-
Notifications
You must be signed in to change notification settings - Fork 0
/
GCNwithUnilm.py
130 lines (110 loc) · 5.69 KB
/
GCNwithUnilm.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
import sys
sys.path.append(".")
import torch
import torch.nn as nn
import torch.nn.functional as F
import copy
import math
import dgl
import os
import numpy as np
import sys
import pickle
sys.path.append(".")
from transformers.models.bert.modeling_bert import BertOnlyMLMHead
from modeling_unilm import LabelSmoothingLoss, BertEncoder, UnilmPreTrainedModel, UnilmModel, UnilmModelIncr, \
BertDecoder, UnilmConfig
from graph_attention import GATlayer
BertLayerNorm = nn.LayerNorm
class chatbotModel(UnilmPreTrainedModel):
def __init__(self, config):
super(chatbotModel, self).__init__(config)
self.config = config
self.num_hidden_layers = config.num_hidden_layers
self.bert = UnilmModel(config)
self.cls = BertOnlyMLMHead(config)
self.crit_mask_lm = nn.CrossEntropyLoss(reduction='none')
if hasattr(config, 'label_smoothing') and config.label_smoothing:
self.crit_mask_lm_smoothed = LabelSmoothingLoss(
config.label_smoothing, config.vocab_size, ignore_index=0, reduction='none')
else:
self.crit_mask_lm_smoothed = None
self.init_weights()
self.tie_weights()
self.graph_fc1 = nn.Linear(config.graph_in_feat, config.graph_out_feat, bias=False)
self.graph_model = nn.ModuleList([GATlayer(config.graph_in_feat, config.graph_out_feat, config.hidden_size,
config.graph_drop, config.leaky_negative_slope) for _ in
range(config.num_attention_graphs)])
def tie_weights(self):
self._tie_or_clone_weights(self.cls.predictions.decoder,
self.bert.embeddings.word_embeddings)
def get_extended_attention_mask(self, input_ids, token_type_ids, attention_mask):
if attention_mask is None:
attention_mask = torch.ones_like(input_ids)
if token_type_ids is None:
token_type_ids = torch.zeros_like(input_ids)
if attention_mask.dim() == 2:
extended_attention_mask = attention_mask.unsqueeze(1).unsqueeze(2)
elif attention_mask.dim() == 3:
extended_attention_mask = attention_mask.unsqueeze(1)
else:
raise NotImplementedError
extended_attention_mask = extended_attention_mask.to(
dtype=torch.float16 if self.config.fp16 else torch.float32) # fp16 compatibility
extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
return extended_attention_mask
def forward(self, graph, input_ids, token_type_ids, encoder_mask=None, decoder_mask=None,
masked_lm_labels=None, masked_pos=None, masked_weights=None, num_tokens_a=None, num_tokens_b=None):
extended_attention_mask = self.get_extended_attention_mask(
input_ids, token_type_ids, encoder_mask)
embedding_output = self.bert.embeddings(
input_ids, token_type_ids)
hidden_states = embedding_output
for l in range(int(self.num_hidden_layers / 2)):
hidden_states = self.bert.encoder.layer[l](hidden_states, extended_attention_mask)
sequence_output = hidden_states
cls = self.bert.pooler(sequence_output)
for gindex, g in enumerate(graph):
g.ndata['q'] = cls[gindex].expand((g.num_nodes(), self.config.hidden_size))
graph_batch = dgl.batch(graph)
graph_batch.ndata['h'] = self.bert.embeddings(graph_batch.ndata['h']).view(-1, self.config.hidden_size * 4)
graph_batch.ndata['h'] = self.graph_fc1(graph_batch.ndata['h'])
for graph_layer in self.graph_model:
graph_batch = graph_layer(graph_batch)
graphs = dgl.unbatch(graph_batch)
knowledge_distill = torch.stack([g.ndata['h'][0].view((4,-1)) for g in graphs])
ks_loss = sum([g.ndata['loss'][0] / (g.ndata['tar_num'][0]+1) for g in graphs])
ks_loss = ks_loss/len(graphs)
hidden_states = torch.cat((knowledge_distill, sequence_output), dim=1)
extended_attention_mask = self.get_extended_attention_mask(
input_ids, token_type_ids, decoder_mask)
for l in range(int(self.num_hidden_layers / 2), self.num_hidden_layers):
hidden_states = self.bert.encoder.layer[l](hidden_states, extended_attention_mask)
sequence_output = hidden_states[:, 4 * self.config.num_attention_graphs:, :]
def gather_seq_out_by_pos(seq, pos):
return torch.gather(seq, 1, pos.unsqueeze(2).expand(-1, -1, seq.size(-1)))
def loss_mask_and_normalize(loss, mask):
mask = mask.type_as(loss)
loss = loss * mask
denominator = torch.sum(mask) + 1e-5
return (loss / denominator).sum()
if masked_lm_labels is None:
if masked_pos is None:
prediction_scores = self.cls(sequence_output)
else:
sequence_output_masked = gather_seq_out_by_pos(
sequence_output, masked_pos)
prediction_scores = self.cls(sequence_output_masked)
return prediction_scores
sequence_output_masked = gather_seq_out_by_pos(
sequence_output, masked_pos)
prediction_scores_masked = self.cls(sequence_output_masked)
if self.crit_mask_lm_smoothed:
masked_lm_loss = self.crit_mask_lm_smoothed(
F.log_softmax(prediction_scores_masked.float(), dim=-1), masked_lm_labels)
else:
masked_lm_loss = self.crit_mask_lm(
prediction_scores_masked.transpose(1, 2).float(), masked_lm_labels)
masked_lm_loss = loss_mask_and_normalize(
masked_lm_loss.float(), masked_weights)
return masked_lm_loss, ks_loss