chore: import upstream snapshot with attribution
This commit is contained in:
wehub-resource-sync
@@ -0,0 +1,282 @@
|
||||
{
|
||||
"cells": [
|
||||
{
|
||||
"cell_type": "code",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"# code by Tae Hwan Jung(Jeff Jung) @graykode, Derek Miller @dmmiller612\n",
|
||||
"# Reference : https://github.com/jadore801120/attention-is-all-you-need-pytorch\n",
|
||||
"# https://github.com/JayParks/transformer\n",
|
||||
"import numpy as np\n",
|
||||
"import torch\n",
|
||||
"import torch.nn as nn\n",
|
||||
"import torch.optim as optim\n",
|
||||
"import matplotlib.pyplot as plt\n",
|
||||
"\n",
|
||||
"# S: Symbol that shows starting of decoding input\n",
|
||||
"# E: Symbol that shows starting of decoding output\n",
|
||||
"# P: Symbol that will fill in blank sequence if current batch data size is short than time steps\n",
|
||||
"\n",
|
||||
"def make_batch():\n",
|
||||
" input_batch = [[src_vocab[n] for n in sentences[0].split()]]\n",
|
||||
" output_batch = [[tgt_vocab[n] for n in sentences[1].split()]]\n",
|
||||
" target_batch = [[tgt_vocab[n] for n in sentences[2].split()]]\n",
|
||||
" return torch.LongTensor(input_batch), torch.LongTensor(output_batch), torch.LongTensor(target_batch)\n",
|
||||
"\n",
|
||||
"def get_sinusoid_encoding_table(n_position, d_model):\n",
|
||||
" def cal_angle(position, hid_idx):\n",
|
||||
" return position / np.power(10000, 2 * (hid_idx // 2) / d_model)\n",
|
||||
" def get_posi_angle_vec(position):\n",
|
||||
" return [cal_angle(position, hid_j) for hid_j in range(d_model)]\n",
|
||||
"\n",
|
||||
" sinusoid_table = np.array([get_posi_angle_vec(pos_i) for pos_i in range(n_position)])\n",
|
||||
" sinusoid_table[:, 0::2] = np.sin(sinusoid_table[:, 0::2]) # dim 2i\n",
|
||||
" sinusoid_table[:, 1::2] = np.cos(sinusoid_table[:, 1::2]) # dim 2i+1\n",
|
||||
" return torch.FloatTensor(sinusoid_table)\n",
|
||||
"\n",
|
||||
"def get_attn_pad_mask(seq_q, seq_k):\n",
|
||||
" # print(seq_q)\n",
|
||||
" batch_size, len_q = seq_q.size()\n",
|
||||
" batch_size, len_k = seq_k.size()\n",
|
||||
" # eq(zero) is PAD token\n",
|
||||
" pad_attn_mask = seq_k.data.eq(0).unsqueeze(1) # batch_size x 1 x len_k(=len_q), one is masking\n",
|
||||
" return pad_attn_mask.expand(batch_size, len_q, len_k) # batch_size x len_q x len_k\n",
|
||||
"\n",
|
||||
"def get_attn_subsequent_mask(seq):\n",
|
||||
" attn_shape = [seq.size(0), seq.size(1), seq.size(1)]\n",
|
||||
" subsequent_mask = np.triu(np.ones(attn_shape), k=1)\n",
|
||||
" subsequent_mask = torch.from_numpy(subsequent_mask).byte()\n",
|
||||
" return subsequent_mask\n",
|
||||
"\n",
|
||||
"class ScaledDotProductAttention(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(ScaledDotProductAttention, self).__init__()\n",
|
||||
"\n",
|
||||
" def forward(self, Q, K, V, attn_mask):\n",
|
||||
" scores = torch.matmul(Q, K.transpose(-1, -2)) / np.sqrt(d_k) # scores : [batch_size x n_heads x len_q(=len_k) x len_k(=len_q)]\n",
|
||||
" scores.masked_fill_(attn_mask, -1e9) # Fills elements of self tensor with value where mask is one.\n",
|
||||
" attn = nn.Softmax(dim=-1)(scores)\n",
|
||||
" context = torch.matmul(attn, V)\n",
|
||||
" return context, attn\n",
|
||||
"\n",
|
||||
"class MultiHeadAttention(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(MultiHeadAttention, self).__init__()\n",
|
||||
" self.W_Q = nn.Linear(d_model, d_k * n_heads)\n",
|
||||
" self.W_K = nn.Linear(d_model, d_k * n_heads)\n",
|
||||
" self.W_V = nn.Linear(d_model, d_v * n_heads)\n",
|
||||
" self.linear = nn.Linear(n_heads * d_v, d_model)\n",
|
||||
" self.layer_norm = nn.LayerNorm(d_model)\n",
|
||||
"\n",
|
||||
" def forward(self, Q, K, V, attn_mask):\n",
|
||||
" # q: [batch_size x len_q x d_model], k: [batch_size x len_k x d_model], v: [batch_size x len_k x d_model]\n",
|
||||
" residual, batch_size = Q, Q.size(0)\n",
|
||||
" # (B, S, D) -proj-> (B, S, D) -split-> (B, S, H, W) -trans-> (B, H, S, W)\n",
|
||||
" q_s = self.W_Q(Q).view(batch_size, -1, n_heads, d_k).transpose(1,2) # q_s: [batch_size x n_heads x len_q x d_k]\n",
|
||||
" k_s = self.W_K(K).view(batch_size, -1, n_heads, d_k).transpose(1,2) # k_s: [batch_size x n_heads x len_k x d_k]\n",
|
||||
" v_s = self.W_V(V).view(batch_size, -1, n_heads, d_v).transpose(1,2) # v_s: [batch_size x n_heads x len_k x d_v]\n",
|
||||
"\n",
|
||||
" attn_mask = attn_mask.unsqueeze(1).repeat(1, n_heads, 1, 1) # attn_mask : [batch_size x n_heads x len_q x len_k]\n",
|
||||
"\n",
|
||||
" # context: [batch_size x n_heads x len_q x d_v], attn: [batch_size x n_heads x len_q(=len_k) x len_k(=len_q)]\n",
|
||||
" context, attn = ScaledDotProductAttention()(q_s, k_s, v_s, attn_mask)\n",
|
||||
" context = context.transpose(1, 2).contiguous().view(batch_size, -1, n_heads * d_v) # context: [batch_size x len_q x n_heads * d_v]\n",
|
||||
" output = self.linear(context)\n",
|
||||
" return self.layer_norm(output + residual), attn # output: [batch_size x len_q x d_model]\n",
|
||||
"\n",
|
||||
"class PoswiseFeedForwardNet(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(PoswiseFeedForwardNet, self).__init__()\n",
|
||||
" self.conv1 = nn.Conv1d(in_channels=d_model, out_channels=d_ff, kernel_size=1)\n",
|
||||
" self.conv2 = nn.Conv1d(in_channels=d_ff, out_channels=d_model, kernel_size=1)\n",
|
||||
" self.layer_norm = nn.LayerNorm(d_model)\n",
|
||||
"\n",
|
||||
" def forward(self, inputs):\n",
|
||||
" residual = inputs # inputs : [batch_size, len_q, d_model]\n",
|
||||
" output = nn.ReLU()(self.conv1(inputs.transpose(1, 2)))\n",
|
||||
" output = self.conv2(output).transpose(1, 2)\n",
|
||||
" return self.layer_norm(output + residual)\n",
|
||||
"\n",
|
||||
"class EncoderLayer(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(EncoderLayer, self).__init__()\n",
|
||||
" self.enc_self_attn = MultiHeadAttention()\n",
|
||||
" self.pos_ffn = PoswiseFeedForwardNet()\n",
|
||||
"\n",
|
||||
" def forward(self, enc_inputs, enc_self_attn_mask):\n",
|
||||
" enc_outputs, attn = self.enc_self_attn(enc_inputs, enc_inputs, enc_inputs, enc_self_attn_mask) # enc_inputs to same Q,K,V\n",
|
||||
" enc_outputs = self.pos_ffn(enc_outputs) # enc_outputs: [batch_size x len_q x d_model]\n",
|
||||
" return enc_outputs, attn\n",
|
||||
"\n",
|
||||
"class DecoderLayer(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(DecoderLayer, self).__init__()\n",
|
||||
" self.dec_self_attn = MultiHeadAttention()\n",
|
||||
" self.dec_enc_attn = MultiHeadAttention()\n",
