File size: 9,151 Bytes
5b36a6d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
from transformers import AutoModel, AutoTokenizer
from scipy.spatial.distance import cosine
import argparse
import json
import pdb
import torch
import torch.nn.functional as F

def read_text(input_file):
    arr = open(input_file).read().split("\n")
    return arr[:-1]


class SimCSEModel:
    def __init__(self):
        self.model = None
        self.tokenizer = None
        self.debug = False
        print("In SimCSE constructor")

    def init_model(self,model_name = None):
        if (model_name == None):
            model_name = "princeton-nlp/sup-simcse-roberta-large"
        #self.model = SimCSE(model_name)
        self.tokenizer = AutoTokenizer.from_pretrained(model_name)
        self.model = AutoModel.from_pretrained(model_name)

    def compute_embeddings(self,input_data,is_file):
        texts = read_text(input_data) if is_file == True else input_data
        inputs = self.tokenizer(texts, padding=True, truncation=True, return_tensors="pt")
        with torch.no_grad():
            embeddings = self.model(**inputs, output_hidden_states=True, return_dict=True).pooler_output
        return texts,embeddings

    def output_results(self,output_file,texts,embeddings,main_index = 0):
        # Calculate cosine similarities
        # Cosine similarities are in [-1, 1]. Higher means more similar
        cosine_dict = {}
        #print("Total sentences",len(texts))
        for i in range(len(texts)):
                cosine_dict[texts[i]] = 1 - cosine(embeddings[main_index], embeddings[i])

        #print("Input sentence:",texts[main_index])
        sorted_dict = dict(sorted(cosine_dict.items(), key=lambda item: item[1],reverse = True))
        if (self.debug):
            for key in sorted_dict:
                print("Cosine similarity with  \"%s\" is: %.3f" % (key, sorted_dict[key]))
        if (output_file is not None):
            with open(output_file,"w") as fp:
                fp.write(json.dumps(sorted_dict,indent=0))
        return sorted_dict



class SGPTModel:
    def __init__(self):
        self.model = None
        self.tokenizer = None
        self.debug = False
        print("In SGPT Constructor")


    def init_model(self,model_name = None):
        # Get our models - The package will take care of downloading the models automatically
        # For best performance: Muennighoff/SGPT-5.8B-weightedmean-nli-bitfit
        if (self.debug):
            print("Init model",model_name)
        if (model_name is None):
            model_name = "Muennighoff/SGPT-125M-weightedmean-nli-bitfit"
        self.tokenizer = AutoTokenizer.from_pretrained(model_name)
        self.model = AutoModel.from_pretrained(model_name)
        #self.tokenizer = AutoTokenizer.from_pretrained("Muennighoff/SGPT-1.3B-weightedmean-msmarco-specb-bitfit")
        #self.model = AutoModel.from_pretrained("Muennighoff/SGPT-1.3B-weightedmean-msmarco-specb-bitfit")
        #self.tokenizer = AutoTokenizer.from_pretrained("Muennighoff/SGPT-5.8B-weightedmean-msmarco-specb-bitfit")
        #self.model = AutoModel.from_pretrained("Muennighoff/SGPT-5.8B-weightedmean-msmarco-specb-bitfit")
        # Deactivate Dropout (There is no dropout in the above models so it makes no difference here but other SGPT models may have dropout)
        self.model.eval()

    def compute_embeddings(self,input_data,is_file):
        if (self.debug):
            print("Computing embeddings for:", input_data[:20])
        model = self.model
        tokenizer = self.tokenizer

        texts = read_text(input_data) if is_file == True else input_data

        # Tokenize input texts
        batch_tokens = tokenizer(texts, padding=True, truncation=True, return_tensors="pt")

    # Get the embeddings
        with torch.no_grad():
            # Get hidden state of shape [bs, seq_len, hid_dim]
            last_hidden_state = model(**batch_tokens, output_hidden_states=True, return_dict=True).last_hidden_state

        # Get weights of shape [bs, seq_len, hid_dim]
        weights = (
            torch.arange(start=1, end=last_hidden_state.shape[1] + 1)
            .unsqueeze(0)
            .unsqueeze(-1)
            .expand(last_hidden_state.size())
            .float().to(last_hidden_state.device)
        )

        # Get attn mask of shape [bs, seq_len, hid_dim]
        input_mask_expanded = (
            batch_tokens["attention_mask"]
            .unsqueeze(-1)
            .expand(last_hidden_state.size())
            .float()
        )

