generator def forward(self, src, tgt, src_mask, tgt_mask): "Take in and process masked src and target sequences." return self.decode(self.encode(src, src_mask), src_mask, tgt, tgt_mask) def encode(self, src, src_mask): #print("encode1:", src.shape) #temp = self.src_embed(src) #print("encode2:", temp.shape) return self.encoder(self.src_embed(src), src_mask) def decode(self, memory, src_mask, tgt, tgt_mask): #print("decode1:", memory.shape) #print("decode2:", src_mask.shape) #print("decode3:", tgt.shape) #print("decode4:", tgt_mask.shape) #temp = self.tgt_embed(tgt) #print("decode5:", temp.shape) return self.decoder(self.tgt_embed(tgt), memory, src_mask, tgt_mask) class Generator(nn.Module): "Define standard linear + softmax generation step." def __init__(self, d_model, vocab): super(Generator, self).__init__() self.proj = nn.Linear(d_model, vocab) def forward(self, x): ''' temp = self.proj(x) print("Generator1:", temp.shape) draw(3, temp[0, :, :], "Gen(lp)_" + str(2)) temp = F.log_softmax(self.proj(x), dim=-1) print("Generator2:", temp.shape) draw(3, temp[0, :, :], "Gen(sm)_" + str(3)) ''' #return F.log_softmax(self.proj(x), dim=-1) return self.proj(x) # ---------------------Encoder-------------------- def clones(module, N): "Produce N identical layers." return nn.ModuleList([copy.deepcopy(module) for _ in range(N)]) class Encoder(nn.Module): "Core encoder is a stack of N layers" def __init__(self, layer, N): super(Encoder, self).__init__() self.layers = clones(layer, N) self.norm = LayerNorm(layer.size) def forward(self, x, mask): "Pass the input (and mask) through each layer in turn." for layer in self.layers: x = layer(x, mask) return self.norm(x) class LayerNorm(nn.Module): "Construct a layernorm module (See citation for details)." def __init__(self, features, eps=1e-6): super(LayerNorm, self).__init__() self.a_2 = nn.Parameter(torch.ones(features)) self.b_2 = nn.Parameter(torch.zeros(features)) self.eps = eps def forward(self, x): mean = x.mean(-1, keepdim=True) std = x.std(-1, keepdim=True) return self.a_2 * (x - mean) / (std + self.eps) + self.b_2 class SublayerConnection(nn.Module): """ A residual connection followed by a layer norm. Note for code simplicity the norm is first as opposed to last. """ def __init__(self, size, dropout): super(SublayerConnection, self).__init__() self.norm = LayerNorm(size) self.dropout = nn.Dropout(dropout) def forward(self, x, sublayer): "Apply residual connection to any sublayer with the same size." return x + self.dropout(sublayer(self.norm(x))) class EncoderLayer(nn.Module): "Encoder is made up of self-attn and feed forward (defined below)" def __init__(self, size, self_attn, feed_forward, dropout): super(EncoderLayer, self).__init__() self.self_attn = self_attn self.feed_forward = feed_forward self.sublayer = clones(SublayerConnection(size, dropout), 2) self.size = size def forward(self, x, mask): "Follow Figure 1 (left) for connections." x = self.sublayer[0](x, lambda x: self.self_attn(x, x, x, mask)) #print("EncoderLayer2: ", x.shape) return self.sublayer[1](x, self.feed_forward) # ---------------------Decoder-------------------- class Decoder(nn.Module): "Generic N layer decoder with masking." def __init__(self, layer, N): super(Decoder, self).__init__() self.layers = clones(layer, N) self.norm = LayerNorm(layer.size) def forward(self, x, memory, src_mask, tgt_mask): for layer in self.layers: x = layer(x, memory, src_mask, tgt_mask) return self.norm(x) class DecoderLayer(nn.Module): "Decoder is made of self-attn, src-attn, and feed forward (defined below)" def __init__(self, size, self_attn, src_attn, feed_forward, dropout): super(DecoderLayer, self).