|
||||
" self.pos_ffn = PoswiseFeedForwardNet()\n",
|
||||
"\n",
|
||||
" def forward(self, dec_inputs, enc_outputs, dec_self_attn_mask, dec_enc_attn_mask):\n",
|
||||
" dec_outputs, dec_self_attn = self.dec_self_attn(dec_inputs, dec_inputs, dec_inputs, dec_self_attn_mask)\n",
|
||||
" dec_outputs, dec_enc_attn = self.dec_enc_attn(dec_outputs, enc_outputs, enc_outputs, dec_enc_attn_mask)\n",
|
||||
" dec_outputs = self.pos_ffn(dec_outputs)\n",
|
||||
" return dec_outputs, dec_self_attn, dec_enc_attn\n",
|
||||
"\n",
|
||||
"class Encoder(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(Encoder, self).__init__()\n",
|
||||
" self.src_emb = nn.Embedding(src_vocab_size, d_model)\n",
|
||||
" self.pos_emb = nn.Embedding.from_pretrained(get_sinusoid_encoding_table(src_len+1, d_model),freeze=True)\n",
|
||||
" self.layers = nn.ModuleList([EncoderLayer() for _ in range(n_layers)])\n",
|
||||
"\n",
|
||||
" def forward(self, enc_inputs): # enc_inputs : [batch_size x source_len]\n",
|
||||
" enc_outputs = self.src_emb(enc_inputs) + self.pos_emb(torch.LongTensor([[1,2,3,4,0]]))\n",
|
||||
" enc_self_attn_mask = get_attn_pad_mask(enc_inputs, enc_inputs)\n",
|
||||
" enc_self_attns = []\n",
|
||||
" for layer in self.layers:\n",
|
||||
" enc_outputs, enc_self_attn = layer(enc_outputs, enc_self_attn_mask)\n",
|
||||
" enc_self_attns.append(enc_self_attn)\n",
|
||||
" return enc_outputs, enc_self_attns\n",
|
||||
"\n",
|
||||
"class Decoder(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(Decoder, self).__init__()\n",
|
||||
" self.tgt_emb = nn.Embedding(tgt_vocab_size, d_model)\n",
|
||||
" self.pos_emb = nn.Embedding.from_pretrained(get_sinusoid_encoding_table(tgt_len+1, d_model),freeze=True)\n",
|
||||
" self.layers = nn.ModuleList([DecoderLayer() for _ in range(n_layers)])\n",
|
||||
"\n",
|
||||
" def forward(self, dec_inputs, enc_inputs, enc_outputs): # dec_inputs : [batch_size x target_len]\n",
|
||||
" dec_outputs = self.tgt_emb(dec_inputs) + self.pos_emb(torch.LongTensor([[5,1,2,3,4]]))\n",
|
||||
" dec_self_attn_pad_mask = get_attn_pad_mask(dec_inputs, dec_inputs)\n",
|
||||
" dec_self_attn_subsequent_mask = get_attn_subsequent_mask(dec_inputs)\n",
|
||||
" dec_self_attn_mask = torch.gt((dec_self_attn_pad_mask + dec_self_attn_subsequent_mask), 0)\n",
|
||||
"\n",
|
||||
" dec_enc_attn_mask = get_attn_pad_mask(dec_inputs, enc_inputs)\n",
|
||||
"\n",
|
||||
" dec_self_attns, dec_enc_attns = [], []\n",
|
||||
" for layer in self.layers:\n",
|
||||
" dec_outputs, dec_self_attn, dec_enc_attn = layer(dec_outputs, enc_outputs, dec_self_attn_mask, dec_enc_attn_mask)\n",
|
||||
" dec_self_attns.append(dec_self_attn)\n",
|
||||
" dec_enc_attns.append(dec_enc_attn)\n",
|
||||
" return dec_outputs, dec_self_attns, dec_enc_attns\n",
|
||||
"\n",
|
||||
"class Transformer(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(Transformer, self).__init__()\n",
|
||||
" self.encoder = Encoder()\n",
|
||||
" self.decoder = Decoder()\n",
|
||||
" self.projection = nn.Linear(d_model, tgt_vocab_size, bias=False)\n",
|
||||
" def forward(self, enc_inputs, dec_inputs):\n",
|
||||
" enc_outputs, enc_self_attns = self.encoder(enc_inputs)\n",
|
||||
" dec_outputs, dec_self_attns, dec_enc_attns = self.decoder(dec_inputs, enc_inputs, enc_outputs)\n",
|
||||
" dec_logits = self.projection(dec_outputs) # dec_logits : [batch_size x src_vocab_size x tgt_vocab_size]\n",
|
||||
" return dec_logits.view(-1, dec_logits.size(-1)), enc_self_attns, dec_self_attns, dec_enc_attns\n",
|
||||
"\n",
|
||||
"def greedy_decoder(model, enc_input, start_symbol):\n",
|
||||
" \"\"\"\n",
|
||||
" For simplicity, a Greedy Decoder is Beam search when K=1. This is necessary for inference as we don't know the\n",
|
||||
" target sequence input. Therefore we try to generate the target input word by word, then feed it into the transformer.\n",
|
||||
" Starting Reference: http://nlp.seas.harvard.edu/2018/04/03/attention.html#greedy-decoding\n",
|
||||
" :param model: Transformer Model\n",
|
||||
" :param enc_input: The encoder input\n",
|
||||
" :param start_symbol: The start symbol. In this example it is 'S' which corresponds to index 4\n",
|
||||
" :return: The target input\n",
|
||||
" \"\"\"\n",
|
||||
" enc_outputs, enc_self_attns = model.encoder(enc_input)\n",
|
||||
" dec_input = torch.zeros(1, 5).type_as(enc_input.data)\n",
|
||||
" next_symbol = start_symbol\n",
|
||||
" for i in range(0, 5):\n",
|
||||
" dec_input[0][i] = next_symbol\n",
|
||||
" dec_outputs, _, _ = model.decoder(dec_input, enc_input, enc_outputs)\n",
|
||||
" projected = model.projection(dec_outputs)\n",
|
||||
" prob = projected.squeeze(0).max(dim=-1, keepdim=False)[1]\n",
|
||||
" next_word = prob.data[i]\n",
|
||||
" next_symbol = next_word.item()\n",
|
||||
" return dec_input\n",
|
||||
"\n",
|
||||
"def showgraph(attn):\n",
|
||||
" attn = attn[-1].squeeze(0)[0]\n",
|
||||
" attn = attn.squeeze(0).data.numpy()\n",
|
||||
" fig = plt.figure(figsize=(n_heads, n_heads)) # [n_heads, n_heads]\n",
|
||||
" ax = fig.add_subplot(1, 1, 1)\n",
|
||||
" ax.matshow(attn, cmap='viridis')\n",
|
||||
" ax.set_xticklabels(['']+sentences[0].split(), fontdict={'fontsize': 14}, rotation=90)\n",
|
||||
" ax.set_yticklabels(['']+sentences[2].split(), fontdict={'fontsize': 14})\n",
|
||||
" plt.show()\n",
|
||||
"\n",
|
||||
"if __name__ == '__main__':\n",
|
||||
" sentences = ['ich mochte ein bier P', 'S i want a beer', 'i want a beer E']\n",
|
||||
" # Transformer Parameters\n",
|
||||
" # Padding Should be Zero index\n",
|
||||
" src_vocab = {'P': 0, 'ich': 1, 'mochte': 2, 'ein': 3, 'bier': 4}\n",
|
||||
" src_vocab_size = len(src_vocab)\n",
|
||||
"\n",
|
||||
" tgt_vocab = {'P': 0, 'i': 1, 'want': 2, 'a': 3, 'beer': 4, 'S': 5, 'E': 6}\n",
|
||||
" number_dict = {i: w for i, w in enumerate(tgt_vocab)}\n",
|
||||
" tgt_vocab_size = len(tgt_vocab)\n",
|
||||
"\n",
|
||||
" src_len = 5 # length of source\n",
|
||||
" tgt_len = 5 # length of target\n",
|
||||
"\n",
|
||||
" d_model = 512 # Embedding Size\n",
|
||||
" d_ff = 2048 # FeedForward dimension\n",
|
||||
" d_k = d_v = 64 # dimension of K(=Q), V\n",
|
||||
" n_layers = 6 # number of Encoder of Decoder Layer\n",
|
||||
" n_heads = 8 # number of heads in Multi-Head Attention\n",
|
||||
"\n",
|
||||
" model = Transformer()\n",
|
||||
"\n",
|
||||
" criterion = nn.CrossEntropyLoss()\n",
|
||||
" optimizer = optim.Adam(model.parameters(), lr=0.001)\n",
|
||||
"\n",
|
||||
