README.md

metadata

license: apache-2.0

Overview

Original from the sentences-transformers library.

Only for researching purposes.

Adapter by Aisuko

Installation

!pip install sentence-transformers==2.3.1

Computing Embeddings for a large set of sentences

import os
import csv
import time

from sentence_transformers import SentenceTransformer
from sentence_transformers.util import http_get

if __name__=='__main__':
    url='http://qim.fs.quoracdn.net/quora_duplicate_questions.tsv'
    dataset_path='quora_duplicate_questions.tsv'
#     max_corpus_size=50000  # max number of sentences to deal with
    
    if not os.path.exists(dataset_path):
        http_get(url, dataset_path)
    
    # get all unique sentences from the file
    corpus_sentences=set()
    with open(dataset_path, encoding='utf8') as fIn:
        reader=csv.DictReader(fIn, delimiter='\t', quoting=csv.QUOTE_MINIMAL)
        for row in reader:
            corpus_sentences.add(row['question1'])
            corpus_sentences.add(row['question2'])
#             if len(corpus_sentences)>=max_corpus_size:
#                 break
                
    corpus_sentences=list(corpus_sentences)
    model=SentenceTransformer('all-MiniLM-L6-v2').to('cuda')
    model.max_seq_length=256
    
    pool=model.start_multi_process_pool()
    
    # computing the embeddings using the multi-process pool
    emb=model.encode_multi_process(corpus_sentences, pool,batch_size=128,chunk_size=1024,normalize_embeddings=True)
    print('Embeddings computed. Shape:', emb.shape)
    
    # optional : stop the processes in the pool
    model.stop_multi_process_pool(pool)

Save the csv file

import pandas as pd

corpus_embedding=pd.DataFrame(emb)
corpus_embedding.to_csv('quora_questions.csv',index=False)