__init__() self.size = size self.self_attn = self_attn self.src_attn = src_attn self.feed_forward = feed_forward self.sublayer = clones(SublayerConnection(size, dropout), 3) def forward(self, x, memory, src_mask, tgt_mask): "Follow Figure 1 (right) for connections." m = memory #print("DecoderLayer1:", x.shape) #print("DecoderLayer2:", memory.shape) x = self.sublayer[0](x, lambda x: self.self_attn(x, x, x, tgt_mask)) #print("DecoderLayer3:", src_mask.shape) #print("DecoderLayer4:", tgt_mask.shape) x = self.sublayer[1](x, lambda x: self.src_attn(x, m, m, src_mask)) #print("DecoderLayer5:", tgt_mask.shape) x = self.sublayer[2](x, self.feed_forward) #print("DecoderLayer6:", tgt_mask.shape) #time.sleep(5) return x def subsequent_mask(size): "Mask out subsequent positions." attn_shape = (1, size, size) subsequent_mask = np.triu(np.ones(attn_shape), k=1).astype('uint8') return torch.from_numpy(subsequent_mask) == 0 #---------------------Attention-------------------- def attention(query, key, value, mask=None, dropout=None): "Compute 'Scaled Dot Product Attention'" #print("attention1:", query.shape, key.transpose(-2, -1).shape, value.shape, mask.shape) d_k = query.size(-1) scores = torch.matmul(query, key.transpose(-2, -1)) \ / math.sqrt(d_k) #print("attention2:", d_k, scores.shape) #draw(3, scores[0, 0, :, :], "scores_" + str(1)) if mask is not None: scores = scores.masked_fill(mask == 0, -1e9) #draw(3, scores[0, 0, :, :], "scores_" + str(2)) p_attn = F.softmax(scores, dim = -1) #draw(3, p_attn[0, 0, :, :], "scores_" + str(3)) if dropout is not None: p_attn = dropout(p_attn) #draw(3, p_attn[0, 0, :, :], "scores_" + str(4)) return torch.matmul(p_attn, value), p_attn class MultiHeadedAttention(nn.Module): def __init__(self, h, d_model, dropout=0.1): "Take in model size and number of heads." super(MultiHeadedAttention, self).__init__() assert d_model % h == 0 # We assume d_v always equals d_k self.d_k = d_model // h self.h = h self.linears = clones(nn.Linear(d_model, d_model), 4) self.attn = None self.dropout = nn.Dropout(p=dropout) def forward(self, query, key, value, mask=None): if mask is not None: # Same mask applied to all h heads. mask = mask.unsqueeze(1) nbatches = query.size(0) #print("MultiHeadedAttention1:", query.shape, key.shape, value.shape, mask.shape) # 1) Do all the linear projections in batch from d_model => h x d_k query, key, value = \ [l(x).view(nbatches, -1, self.h, self.d_k).transpose(1, 2) for l, x in zip(self.linears, (query, key, value))] ''' print("MultiHeadedAttention2:", query.shape, key.shape, value.shape) for i in range(self.h): draw(3, query[0,i,:,:], "Query_"+str(i)) draw(3, key[0, i, :, :], "Key_" + str(i)) draw(3, value[0, i, :, :], "Value_" + str(i)) ''' # 2) Apply attention on all the projected vectors in batch. x, self.attn = attention(query, key, value, mask=mask, dropout=self.dropout) ''' print("MultiHeadedAttention3:", x.shape, self.attn.shape) for i in range(self.h): draw(3, self.attn[0, i, :, :], "Atten map_" + str(i)) draw(3, x[0, i, :, :], "X_" + str(i)) ''' # 3) "Concat" using a view and apply a final linear. x = x.transpose(1, 2).contiguous() \ .view(nbatches, -1, self.h * self.d_k) ''' print("MultiHeadedAttention4:", x.shape, self.linears[-1]) draw(3, x[0, :, :], "Concat") temp = self.linears[-1](x) print("MultiHeadedAttention5:", temp.shape) draw(3, temp[0, :, :], "Linear Proj") ''' return self.linears[-1](x) # ---------------------Position-wise Feed-Forward Networks-------------------- class PositionwiseFeedForward(nn.Module): "Implements FFN equation." def __init__(self, d_model, d_ff, dropout=0.1): super(PositionwiseFeedForward, self).