" enc_inputs, dec_inputs, target_batch = make_batch()\n",
|
||||
"\n",
|
||||
" for epoch in range(20):\n",
|
||||
" optimizer.zero_grad()\n",
|
||||
" outputs, enc_self_attns, dec_self_attns, dec_enc_attns = model(enc_inputs, dec_inputs)\n",
|
||||
" loss = criterion(outputs, target_batch.contiguous().view(-1))\n",
|
||||
" print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.6f}'.format(loss))\n",
|
||||
" loss.backward()\n",
|
||||
" optimizer.step()\n",
|
||||
"\n",
|
||||
" # Test\n",
|
||||
" greedy_dec_input = greedy_decoder(model, enc_inputs, start_symbol=tgt_vocab[\"S\"])\n",
|
||||
" predict, _, _, _ = model(enc_inputs, greedy_dec_input)\n",
|
||||
" predict = predict.data.max(1, keepdim=True)[1]\n",
|
||||
" print(sentences[0], '->', [number_dict[n.item()] for n in predict.squeeze()])\n",
|
||||
"\n",
|
||||
" print('first head of last state enc_self_attns')\n",
|
||||
" showgraph(enc_self_attns)\n",
|
||||
"\n",
|
||||
" print('first head of last state dec_self_attns')\n",
|
||||
" showgraph(dec_self_attns)\n",
|
||||
"\n",
|
||||
" print('first head of last state dec_enc_attns')\n",
|
||||
" showgraph(dec_enc_attns)"
|
||||
],
|
||||
"outputs": [],
|
||||
"execution_count": null
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
"anaconda-cloud": {},
|
||||
"kernelspec": {
|
||||
"display_name": "Python 3",
|
||||
"language": "python",
|
||||
"name": "python3"
|
||||
},
|
||||
"language_info": {
|
||||
"codemirror_mode": {
|
||||
"name": "ipython",
|
||||
"version": 3
|
||||
},
|
||||
"file_extension": ".py",
|
||||
"mimetype": "text/x-python",
|
||||
"name": "python",
|
||||
"nbconvert_exporter": "python",
|
||||
"pygments_lexer": "ipython3",
|
||||
"version": "3.6.1"
|
||||
}
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 4
|
||||
}
|
||||
@@ -0,0 +1,249 @@
|
||||
# %%
|
||||
# code by Tae Hwan Jung(Jeff Jung) @graykode, Derek Miller @dmmiller612
|
||||
# Reference : https://github.com/jadore801120/attention-is-all-you-need-pytorch
|
||||
# https://github.com/JayParks/transformer
|
||||
import numpy as np
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
import torch.optim as optim
|
||||
import matplotlib.pyplot as plt
|
||||
|
||||
# S: Symbol that shows starting of decoding input
|
||||
# E: Symbol that shows starting of decoding output
|
||||
# P: Symbol that will fill in blank sequence if current batch data size is short than time steps
|
||||
|
||||
def make_batch():
|
||||
input_batch = [[src_vocab[n] for n in sentences[0].split()]]
|
||||
output_batch = [[tgt_vocab[n] for n in sentences[1].split()]]
|
||||
target_batch = [[tgt_vocab[n] for n in sentences[2].split()]]
|
||||
return torch.LongTensor(input_batch), torch.LongTensor(output_batch), torch.LongTensor(target_batch)
|
||||
|
||||
def get_sinusoid_encoding_table(n_position, d_model):
|
||||
def cal_angle(position, hid_idx):
|
||||
return position / np.power(10000, 2 * (hid_idx // 2) / d_model)
|
||||
def get_posi_angle_vec(position):
|
||||
return [cal_angle(position, hid_j) for hid_j in range(d_model)]
|
||||
|
||||
sinusoid_table = np.array([get_posi_angle_vec(pos_i) for pos_i in range(n_position)])
|
||||
sinusoid_table[:, 0::2] = np.sin(sinusoid_table[:, 0::2]) # dim 2i
|
||||
sinusoid_table[:, 1::2] = np.cos(sinusoid_table[:, 1::2]) # dim 2i+1
|
||||
return torch.FloatTensor(sinusoid_table)
|
||||
|
||||
def get_attn_pad_mask(seq_q, seq_k):
|
||||
# print(seq_q)
|
||||
batch_size, len_q = seq_q.size()
|
||||
batch_size, len_k = seq_k.size()
|
||||
# eq(zero) is PAD token
|
||||
pad_attn_mask = seq_k.data.eq(0).unsqueeze(1) # batch_size x 1 x len_k(=len_q), one is masking
|
||||
return pad_attn_mask.expand(batch_size, len_q, len_k) # batch_size x len_q x len_k
|
||||
|
||||
def get_attn_subsequent_mask(seq):
|
||||
attn_shape = [seq.size(0), seq.size(1), seq.size(1)]
|
||||
subsequent_mask = np.triu(np.ones(attn_shape), k=1)
|
||||
subsequent_mask = torch.from_numpy(subsequent_mask).byte()
|
||||
return subsequent_mask
|
||||
|
||||
class ScaledDotProductAttention(nn.Module):
|
||||
def __init__(self):
|
||||
super(ScaledDotProductAttention, self).__init__()
|
||||
|
||||
def forward(self, Q, K, V, attn_mask):
|
||||
scores = torch.matmul(Q, K.transpose(-1, -2)) / np.sqrt(d_k) # scores : [batch_size x n_heads x len_q(=len_k) x len_k(=len_q)]
|
||||
scores.masked_fill_(attn_mask, -1e9) # Fills elements of self tensor with value where mask is one.
|
||||
attn = nn.Softmax(dim=-1)(scores)
|
||||
context = torch.matmul(attn, V)
|
||||
return context, attn
|
||||
|
||||
class MultiHeadAttention(nn.Module):
|
||||
def __init__(self):
|
||||
super(MultiHeadAttention, self).__init__()
|
||||
self.W_Q = nn.Linear(d_model, d_k * n_heads)
|
||||
self.W_K = nn.Linear(d_model, d_k * n_heads)
|
||||
self.W_V = nn.Linear(d_model, d_v * n_heads)
|
||||
self.linear = nn.Linear(n_heads * d_v, d_model)
|
||||
self.layer_norm = nn.LayerNorm(d_model)
|
||||
|
||||
def forward(self, Q, K, V, attn_mask):
|
||||
# q: [batch_size x len_q x d_model], k: [batch_size x len_k x d_model], v: [batch_size x len_k x d_model]
|
||||
residual, batch_size = Q, Q.size(0)
|
||||
# (B, S, D) -proj-> (B, S, D) -split-> (B, S, H, W) -trans-> (B, H, S, W)
|
||||
q_s = self.W_Q(Q).view(batch_size, -1, n_heads, d_k).transpose(1,2) # q_s: [batch_size x n_heads x len_q x d_k]
|
||||
k_s = self.W_K(K).view(batch_size, -1, n_heads, d_k).transpose(1,2) # k_s: [batch_size x n_heads x len_k x d_k]
|
||||
v_s = self.W_V(V).view(batch_size, -1, n_heads, d_v).transpose(1,2) # v_s: [batch_size x n_heads x len_k x d_v]
|
||||
|
||||
attn_mask = attn_mask.unsqueeze(1).repeat(1, n_heads, 1, 1) # attn_mask : [batch_size x n_heads x len_q x len_k]
|
||||
|
||||
# context: [batch_size x n_heads x len_q x d_v], attn: [batch_size x n_heads x len_q(=len_k) x len_k(=len_q)]
|
||||
context, attn = ScaledDotProductAttention()(q_s, k_s, v_s, attn_mask)
|
||||
context = context.transpose(1, 2).contiguous().view(batch_size, -1, n_heads * d_v) # context: [batch_size x len_q x n_heads * d_v]
|
||||
output = self.linear(context)
|
||||
return self.layer_norm(output + residual), attn # output: [batch_size x len_q x d_model]
|
||||
|
||||
class PoswiseFeedForwardNet(nn.Module):
|
||||
def __init__(self):
|
||||
super(PoswiseFeedForwardNet, self).__init__()
|
||||
self.conv1 = nn.Conv1d(in_channels=d_model, out_channels=d_ff, kernel_size=1)
|
||||
self.conv2 = nn.Conv1d(in_channels=d_ff, out_channels=d_model, kernel_size=1)
|
||||
self.layer_norm = nn.LayerNorm(d_model)
