sentence classification

1from transformers import AutoTokenizer, AutoModel
2import torch
3
4
5#Mean Pooling - Take attention mask into account for correct averaging
6def mean_pooling(model_output, attention_mask):
7    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
8    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
9    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
10
11
12# Sentences we want sentence embeddings for
13sentences = ['This is an example sentence', 'Each sentence is converted']
14
15# Load model from HuggingFace Hub
16tokenizer = AutoTokenizer.from_pretrained('sentence-transformers/paraphrase-xlm-r-multilingual-v1')
17model = AutoModel.from_pretrained('sentence-transformers/paraphrase-xlm-r-multilingual-v1')
18
19# Tokenize sentences
20encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
21
22# Compute token embeddings
23with torch.no_grad():
24    model_output = model(**encoded_input)
25
26# Perform pooling. In this case, max pooling.
27sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
28
29print("Sentence embeddings:")
30print(sentence_embeddings)