__init__() self.w_1 = nn.Linear(d_model, d_ff) self.w_2 = nn.Linear(d_ff, d_model) self.dropout = nn.Dropout(dropout) def forward(self, x): ''' temp = self.w_1(x) print("PositionwiseFeedForward1:", temp.shape) draw(3, temp[0, :, :], "w_1") temp = F.relu(self.w_1(x)) print("PositionwiseFeedForward2:", temp.shape) draw(3, temp[0, :, :], "w_1_relu") temp = self.dropout(F.relu(self.w_1(x))) print("PositionwiseFeedForward3:", temp.shape) draw(3, temp[0, :, :], "dropout") temp = self.w_2(self.dropout(F.relu(self.w_1(x)))) print("PositionwiseFeedForward4:", temp.shape) draw(3, temp[0, :, :], "w_2") ''' return self.w_2(self.dropout(F.relu(self.w_1(x)))) # ---------------------Embeddings-------------------- class Embeddings(nn.Module): def __init__(self, d_model, vocab): super(Embeddings, self).__init__() self.lut = nn.Embedding(vocab, d_model) self.d_model = d_model def forward(self, x): #a = self.lut(x) #print("Embeddings:", a.shape) return self.lut(x) * math.sqrt(self.d_model) # ---------------------ExpandConv-------------------- class ExpandConv(nn.Module): def __init__(self, d_model, vocab): super(ExpandConv, self).__init__() self.lut = nn.Conv1d(in_channels=vocab, out_channels=d_model, kernel_size=1) self.d_model = d_model def forward(self, x): #print("ExpandConv:", x.shape) convoluted_x = self.lut(x) convoluted_x = convoluted_x.permute(0, 2, 1) return convoluted_x * math.sqrt(self.d_model) # ---------------------Positional Encoding-------------------- class PositionalEncoding(nn.Module): "Implement the PE function." def __init__(self, d_model, dropout, max_len=5000): super(PositionalEncoding, self).__init__() self.dropout = nn.Dropout(p=dropout) # Compute the positional encodings once in log space. pe = torch.zeros(max_len, d_model) position = torch.arange(0, max_len).unsqueeze(1) # scaling term that decreases exponentially as the depth (i.e., column index in pe) increases. div_term = torch.exp(torch.arange(0, d_model, 2) * -(math.log(10000.0) / d_model)) pe[:, 0::2] = torch.sin(position * div_term) pe[:, 1::2] = torch.cos(position * div_term) pe = pe.unsqueeze(0) self.register_buffer('pe', pe) def forward(self, x): x = x + Variable(self.pe[:, :x.size(1)], requires_grad=False) return self.dropout(x) def make_model(src_vocab, tgt_vocab, N=6, d_model=128, d_ff=2048, h=8, dropout=0.1): "Helper: Construct a model from hyperparameters." c = copy.deepcopy attn = MultiHeadedAttention(h, d_model) ff = PositionwiseFeedForward(d_model, d_ff, dropout) position = PositionalEncoding(d_model, dropout) model = EncoderDecoder( Encoder(EncoderLayer(d_model, c(attn), c(ff), dropout), N), Decoder(DecoderLayer(d_model, c(attn), c(attn), c(ff), dropout), N), #nn.Sequential(Embeddings(d_model, src_vocab), c(position)), #nn.Sequential(Embeddings(d_model, tgt_vocab), c(position)), nn.Sequential(ExpandConv(d_model, src_vocab), c(position)), nn.Sequential(ExpandConv(d_model, tgt_vocab), c(position)), Generator(d_model, tgt_vocab)) # This was important from their code. # Initialize parameters with Glorot / fan_avg. for p in model.parameters(): if p.dim() > 1: nn.init.xavier_uniform(p) return model #tmp_model = make_model(10, 10, 2) #print("all pass", tmp_model)