|
||||
|
||||
def forward(self, inputs):
|
||||
residual = inputs # inputs : [batch_size, len_q, d_model]
|
||||
output = nn.ReLU()(self.conv1(inputs.transpose(1, 2)))
|
||||
output = self.conv2(output).transpose(1, 2)
|
||||
return self.layer_norm(output + residual)
|
||||
|
||||
class EncoderLayer(nn.Module):
|
||||
def __init__(self):
|
||||
super(EncoderLayer, self).__init__()
|
||||
self.enc_self_attn = MultiHeadAttention()
|
||||
self.pos_ffn = PoswiseFeedForwardNet()
|
||||
|
||||
def forward(self, enc_inputs, enc_self_attn_mask):
|
||||
enc_outputs, attn = self.enc_self_attn(enc_inputs, enc_inputs, enc_inputs, enc_self_attn_mask) # enc_inputs to same Q,K,V
|
||||
enc_outputs = self.pos_ffn(enc_outputs) # enc_outputs: [batch_size x len_q x d_model]
|
||||
return enc_outputs, attn
|
||||
|
||||
class DecoderLayer(nn.Module):
|
||||
def __init__(self):
|
||||
super(DecoderLayer, self).__init__()
|
||||
self.dec_self_attn = MultiHeadAttention()
|
||||
self.dec_enc_attn = MultiHeadAttention()
|
||||
self.pos_ffn = PoswiseFeedForwardNet()
|
||||
|
||||
def forward(self, dec_inputs, enc_outputs, dec_self_attn_mask, dec_enc_attn_mask):
|
||||
dec_outputs, dec_self_attn = self.dec_self_attn(dec_inputs, dec_inputs, dec_inputs, dec_self_attn_mask)
|
||||
dec_outputs, dec_enc_attn = self.dec_enc_attn(dec_outputs, enc_outputs, enc_outputs, dec_enc_attn_mask)
|
||||
dec_outputs = self.pos_ffn(dec_outputs)
|
||||
return dec_outputs, dec_self_attn, dec_enc_attn
|
||||
|
||||
class Encoder(nn.Module):
|
||||
def __init__(self):
|
||||
super(Encoder, self).__init__()
|
||||
self.src_emb = nn.Embedding(src_vocab_size, d_model)
|
||||
self.pos_emb = nn.Embedding.from_pretrained(get_sinusoid_encoding_table(src_len+1, d_model),freeze=True)
|
||||
self.layers = nn.ModuleList([EncoderLayer() for _ in range(n_layers)])
|
||||
|
||||
def forward(self, enc_inputs): # enc_inputs : [batch_size x source_len]
|
||||
enc_outputs = self.src_emb(enc_inputs) + self.pos_emb(torch.LongTensor([[1,2,3,4,0]]))
|
||||
enc_self_attn_mask = get_attn_pad_mask(enc_inputs, enc_inputs)
|
||||
enc_self_attns = []
|
||||
for layer in self.layers:
|
||||
enc_outputs, enc_self_attn = layer(enc_outputs, enc_self_attn_mask)
|
||||
enc_self_attns.append(enc_self_attn)
|
||||
return enc_outputs, enc_self_attns
|
||||
|
||||
class Decoder(nn.Module):
|
||||
def __init__(self):
|
||||
super(Decoder, self).__init__()
|
||||
self.tgt_emb = nn.Embedding(tgt_vocab_size, d_model)
|
||||
self.pos_emb = nn.Embedding.from_pretrained(get_sinusoid_encoding_table(tgt_len+1, d_model),freeze=True)
|
||||
self.layers = nn.ModuleList([DecoderLayer() for _ in range(n_layers)])
|
||||
|
||||
def forward(self, dec_inputs, enc_inputs, enc_outputs): # dec_inputs : [batch_size x target_len]
|
||||
dec_outputs = self.tgt_emb(dec_inputs) + self.pos_emb(torch.LongTensor([[5,1,2,3,4]]))
|
||||
dec_self_attn_pad_mask = get_attn_pad_mask(dec_inputs, dec_inputs)
|
||||
dec_self_attn_subsequent_mask = get_attn_subsequent_mask(dec_inputs)
|
||||
dec_self_attn_mask = torch.gt((dec_self_attn_pad_mask + dec_self_attn_subsequent_mask), 0)
|
||||
|
||||
dec_enc_attn_mask = get_attn_pad_mask(dec_inputs, enc_inputs)
|
||||
|
||||
dec_self_attns, dec_enc_attns = [], []
|
||||
for layer in self.layers:
|
||||
dec_outputs, dec_self_attn, dec_enc_attn = layer(dec_outputs, enc_outputs, dec_self_attn_mask, dec_enc_attn_mask)
|
||||
dec_self_attns.append(dec_self_attn)
|
||||
dec_enc_attns.append(dec_enc_attn)
|
||||
return dec_outputs, dec_self_attns, dec_enc_attns
|
||||
|
||||
class Transformer(nn.Module):
|
||||
def __init__(self):
|
||||
super(Transformer, self).__init__()
|
||||
self.encoder = Encoder()
|
||||
self.decoder = Decoder()
|
||||
self.projection = nn.Linear(d_model, tgt_vocab_size, bias=False)
|
||||
def forward(self, enc_inputs, dec_inputs):
|
||||
enc_outputs, enc_self_attns = self.encoder(enc_inputs)
|
||||
dec_outputs, dec_self_attns, dec_enc_attns = self.decoder(dec_inputs, enc_inputs, enc_outputs)
|
||||
dec_logits = self.projection(dec_outputs) # dec_logits : [batch_size x src_vocab_size x tgt_vocab_size]
|
||||
return dec_logits.view(-1, dec_logits.size(-1)), enc_self_attns, dec_self_attns, dec_enc_attns
|
||||
|
||||
def greedy_decoder(model, enc_input, start_symbol):
|
||||
"""
|
||||
For simplicity, a Greedy Decoder is Beam search when K=1. This is necessary for inference as we don't know the
|
||||
target sequence input. Therefore we try to generate the target input word by word, then feed it into the transformer.
|
||||
Starting Reference: http://nlp.seas.harvard.edu/2018/04/03/attention.html#greedy-decoding
|
||||
:param model: Transformer Model
|
||||
:param enc_input: The encoder input
|
||||
:param start_symbol: The start symbol. In this example it is 'S' which corresponds to index 4
|
||||
:return: The target input
|
||||
"""
|
||||
enc_outputs, enc_self_attns = model.encoder(enc_input)
|
||||
dec_input = torch.zeros(1, 5).type_as(enc_input.data)
|
||||
next_symbol = start_symbol
|
||||
for i in range(0, 5):
|
||||
dec_input[0][i] = next_symbol
|
||||
dec_outputs, _, _ = model.decoder(dec_input, enc_input, enc_outputs)
|
||||
projected = model.projection(dec_outputs)
|
||||
prob = projected.squeeze(0).max(dim=-1, keepdim=False)[1]
|
||||
next_word = prob.data[i]
|
||||
next_symbol = next_word.item()
|
||||
return dec_input
|
||||
|
||||
def showgraph(attn):
|
||||
attn = attn[-1].squeeze(0)[0]
|
||||
attn = attn.squeeze(0).data.numpy()
|
||||
fig = plt.figure(figsize=(n_heads, n_heads)) # [n_heads, n_heads]
|
||||
ax = fig.add_subplot(1, 1, 1)
|
||||
ax.matshow(attn, cmap='viridis')
|
||||
ax.set_xticklabels(['']+sentences[0].split(), fontdict={'fontsize': 14}, rotation=90)
|
||||
ax.set_yticklabels(['']+sentences[2].split(), fontdict={'fontsize': 14})
|
||||
plt.show()
|
||||
|
||||
if __name__ == '__main__':
|
||||
sentences = ['ich mochte ein bier P', 'S i want a beer', 'i want a beer E']
|
||||
# Transformer Parameters
|
||||
# Padding Should be Zero index
|
||||
src_vocab = {'P': 0, 'ich': 1, 'mochte': 2, 'ein': 3, 'bier': 4}
|
||||
src_vocab_size = len(src_vocab)
|
||||
|
||||
tgt_vocab = {'P': 0, 'i': 1, 'want': 2, 'a': 3, 'beer': 4, 'S': 5, 'E': 6}
|
||||
number_dict = {i: w for i, w in enumerate(tgt_vocab)}
|
||||
tgt_vocab_size = len(tgt_vocab)
|
||||
|
||||
src_len = 5 # length of source
|
||||
tgt_len = 5 # length of target
|
||||
|
||||
d_model = 512 # Embedding Size
|
||||
d_ff = 2048 # FeedForward dimension