        # Perform weighted mean pooling across seq_len: bs, seq_len, hidden_dim -> bs, hidden_dim
        sum_embeddings = torch.sum(last_hidden_state * input_mask_expanded * weights, dim=1)
        sum_mask = torch.sum(input_mask_expanded * weights, dim=1)

        embeddings = sum_embeddings / sum_mask
        return texts,embeddings

    def output_results(self,output_file,texts,embeddings,main_index = 0):
        # Calculate cosine similarities
        # Cosine similarities are in [-1, 1]. Higher means more similar
        cosine_dict = {}
        if (self.debug):
            print("Total sentences",len(texts))
        for i in range(len(texts)):
                cosine_dict[texts[i]] = 1 - cosine(embeddings[main_index], embeddings[i])

        if (self.debug):
            print("Input sentence:",texts[main_index])
        sorted_dict = dict(sorted(cosine_dict.items(), key=lambda item: item[1],reverse = True))
        if (self.debug):
            for key in sorted_dict:
                print("Cosine similarity with  \"%s\" is: %.3f" % (key, sorted_dict[key]))
        if (output_file is not None):
            with open(output_file,"w") as fp:
                fp.write(json.dumps(sorted_dict,indent=0))
        return sorted_dict





class HFModel:
    def __init__(self):
        self.model = None
        self.tokenizer = None
        self.debug = False
        print("In HF Constructor")


    def init_model(self,model_name = None):
        # Get our models - The package will take care of downloading the models automatically
        # For best performance: Muennighoff/SGPT-5.8B-weightedmean-nli-bitfit
        #print("Init model",model_name)
        if (model_name is None):
            model_name = "sentence-transformers/all-MiniLM-L6-v2"
        self.tokenizer = AutoTokenizer.from_pretrained(model_name)
        self.model = AutoModel.from_pretrained(model_name)
        self.model.eval()

    def mean_pooling(self,model_output, attention_mask):
        token_embeddings = model_output[0] #First element of model_output contains all token embeddings
        input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
        return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)

    def compute_embeddings(self,input_data,is_file):
        #print("Computing embeddings for:", input_data[:20])
        model = self.model
        tokenizer = self.tokenizer

        texts = read_text(input_data) if is_file == True else input_data

        encoded_input = tokenizer(texts, padding=True, truncation=True, return_tensors='pt')

        # Compute token embeddings
        with torch.no_grad():
            model_output = model(**encoded_input)

        # Perform pooling
        sentence_embeddings = self.mean_pooling(model_output, encoded_input['attention_mask'])

        # Normalize embeddings
        sentence_embeddings = F.normalize(sentence_embeddings, p=2, dim=1)

        return texts,sentence_embeddings

    def output_results(self,output_file,texts,embeddings,main_index = 0):
        # Calculate cosine similarities
        # Cosine similarities are in [-1, 1]. Higher means more similar
        cosine_dict = {}
        #print("Total sentences",len(texts))
        for i in range(len(texts)):
                cosine_dict[texts[i]] = 1 - cosine(embeddings[main_index], embeddings[i])

        #print("Input sentence:",texts[main_index])
        sorted_dict = dict(sorted(cosine_dict.items(), key=lambda item: item[1],reverse = True))
        if (self.debug):
            for key in sorted_dict:
                print("Cosine similarity with  \"%s\" is: %.3f" % (key, sorted_dict[key]))
        if (output_file is not None):
            with open(output_file,"w") as fp:
                fp.write(json.dumps(sorted_dict,indent=0))
        return sorted_dict



if __name__ == '__main__':
        parser = argparse.ArgumentParser(description='SGPT model for sentence embeddings ',formatter_class=argparse.ArgumentDefaultsHelpFormatter)
        parser.add_argument('-input', action="store", dest="input",required=True,help="Input file with sentences")
        parser.add_argument('-output', action="store", dest="output",default="output.txt",help="Output file with results")
        parser.add_argument('-model', action="store", dest="model",default="sentence-transformers/all-MiniLM-L6-v2",help="model name")

        results = parser.parse_args()
        obj = HFModel()
        obj.init_model(results.model)
        texts, embeddings = obj.compute_embeddings(results.input,is_file = True)
        results = obj.output_results(results.output,texts,embeddings)