File size: 1,827 Bytes
646aebf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch.nn as nn
import torch
from transformers import AutoModel, AutoConfig

class RefactorSpanModel(nn.Module):
    def __init__(self):
        super().__init__()
        base_model_path = 'microsoft/codebert-base'
        self.base_config = AutoConfig.from_pretrained(base_model_path)
        self.base_model = AutoModel.from_config(self.base_config)
        self.dropout = nn.Dropout(0.5)
        self.classifier = nn.Linear(768, 1)
        self.start_span = nn.Linear(768, 1)

    def forward(self, input_ids):
        outputs = self.base_model(input_ids)
        outputs_pool = self.dropout(outputs[1]) #use pooler output...
        outputs_hidden = self.dropout(outputs[0]) 
        refactor = self.classifier(outputs_pool)
        span = self.start_span(outputs_hidden)
        return refactor, span
    
class RefactorModel(nn.Module):
    def __init__(self):
        super().__init__()
        base_model_path = 'microsoft/codebert-base'
        self.base_config = AutoConfig.from_pretrained(base_model_path)
        self.base_model = AutoModel.from_config(self.base_config)
        self.dropout = nn.Dropout(0.5)
        self.classifier = nn.Linear(768, 1)

    def forward(self, input_ids):
        outputs = self.base_model(input_ids)
        outputs_pool = self.dropout(outputs[1]) #use pooler output...
        refactor = self.classifier(outputs_pool)
        return refactor
    
if __name__ == "__main__":
    checkpoint = 'pytorch_model_RSP.bin'
    model = RefactorSpanModel() 
    model.load_state_dict(torch.load(checkpoint), strict=True)
    #print(model.base_model.embeddings.word_embeddings.weight)
    
    checkpoint = 'pytorch_model_RP.bin'
    model = RefactorModel() 
    model.load_state_dict(torch.load(checkpoint), strict=True)
    #print(model.base_model.embeddings.word_embeddings.weight)