|
||||
d_k = d_v = 64 # dimension of K(=Q), V
|
||||
n_layers = 6 # number of Encoder of Decoder Layer
|
||||
n_heads = 8 # number of heads in Multi-Head Attention
|
||||
|
||||
model = Transformer()
|
||||
|
||||
criterion = nn.CrossEntropyLoss()
|
||||
optimizer = optim.Adam(model.parameters(), lr=0.001)
|
||||
|
||||
enc_inputs, dec_inputs, target_batch = make_batch()
|
||||
|
||||
for epoch in range(20):
|
||||
optimizer.zero_grad()
|
||||
outputs, enc_self_attns, dec_self_attns, dec_enc_attns = model(enc_inputs, dec_inputs)
|
||||
loss = criterion(outputs, target_batch.contiguous().view(-1))
|
||||
print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.6f}'.format(loss))
|
||||
loss.backward()
|
||||
optimizer.step()
|
||||
|
||||
# Test
|
||||
greedy_dec_input = greedy_decoder(model, enc_inputs, start_symbol=tgt_vocab["S"])
|
||||
predict, _, _, _ = model(enc_inputs, greedy_dec_input)
|
||||
predict = predict.data.max(1, keepdim=True)[1]
|
||||
print(sentences[0], '->', [number_dict[n.item()] for n in predict.squeeze()])
|
||||
|
||||
print('first head of last state enc_self_attns')
|
||||
showgraph(enc_self_attns)
|
||||
|
||||
print('first head of last state dec_self_attns')
|
||||
showgraph(dec_self_attns)
|
||||
|
||||
print('first head of last state dec_enc_attns')
|
||||
showgraph(dec_enc_attns)
|
||||
@@ -0,0 +1,259 @@
|
||||
{
|
||||
"cells": [
|
||||
{
|
||||
"cell_type": "code",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"# code by Tae Hwan Jung(Jeff Jung) @graykode, Derek Miller @dmmiller612\n",
|
||||
"# Reference : https://github.com/jadore801120/attention-is-all-you-need-pytorch\n",
|
||||
"# https://github.com/JayParks/transformer\n",
|
||||
"import numpy as np\n",
|
||||
"import torch\n",
|
||||
"import torch.nn as nn\n",
|
||||
"import torch.optim as optim\n",
|
||||
"import matplotlib.pyplot as plt\n",
|
||||
"\n",
|
||||
"# S: Symbol that shows starting of decoding input\n",
|
||||
"# E: Symbol that shows starting of decoding output\n",
|
||||
"# P: Symbol that will fill in blank sequence if current batch data size is short than time steps\n",
|
||||
"\n",
|
||||
"def make_batch(sentences):\n",
|
||||
" input_batch = [[src_vocab[n] for n in sentences[0].split()]]\n",
|
||||
" output_batch = [[tgt_vocab[n] for n in sentences[1].split()]]\n",
|
||||
" target_batch = [[tgt_vocab[n] for n in sentences[2].split()]]\n",
|
||||
" return torch.LongTensor(input_batch), torch.LongTensor(output_batch), torch.LongTensor(target_batch)\n",
|
||||
"\n",
|
||||
"def get_sinusoid_encoding_table(n_position, d_model):\n",
|
||||
" def cal_angle(position, hid_idx):\n",
|
||||
" return position / np.power(10000, 2 * (hid_idx // 2) / d_model)\n",
|
||||
" def get_posi_angle_vec(position):\n",
|
||||
" return [cal_angle(position, hid_j) for hid_j in range(d_model)]\n",
|
||||
"\n",
|
||||
" sinusoid_table = np.array([get_posi_angle_vec(pos_i) for pos_i in range(n_position)])\n",
|
||||
" sinusoid_table[:, 0::2] = np.sin(sinusoid_table[:, 0::2]) # dim 2i\n",
|
||||
" sinusoid_table[:, 1::2] = np.cos(sinusoid_table[:, 1::2]) # dim 2i+1\n",
|
||||
" return torch.FloatTensor(sinusoid_table)\n",
|
||||
"\n",
|
||||
"def get_attn_pad_mask(seq_q, seq_k):\n",
|
||||
" batch_size, len_q = seq_q.size()\n",
|
||||
" batch_size, len_k = seq_k.size()\n",
|
||||
" # eq(zero) is PAD token\n",
|
||||
" pad_attn_mask = seq_k.data.eq(0).unsqueeze(1) # batch_size x 1 x len_k(=len_q), one is masking\n",
|
||||
" return pad_attn_mask.expand(batch_size, len_q, len_k) # batch_size x len_q x len_k\n",
|
||||
"\n",
|
||||
"def get_attn_subsequent_mask(seq):\n",
|
||||
" attn_shape = [seq.size(0), seq.size(1), seq.size(1)]\n",
|
||||
" subsequent_mask = np.triu(np.ones(attn_shape), k=1)\n",
|
||||
" subsequent_mask = torch.from_numpy(subsequent_mask).byte()\n",
|
||||
" return subsequent_mask\n",
|
||||
"\n",
|
||||
"class ScaledDotProductAttention(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(ScaledDotProductAttention, self).__init__()\n",
|
||||
"\n",
|
||||
" def forward(self, Q, K, V, attn_mask):\n",
|
||||
" scores = torch.matmul(Q, K.transpose(-1, -2)) / np.sqrt(d_k) # scores : [batch_size x n_heads x len_q(=len_k) x len_k(=len_q)]\n",
|
||||
" scores.masked_fill_(attn_mask, -1e9) # Fills elements of self tensor with value where mask is one.\n",
|
||||
" attn = nn.Softmax(dim=-1)(scores)\n",
|
||||
" context = torch.matmul(attn, V)\n",
|
||||
" return context, attn\n",
|
||||
"\n",
|
||||
"class MultiHeadAttention(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(MultiHeadAttention, self).__init__()\n",
|
||||
" self.W_Q = nn.Linear(d_model, d_k * n_heads)\n",
|
||||
" self.W_K = nn.Linear(d_model, d_k * n_heads)\n",
|
||||
" self.W_V = nn.Linear(d_model, d_v * n_heads)\n",
|
||||
" self.linear = nn.Linear(n_heads * d_v, d_model)\n",
|
||||
" self.layer_norm = nn.LayerNorm(d_model)\n",
|
||||
"\n",
|
||||
" def forward(self, Q, K, V, attn_mask):\n",
|
||||
" # q: [batch_size x len_q x d_model], k: [batch_size x len_k x d_model], v: [batch_size x len_k x d_model]\n",
|
||||
" residual, batch_size = Q, Q.size(0)\n",
|
||||
" # (B, S, D) -proj-> (B, S, D) -split-> (B, S, H, W) -trans-> (B, H, S, W)\n",
|
||||
" q_s = self.W_Q(Q).view(batch_size, -1, n_heads, d_k).transpose(1,2) # q_s: [batch_size x n_heads x len_q x d_k]\n",
|
||||
" k_s = self.W_K(K).view(batch_size, -1, n_heads, d_k).transpose(1,2) # k_s: [batch_size x n_heads x len_k x d_k]\n",
|
||||
" v_s = self.W_V(V).view(batch_size, -1, n_heads, d_v).transpose(1,2) # v_s: [batch_size x n_heads x len_k x d_v]\n",
|
||||
"\n",
|
||||
" attn_mask = attn_mask.unsqueeze(1).repeat(1, n_heads, 1, 1) # attn_mask : [batch_size x n_heads x len_q x len_k]\n",
|
||||
"\n",
|
||||
" # context: [batch_size x n_heads x len_q x d_v], attn: [batch_size x n_heads x len_q(=len_k) x len_k(=len_q)]\n",
|
||||
" context, attn = ScaledDotProductAttention()(q_s, k_s, v_s, attn_mask)\n",
|
||||
" context = context.transpose(1, 2).contiguous().view(batch_size, -1, n_heads * d_v) # context: [batch_size x len_q x n_heads * d_v]\n",
|
||||
" output = self.linear(context)\n",
|
||||
" return self.layer_norm(output + residual), attn # output: [batch_size x len_q x d_model]\n",
|
||||
"\n",
|
||||
"class PoswiseFeedForwardNet(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(PoswiseFeedForwardNet, self).__init__()\n",
|
||||
" self.conv1 = nn.Conv1d(in_channels=d_model, out_channels=d_ff, kernel_size=1)\n",
|
||||
