{ "cells": [ { "cell_type": "code", "metadata": {}, "source": [ "# code by Tae Hwan Jung(Jeff Jung) @graykode\n", "# Reference : https://github.com/jadore801120/attention-is-all-you-need-pytorch\n", "# https://github.com/JayParks/transformer, https://github.com/dhlee347/pytorchic-bert\n", "import math\n", "import re\n", "from random import *\n", "import numpy as np\n", "import torch\n", "import torch.nn as nn\n", "import torch.optim as optim\n", "\n", "# sample IsNext and NotNext to be same in small batch size\n", "def make_batch():\n", " batch = []\n", " positive = negative = 0\n", " while positive != batch_size/2 or negative != batch_size/2:\n", " tokens_a_index, tokens_b_index= randrange(len(sentences)), randrange(len(sentences)) # sample random index in sentences\n", " tokens_a, tokens_b= token_list[tokens_a_index], token_list[tokens_b_index]\n", " input_ids = [word_dict['[CLS]']] + tokens_a + [word_dict['[SEP]']] + tokens_b + [word_dict['[SEP]']]\n", " segment_ids = [0] * (1 + len(tokens_a) + 1) + [1] * (len(tokens_b) + 1)\n", "\n", " # MASK LM\n", " n_pred = min(max_pred, max(1, int(round(len(input_ids) * 0.15)))) # 15 % of tokens in one sentence\n", " cand_maked_pos = [i for i, token in enumerate(input_ids)\n", " if token != word_dict['[CLS]'] and token != word_dict['[SEP]']]\n", " shuffle(cand_maked_pos)\n", " masked_tokens, masked_pos = [], []\n", " for pos in cand_maked_pos[:n_pred]:\n", " masked_pos.append(pos)\n", " masked_tokens.append(input_ids[pos])\n", " if random() < 0.8: # 80%\n", " input_ids[pos] = word_dict['[MASK]'] # make mask\n", " elif random() < 0.5: # 10%\n", " index = randint(0, vocab_size - 1) # random index in vocabulary\n", " input_ids[pos] = word_dict[number_dict[index]] # replace\n", "\n", " # Zero Paddings\n", " n_pad = maxlen - len(input_ids)\n", " input_ids.extend([0] * n_pad)\n", " segment_ids.extend([0] * n_pad)\n", "\n", " # Zero Padding (100% - 15%) tokens\n", " if max_pred > n_pred:\n", " n_pad = max_pred - n_pred\n", " masked_tokens.extend([0] * n_pad)\n", " masked_pos.extend([0] * n_pad)\n", "\n", " if tokens_a_index + 1 == tokens_b_index and positive < batch_size/2:\n", " batch.append([input_ids, segment_ids, masked_tokens, masked_pos, True]) # IsNext\n", " positive += 1\n", " elif tokens_a_index + 1 != tokens_b_index and negative < batch_size/2:\n", " batch.append([input_ids, segment_ids, masked_tokens, masked_pos, False]) # NotNext\n", " negative += 1\n", " return batch\n", "# Proprecessing Finished\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 gelu(x):\n", " \"Implementation of the gelu activation function by Hugging Face\"\n", " return x * 0.5 * (1.0 + torch.erf(x / math.sqrt(2.0)))\n", "\n", "class Embedding(nn.Module):\n", " def __init__(self):\n", " super(Embedding, self).__init__()\n", " self.tok_embed = nn.Embedding(vocab_size, d_model) # token embedding\n", " self.pos_embed = nn.Embedding(maxlen, d_model) # position embedding\n", " self.seg_embed = nn.Embedding(n_segments, d_model) # segment(token type) embedding\n", " self.norm = nn.LayerNorm(d_model)\n", "\n", " def forward(self, x, seg):\n", " seq_len = x.size(1)\n", " pos = torch.arange(seq_len, dtype=torch.long)\n", " pos = pos.unsqueeze(0).expand_as(x) # (seq_len,) -> (batch_size, seq_len)\n", " embedding = self.tok_embed(x) + self.pos_embed(pos) + self.seg_embed(seg)\n", " return self.norm(embedding)\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", " 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 = nn.Linear(n_heads * d_v, d_model)(context)\n", " return nn.LayerNorm(d_model)(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.fc1 = nn.Linear(d_model, d_ff)\n", " self.fc2 = nn.Linear(d_ff, d_model)\n", "\n", " def forward(self, x):\n", " # (batch_size, len_seq, d_model) -> (batch_size, len_seq, d_ff) -> (batch_size, len_seq, d_model)\n", " return self.fc2(gelu(self.fc1(x)))\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 BERT(nn.Module):\n", " def __init__(self):\n", " super(BERT, self).