" self.conv2 = nn.Conv1d(in_channels=d_ff, out_channels=d_model, kernel_size=1)\n",
|
||||
" self.layer_norm = nn.LayerNorm(d_model)\n",
|
||||
"\n",
|
||||
" def forward(self, inputs):\n",
|
||||
" residual = inputs # inputs : [batch_size, len_q, d_model]\n",
|
||||
" output = nn.ReLU()(self.conv1(inputs.transpose(1, 2)))\n",
|
||||
" output = self.conv2(output).transpose(1, 2)\n",
|
||||
" return self.layer_norm(output + residual)\n",
|
||||
"\n",
|
||||
"class EncoderLayer(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(EncoderLayer, self).__init__()\n",
|
||||
" self.enc_self_attn = MultiHeadAttention()\n",
|
||||
" self.pos_ffn = PoswiseFeedForwardNet()\n",
|
||||
"\n",
|
||||
" def forward(self, enc_inputs, enc_self_attn_mask):\n",
|
||||
" enc_outputs, attn = self.enc_self_attn(enc_inputs, enc_inputs, enc_inputs, enc_self_attn_mask) # enc_inputs to same Q,K,V\n",
|
||||
" enc_outputs = self.pos_ffn(enc_outputs) # enc_outputs: [batch_size x len_q x d_model]\n",
|
||||
" return enc_outputs, attn\n",
|
||||
"\n",
|
||||
"class DecoderLayer(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(DecoderLayer, self).__init__()\n",
|
||||
" self.dec_self_attn = MultiHeadAttention()\n",
|
||||
" self.dec_enc_attn = MultiHeadAttention()\n",
|
||||
" self.pos_ffn = PoswiseFeedForwardNet()\n",
|
||||
"\n",
|
||||
" def forward(self, dec_inputs, enc_outputs, dec_self_attn_mask, dec_enc_attn_mask):\n",
|
||||
" dec_outputs, dec_self_attn = self.dec_self_attn(dec_inputs, dec_inputs, dec_inputs, dec_self_attn_mask)\n",
|
||||
" dec_outputs, dec_enc_attn = self.dec_enc_attn(dec_outputs, enc_outputs, enc_outputs, dec_enc_attn_mask)\n",
|
||||
" dec_outputs = self.pos_ffn(dec_outputs)\n",
|
||||
" return dec_outputs, dec_self_attn, dec_enc_attn\n",
|
||||
"\n",
|
||||
"class Encoder(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(Encoder, self).__init__()\n",
|
||||
" self.src_emb = nn.Embedding(src_vocab_size, d_model)\n",
|
||||
" self.pos_emb = nn.Embedding.from_pretrained(get_sinusoid_encoding_table(src_len+1, d_model),freeze=True)\n",
|
||||
" self.layers = nn.ModuleList([EncoderLayer() for _ in range(n_layers)])\n",
|
||||
"\n",
|
||||
" def forward(self, enc_inputs): # enc_inputs : [batch_size x source_len]\n",
|
||||
" enc_outputs = self.src_emb(enc_inputs) + self.pos_emb(torch.LongTensor([[1,2,3,4,0]]))\n",
|
||||
" enc_self_attn_mask = get_attn_pad_mask(enc_inputs, enc_inputs)\n",
|
||||
" enc_self_attns = []\n",
|
||||
" for layer in self.layers:\n",
|
||||
" enc_outputs, enc_self_attn = layer(enc_outputs, enc_self_attn_mask)\n",
|
||||
" enc_self_attns.append(enc_self_attn)\n",
|
||||
" return enc_outputs, enc_self_attns\n",
|
||||
"\n",
|
||||
"class Decoder(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(Decoder, self).__init__()\n",
|
||||
" self.tgt_emb = nn.Embedding(tgt_vocab_size, d_model)\n",
|
||||
" self.pos_emb = nn.Embedding.from_pretrained(get_sinusoid_encoding_table(tgt_len+1, d_model),freeze=True)\n",
|
||||
" self.layers = nn.ModuleList([DecoderLayer() for _ in range(n_layers)])\n",
|
||||
"\n",
|
||||
" def forward(self, dec_inputs, enc_inputs, enc_outputs): # dec_inputs : [batch_size x target_len]\n",
|
||||
" dec_outputs = self.tgt_emb(dec_inputs) + self.pos_emb(torch.LongTensor([[5,1,2,3,4]]))\n",
|
||||
" dec_self_attn_pad_mask = get_attn_pad_mask(dec_inputs, dec_inputs)\n",
|
||||
" dec_self_attn_subsequent_mask = get_attn_subsequent_mask(dec_inputs)\n",
|
||||
" dec_self_attn_mask = torch.gt((dec_self_attn_pad_mask + dec_self_attn_subsequent_mask), 0)\n",
|
||||
"\n",
|
||||
" dec_enc_attn_mask = get_attn_pad_mask(dec_inputs, enc_inputs)\n",
|
||||
"\n",
|
||||
" dec_self_attns, dec_enc_attns = [], []\n",
|
||||
" for layer in self.layers:\n",
|
||||
" dec_outputs, dec_self_attn, dec_enc_attn = layer(dec_outputs, enc_outputs, dec_self_attn_mask, dec_enc_attn_mask)\n",
|
||||
" dec_self_attns.append(dec_self_attn)\n",
|
||||
" dec_enc_attns.append(dec_enc_attn)\n",
|
||||
" return dec_outputs, dec_self_attns, dec_enc_attns\n",
|
||||
"\n",
|
||||
"class Transformer(nn.Module):\n",
|
||||
" def __init__(self):\n",
|
||||
" super(Transformer, self).__init__()\n",
|
||||
" self.encoder = Encoder()\n",
|
||||
" self.decoder = Decoder()\n",
|
||||
" self.projection = nn.Linear(d_model, tgt_vocab_size, bias=False)\n",
|
||||
" def forward(self, enc_inputs, dec_inputs):\n",
|
||||
" enc_outputs, enc_self_attns = self.encoder(enc_inputs)\n",
|
||||
" dec_outputs, dec_self_attns, dec_enc_attns = self.decoder(dec_inputs, enc_inputs, enc_outputs)\n",
|
||||
" dec_logits = self.projection(dec_outputs) # dec_logits : [batch_size x src_vocab_size x tgt_vocab_size]\n",
|
||||
" return dec_logits.view(-1, dec_logits.size(-1)), enc_self_attns, dec_self_attns, dec_enc_attns\n",
|
||||
"\n",
|
||||
"def showgraph(attn):\n",
|
||||
" attn = attn[-1].squeeze(0)[0]\n",
|
||||
" attn = attn.squeeze(0).data.numpy()\n",
|
||||
" fig = plt.figure(figsize=(n_heads, n_heads)) # [n_heads, n_heads]\n",
|
||||
" ax = fig.add_subplot(1, 1, 1)\n",
|
||||
" ax.matshow(attn, cmap='viridis')\n",
|
||||
" ax.set_xticklabels(['']+sentences[0].split(), fontdict={'fontsize': 14}, rotation=90)\n",
|
||||
" ax.set_yticklabels(['']+sentences[2].split(), fontdict={'fontsize': 14})\n",
|
||||
" plt.show()\n",
|
||||
"\n",
|
||||
"if __name__ == '__main__':\n",
|
||||
" sentences = ['ich mochte ein bier P', 'S i want a beer', 'i want a beer E']\n",
|
||||
"\n",
|
||||
" # Transformer Parameters\n",
|
||||
" # Padding Should be Zero\n",
|
||||
" src_vocab = {'P': 0, 'ich': 1, 'mochte': 2, 'ein': 3, 'bier': 4}\n",
|
||||
" src_vocab_size = len(src_vocab)\n",
|
||||
"\n",
|
||||
" tgt_vocab = {'P': 0, 'i': 1, 'want': 2, 'a': 3, 'beer': 4, 'S': 5, 'E': 6}\n",
|
||||
" number_dict = {i: w for i, w in enumerate(tgt_vocab)}\n",
|
||||
" tgt_vocab_size = len(tgt_vocab)\n",
|
||||
"\n",
|
||||
" src_len = 5 # length of source\n",
|
||||
" tgt_len = 5 # length of target\n",
|
||||
"\n",
|
||||
" d_model = 512 # Embedding Size\n",
|
||||
" d_ff = 2048 # FeedForward dimension\n",
|
||||
" d_k = d_v = 64 # dimension of K(=Q), V\n",
|
||||
" n_layers = 6 # number of Encoder of Decoder Layer\n",
|
||||
" n_heads = 8 # number of heads in Multi-Head Attention\n",
|
||||
"\n",
|
||||
" model = Transformer()\n",
|
||||
"\n",
|
||||
" criterion = nn.CrossEntropyLoss()\n",
|
||||
" optimizer = optim.Adam(model.parameters(), lr=0.001)\n",
|
||||
"\n",
|
||||
" enc_inputs, dec_inputs, target_batch = make_batch(sentences)\n",