__init__()\n", " self.embedding = Embedding()\n", " self.layers = nn.ModuleList([EncoderLayer() for _ in range(n_layers)])\n", " self.fc = nn.Linear(d_model, d_model)\n", " self.activ1 = nn.Tanh()\n", " self.linear = nn.Linear(d_model, d_model)\n", " self.activ2 = gelu\n", " self.norm = nn.LayerNorm(d_model)\n", " self.classifier = nn.Linear(d_model, 2)\n", " # decoder is shared with embedding layer\n", " embed_weight = self.embedding.tok_embed.weight\n", " n_vocab, n_dim = embed_weight.size()\n", " self.decoder = nn.Linear(n_dim, n_vocab, bias=False)\n", " self.decoder.weight = embed_weight\n", " self.decoder_bias = nn.Parameter(torch.zeros(n_vocab))\n", "\n", " def forward(self, input_ids, segment_ids, masked_pos):\n", " output = self.embedding(input_ids, segment_ids)\n", " enc_self_attn_mask = get_attn_pad_mask(input_ids, input_ids)\n", " for layer in self.layers:\n", " output, enc_self_attn = layer(output, enc_self_attn_mask)\n", " # output : [batch_size, len, d_model], attn : [batch_size, n_heads, d_mode, d_model]\n", " # it will be decided by first token(CLS)\n", " h_pooled = self.activ1(self.fc(output[:, 0])) # [batch_size, d_model]\n", " logits_clsf = self.classifier(h_pooled) # [batch_size, 2]\n", "\n", " masked_pos = masked_pos[:, :, None].expand(-1, -1, output.size(-1)) # [batch_size, max_pred, d_model]\n", " # get masked position from final output of transformer.\n", " h_masked = torch.gather(output, 1, masked_pos) # masking position [batch_size, max_pred, d_model]\n", " h_masked = self.norm(self.activ2(self.linear(h_masked)))\n", " logits_lm = self.decoder(h_masked) + self.decoder_bias # [batch_size, max_pred, n_vocab]\n", "\n", " return logits_lm, logits_clsf\n", "\n", "if __name__ == '__main__':\n", " # BERT Parameters\n", " maxlen = 30 # maximum of length\n", " batch_size = 6\n", " max_pred = 5 # max tokens of prediction\n", " n_layers = 6 # number of Encoder of Encoder Layer\n", " n_heads = 12 # number of heads in Multi-Head Attention\n", " d_model = 768 # Embedding Size\n", " d_ff = 768 * 4 # 4*d_model, FeedForward dimension\n", " d_k = d_v = 64 # dimension of K(=Q), V\n", " n_segments = 2\n", "\n", " text = (\n", " 'Hello, how are you? I am Romeo.\\n'\n", " 'Hello, Romeo My name is Juliet. Nice to meet you.\\n'\n", " 'Nice meet you too. How are you today?\\n'\n", " 'Great. My baseball team won the competition.\\n'\n", " 'Oh Congratulations, Juliet\\n'\n", " 'Thanks you Romeo'\n", " )\n", " sentences = re.sub(\"[.,!?\\\\-]\", '', text.lower()).split('\\n') # filter '.', ',', '?', '!'\n", " word_list = list(set(\" \".join(sentences).split()))\n", " word_dict = {'[PAD]': 0, '[CLS]': 1, '[SEP]': 2, '[MASK]': 3}\n", " for i, w in enumerate(word_list):\n", " word_dict[w] = i + 4\n", " number_dict = {i: w for i, w in enumerate(word_dict)}\n", " vocab_size = len(word_dict)\n", "\n", " token_list = list()\n", " for sentence in sentences:\n", " arr = [word_dict[s] for s in sentence.split()]\n", " token_list.append(arr)\n", "\n", " model = BERT()\n", " criterion = nn.CrossEntropyLoss()\n", " optimizer = optim.Adam(model.parameters(), lr=0.001)\n", "\n", " batch = make_batch()\n", " input_ids, segment_ids, masked_tokens, masked_pos, isNext = map(torch.LongTensor, zip(*batch))\n", "\n", " for epoch in range(100):\n", " optimizer.zero_grad()\n", " logits_lm, logits_clsf = model(input_ids, segment_ids, masked_pos)\n", " loss_lm = criterion(logits_lm.transpose(1, 2), masked_tokens) # for masked LM\n", " loss_lm = (loss_lm.float()).mean()\n", " loss_clsf = criterion(logits_clsf, isNext) # for sentence classification\n", " loss = loss_lm + loss_clsf\n", " if (epoch + 1) % 10 == 0:\n", " print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.6f}'.format(loss))\n", " loss.backward()\n", " optimizer.step()\n", "\n", " # Predict mask tokens ans isNext\n", " input_ids, segment_ids, masked_tokens, masked_pos, isNext = map(torch.LongTensor, zip(batch[0]))\n", " print(text)\n", " print([number_dict[w.item()] for w in input_ids[0] if number_dict[w.item()] != '[PAD]'])\n", "\n", " logits_lm, logits_clsf = model(input_ids, segment_ids, masked_pos)\n", " logits_lm = logits_lm.data.max(2)[1][0].data.numpy()\n", " print('masked tokens list : ',[pos.item() for pos in masked_tokens[0] if pos.item() != 0])\n", " print('predict masked tokens list : ',[pos for pos in logits_lm if pos != 0])\n", "\n", " logits_clsf = logits_clsf.data.max(1)[1].data.numpy()[0]\n", " print('isNext : ', True if isNext else False)\n", " print('predict isNext : ',True if logits_clsf else False)\n" ], "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 }