|
||||
"\n",
|
||||
" for epoch in range(20):\n",
|
||||
" optimizer.zero_grad()\n",
|
||||
" outputs, enc_self_attns, dec_self_attns, dec_enc_attns = model(enc_inputs, dec_inputs)\n",
|
||||
" loss = criterion(outputs, target_batch.contiguous().view(-1))\n",
|
||||
" print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.6f}'.format(loss))\n",
|
||||
" loss.backward()\n",
|
||||
" optimizer.step()\n",
|
||||
"\n",
|
||||
" # Test\n",
|
||||
" predict, _, _, _ = model(enc_inputs, dec_inputs)\n",
|
||||
" predict = predict.data.max(1, keepdim=True)[1]\n",
|
||||
" print(sentences[0], '->', [number_dict[n.item()] for n in predict.squeeze()])\n",
|
||||
"\n",
|
||||
" print('first head of last state enc_self_attns')\n",
|
||||
" showgraph(enc_self_attns)\n",
|
||||
"\n",
|
||||
" print('first head of last state dec_self_attns')\n",
|
||||
" showgraph(dec_self_attns)\n",
|
||||
"\n",
|
||||
" print('first head of last state dec_enc_attns')\n",
|
||||
" showgraph(dec_enc_attns)"
|
||||
],
|
||||
"outputs": [],
|
||||
"execution_count": null
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
"anaconda-cloud": {},
|
||||
"kernelspec": {
|
||||
"display_name": "Python 3",
|
||||
"language": "python",
|
||||
"name": "python3"
|
||||
},
|
||||
"language_info": {
|
||||
"codemirror_mode": {
|
||||
"name": "ipython",
|
||||
"version": 3
|
||||
},
|
||||
"file_extension": ".py",
|
||||
"mimetype": "text/x-python",
|
||||
"name": "python",
|
||||
"nbconvert_exporter": "python",
|
||||
"pygments_lexer": "ipython3",
|
||||
"version": "3.6.1"
|
||||
}
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 4
|
||||
}
|
||||
@@ -0,0 +1,226 @@
|
||||
# %%
|
||||
# code by Tae Hwan Jung(Jeff Jung) @graykode, Derek Miller @dmmiller612
|
||||
# Reference : https://github.com/jadore801120/attention-is-all-you-need-pytorch
|
||||
# https://github.com/JayParks/transformer
|
||||
import numpy as np
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
import torch.optim as optim
|
||||
import matplotlib.pyplot as plt
|
||||
|
||||
# S: Symbol that shows starting of decoding input
|
||||
# E: Symbol that shows starting of decoding output
|
||||
# P: Symbol that will fill in blank sequence if current batch data size is short than time steps
|
||||
|
||||
def make_batch(sentences):
|
||||
input_batch = [[src_vocab[n] for n in sentences[0].split()]]
|
||||
output_batch = [[tgt_vocab[n] for n in sentences[1].split()]]
|
||||
target_batch = [[tgt_vocab[n] for n in sentences[2].split()]]
|
||||
return torch.LongTensor(input_batch), torch.LongTensor(output_batch), torch.LongTensor(target_batch)
|
||||
|
||||
def get_sinusoid_encoding_table(n_position, d_model):
|
||||
def cal_angle(position, hid_idx):
|
||||
return position / np.power(10000, 2 * (hid_idx // 2) / d_model)
|
||||
def get_posi_angle_vec(position):
|
||||
return [cal_angle(position, hid_j) for hid_j in range(d_model)]
|
||||
|
||||
sinusoid_table = np.array([get_posi_angle_vec(pos_i) for pos_i in range(n_position)])
|
||||
sinusoid_table[:, 0::2] = np.sin(sinusoid_table[:, 0::2]) # dim 2i
|
||||
sinusoid_table[:, 1::2] = np.cos(sinusoid_table[:, 1::2]) # dim 2i+1
|
||||
return torch.FloatTensor(sinusoid_table)
|
||||
|
||||
def get_attn_pad_mask(seq_q, seq_k):
|
||||
batch_size, len_q = seq_q.size()
|
||||
batch_size, len_k = seq_k.size()
|
||||
# eq(zero) is PAD token
|
||||
pad_attn_mask = seq_k.data.eq(0).unsqueeze(1) # batch_size x 1 x len_k(=len_q), one is masking
|
||||
return pad_attn_mask.expand(batch_size, len_q, len_k) # batch_size x len_q x len_k
|
||||
|
||||
def get_attn_subsequent_mask(seq):
|
||||
attn_shape = [seq.size(0), seq.size(1), seq.size(1)]
|
||||
subsequent_mask = np.triu(np.ones(attn_shape), k=1)
|
||||
subsequent_mask = torch.from_numpy(subsequent_mask).byte()
|
||||
return subsequent_mask
|
||||
|
||||
class ScaledDotProductAttention(nn.Module):
|
||||
def __init__(self):
|
||||
super(ScaledDotProductAttention, self).__init__()
|
||||
|
||||
def forward(self, Q, K, V, attn_mask):
|
||||
scores = torch.matmul(Q, K.transpose(-1, -2)) / np.sqrt(d_k) # scores : [batch_size x n_heads x len_q(=len_k) x len_k(=len_q)]
|
||||
scores.masked_fill_(attn_mask, -1e9) # Fills elements of self tensor with value where mask is one.
|
||||
attn = nn.Softmax(dim=-1)(scores)
|
||||
context = torch.matmul(attn, V)
|
||||
return context, attn
|
||||
|
||||
class MultiHeadAttention(nn.Module):
|
||||
def __init__(self):
|
||||
super(MultiHeadAttention, self).__init__()
|
||||
self.W_Q = nn.Linear(d_model, d_k * n_heads)
|
||||
self.W_K = nn.Linear(d_model, d_k * n_heads)
|
||||
self.W_V = nn.Linear(d_model, d_v * n_heads)
|
||||
self.linear = nn.Linear(n_heads * d_v, d_model)
|
||||
self.layer_norm = nn.LayerNorm(d_model)
|
||||
|
||||
def forward(self, Q, K, V, attn_mask):
|
||||
# q: [batch_size x len_q x d_model], k: [batch_size x len_k x d_model], v: [batch_size x len_k x d_model]
|
||||
residual, batch_size = Q, Q.size(0)
|
||||
# (B, S, D) -proj-> (B, S, D) -split-> (B, S, H, W) -trans-> (B, H, S, W)
|
||||
q_s = self.W_Q(Q).view(batch_size, -1, n_heads, d_k).transpose(1,2) # q_s: [batch_size x n_heads x len_q x d_k]
|
||||
k_s = self.W_K(K).view(batch_size, -1, n_heads, d_k).transpose(1,2) # k_s: [batch_size x n_heads x len_k x d_k]
|
||||
v_s = self.W_V(V).view(batch_size, -1, n_heads, d_v).transpose(1,2) # v_s: [batch_size x n_heads x len_k x d_v]
|
||||
|
||||
attn_mask = attn_mask.unsqueeze(1).repeat(1, n_heads, 1, 1) # attn_mask : [batch_size x n_heads x len_q x len_k]
|
||||
|
||||
# context: [batch_size x n_heads x len_q x d_v], attn: [batch_size x n_heads x len_q(=len_k) x len_k(=len_q)]
|
||||
context, attn = ScaledDotProductAttention()(q_s, k_s, v_s, attn_mask)
|
||||
context = context.transpose(1, 2).contiguous().view(batch_size, -1, n_heads * d_v) # context: [batch_size x len_q x n_heads * d_v]
|
||||
output = self.linear(context)
|
||||
return self.layer_norm(output + residual), attn # output: [batch_size x len_q x d_model]
|
||||
|
||||
class PoswiseFeedForwardNet(nn.Module):
|
||||
def __init__(self):
|
||||
super(PoswiseFeedForwardNet, self).__init__()
|
||||
self.conv1 = nn.Conv1d(in_channels=d_model, out_channels=d_ff, kernel_size=1)
|
||||
self.conv2 = nn.Conv1d(in_channels=d_ff, out_channels=d_model, kernel_size=1)
|
||||
self.layer_norm = nn.LayerNorm(d_model)
|
||||
|
||||
def forward(self, inputs):
|
||||
residual = inputs # inputs : [batch_size, len_q, d_model]
|
||||
output = nn.ReLU()(self.conv1(inputs.transpose(1, 2)))
|
||||
output = self.conv2(output).transpose(1, 2)
|
||||
return self.layer_norm(output + residual)
|
||||
|
||||
class EncoderLayer(nn.Module):
|
||||
def __init__(self):
|
||||
super(EncoderLayer, self).__init__()
|
||||
self.enc_self_attn = MultiHeadAttention()
|
||||
self.pos_ffn = PoswiseFeedForwardNet()
|
||||
|
||||
def forward(self, enc_inputs, enc_self_attn_mask):
|
||||
enc_outputs, attn = self.enc_self_attn(enc_inputs, enc_inputs, enc_inputs, enc_self_attn_mask) # enc_inputs to same Q,K,V
|
||||
enc_outputs = self.pos_ffn(enc_outputs) # enc_outputs: [batch_size x len_q x d_model]
|
||||
return enc_outputs, attn
|
||||
|
||||
class DecoderLayer(nn.Module):
|
||||
def __init__(self):
|
||||
super(DecoderLayer, self).__init__()
|
||||
self.dec_self_attn = MultiHeadAttention()
|
||||
self.dec_enc_attn = MultiHeadAttention()
|
||||
self.pos_ffn = PoswiseFeedForwardNet()
|
||||
|
||||
def forward(self, dec_inputs, enc_outputs, dec_self_attn_mask, dec_enc_attn_mask):
|
||||
dec_outputs, dec_self_attn = self.dec_self_attn(dec_inputs, dec_inputs, dec_inputs, dec_self_attn_mask)
|
||||
dec_outputs, dec_enc_attn = self.dec_enc_attn(dec_outputs, enc_outputs, enc_outputs, dec_enc_attn_mask)
|
||||
dec_outputs = self.pos_ffn(dec_outputs)
|
||||
return dec_outputs, dec_self_attn, dec_enc_attn
|
||||
|
||||
class Encoder(nn.Module):
|
||||
def __init__(self):
|
||||
super(Encoder, self).__init__()
|
||||
self.src_emb = nn.Embedding(src_vocab_size, d_model)
|
||||
self.pos_emb = nn.Embedding.from_pretrained(get_sinusoid_encoding_table(src_len+1, d_model),freeze=True)
|
||||
self.layers = nn.ModuleList([EncoderLayer() for _ in range(n_layers)])
|
||||
|
||||
def forward(self, enc_inputs): # enc_inputs : [batch_size x source_len]
|
||||
enc_outputs = self.src_emb(enc_inputs) + self.pos_emb(torch.LongTensor([[1,2,3,4,0]]))
|
||||
enc_self_attn_mask = get_attn_pad_mask(enc_inputs, enc_inputs)
|
||||
enc_self_attns = []
|
||||
for layer in self.layers:
|
||||
enc_outputs, enc_self_attn = layer(enc_outputs, enc_self_attn_mask)
|
||||
enc_self_attns.append(enc_self_attn)
|
||||
return enc_outputs, enc_self_attns
|
||||
|
||||
class Decoder(nn.Module):
|
||||
def __init__(self):
|
||||
super(Decoder, self).__init__()
|
||||
self.tgt_emb = nn.Embedding(tgt_vocab_size, d_model)
|
||||
self.pos_emb = nn.Embedding.from_pretrained(get_sinusoid_encoding_table(tgt_len+1, d_model),freeze=True)
|
||||
self.layers = nn.ModuleList([DecoderLayer() for _ in range(n_layers)])
|
||||
|
||||
def forward(self, dec_inputs, enc_inputs, enc_outputs): # dec_inputs : [batch_size x target_len]
|
||||
dec_outputs = self.tgt_emb(dec_inputs) + self.pos_emb(torch.LongTensor([[5,1,2,3,4]]))
|
||||
dec_self_attn_pad_mask = get_attn_pad_mask(dec_inputs, dec_inputs)
|
||||
dec_self_attn_subsequent_mask = get_attn_subsequent_mask(dec_inputs)
|
||||
dec_self_attn_mask = torch.gt((dec_self_attn_pad_mask + dec_self_attn_subsequent_mask), 0)
|
||||
|
||||
dec_enc_attn_mask = get_attn_pad_mask(dec_inputs, enc_inputs)
|
||||
|
||||
dec_self_attns, dec_enc_attns = [], []
|
||||
for layer in self.layers:
|
||||
dec_outputs, dec_self_attn, dec_enc_attn = layer(dec_outputs, enc_outputs, dec_self_attn_mask, dec_enc_attn_mask)
|
||||
dec_self_attns.append(dec_self_attn)
|
||||
dec_enc_attns.append(dec_enc_attn)
|
||||
return dec_outputs, dec_self_attns, dec_enc_attns
|
||||
|
||||
class Transformer(nn.Module):
|
||||
def __init__(self):
|
||||
super(Transformer, self).__init__()
|
||||
self.encoder = Encoder()
|
||||
self.decoder = Decoder()
|
||||
self.projection = nn.Linear(d_model, tgt_vocab_size, bias=False)
|
||||
def forward(self, enc_inputs, dec_inputs):
|
||||
enc_outputs, enc_self_attns = self.encoder(enc_inputs)
|
||||
dec_outputs, dec_self_attns, dec_enc_attns = self.decoder(dec_inputs, enc_inputs, enc_outputs)
|
||||
dec_logits = self.projection(dec_outputs) # dec_logits : [batch_size x src_vocab_size x tgt_vocab_size]
|
||||
return dec_logits.view(-1, dec_logits.size(-1)), enc_self_attns, dec_self_attns, dec_enc_attns
|
||||
|
||||
def showgraph(attn):
|
||||
attn = attn[-1].squeeze(0)[0]
|
||||
attn = attn.squeeze(0).data.numpy()
|
||||
fig = plt.figure(figsize=(n_heads, n_heads)) # [n_heads, n_heads]
|
||||
ax = fig.add_subplot(1, 1, 1)
|
||||
ax.matshow(attn, cmap='viridis')
|
||||
ax.set_xticklabels(['']+sentences[0].split(), fontdict={'fontsize': 14}, rotation=90)
|
||||
ax.set_yticklabels(['']+sentences[2].split(), fontdict={'fontsize': 14})
|
||||
plt.show()
|
||||
|
||||
if __name__ == '__main__':
|
||||
sentences = ['ich mochte ein bier P', 'S i want a beer', 'i want a beer E']
|
||||
|
||||
# Transformer Parameters
|
||||
# Padding Should be Zero
|
||||
src_vocab = {'P': 0, 'ich': 1, 'mochte': 2, 'ein': 3, 'bier': 4}
|
||||
src_vocab_size = len(src_vocab)
|
||||
|
||||
tgt_vocab = {'P': 0, 'i': 1, 'want': 2, 'a': 3, 'beer': 4, 'S': 5, 'E': 6}
|
||||
number_dict = {i: w for i, w in enumerate(tgt_vocab)}
|
||||
tgt_vocab_size = len(tgt_vocab)
|
||||
|
||||
src_len = 5 # length of source
|
||||
tgt_len = 5 # length of target
|
||||
|
||||
d_model = 512 # Embedding Size
|
||||
d_ff = 2048 # FeedForward dimension
|
||||
d_k = d_v = 64 # dimension of K(=Q), V
|
||||
n_layers = 6 # number of Encoder of Decoder Layer
|
||||
n_heads = 8 # number of heads in Multi-Head Attention
|
||||
|
||||
model = Transformer()
|
||||
|
||||
criterion = nn.CrossEntropyLoss()
|
||||
optimizer = optim.Adam(model.parameters(), lr=0.001)
|
||||
|
||||
enc_inputs, dec_inputs, target_batch = make_batch(sentences)
|
||||
|
||||
for epoch in range(20):
|
||||
optimizer.zero_grad()
|
||||
outputs, enc_self_attns, dec_self_attns, dec_enc_attns = model(enc_inputs, dec_inputs)
|
||||
loss = criterion(outputs, target_batch.contiguous().view(-1))
|
||||
print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.6f}'.format(loss))
|
||||
loss.backward()
|
||||
optimizer.step()
|
||||
|
||||
# Test
|
||||
predict, _, _, _ = model(enc_inputs, dec_inputs)
|
||||
predict = predict.data.max(1, keepdim=True)[1]
|
||||
print(sentences[0], '->', [number_dict[n.item()] for n in predict.squeeze()])
|
||||
|
||||
print('first head of last state enc_self_attns')
|
||||
showgraph(enc_self_attns)
|
||||
|
||||
print('first head of last state dec_self_attns')
|
||||
showgraph(dec_self_attns)
|
||||
|
||||
print('first head of last state dec_enc_attns')
|
||||
showgraph(dec_enc_attns)
|
||||
Reference in New Issue
Block a user