|
indexpath= "./wiki-index/knn.index" |
|
wiki_sentence_path="wikipedia-en-sentences.parquet" |
|
|
|
|
|
import faiss |
|
import glob |
|
import numpy as np |
|
import pandas as pd |
|
pd.set_option("display.max_colwidth", 1000) |
|
|
|
import nltk.data |
|
import numpy as np |
|
import time |
|
|
|
import os |
|
|
|
import torch |
|
from transformers import AutoTokenizer, AutoModel |
|
|
|
tokenizer = AutoTokenizer.from_pretrained('facebook/contriever-msmarco') |
|
contriever = AutoModel.from_pretrained('facebook/contriever-msmarco') |
|
|
|
|
|
device = 'cuda' if torch.cuda.is_available() else 'cpu' |
|
contriever.to(device) |
|
|
|
|
|
|
|
|
|
def cos_sim_2d(x, y): |
|
norm_x = x / np.linalg.norm(x, axis=1, keepdims=True) |
|
norm_y = y / np.linalg.norm(y, axis=1, keepdims=True) |
|
return np.matmul(norm_x, norm_y.T) |
|
|
|
|
|
|
|
print(device) |
|
|
|
|
|
|
|
def mean_pooling(token_embeddings, mask): |
|
token_embeddings = token_embeddings.masked_fill(~mask[..., None].bool(), 0.) |
|
sentence_embeddings = token_embeddings.sum(dim=1) / mask.sum(dim=1)[..., None] |
|
return sentence_embeddings |
|
print("loading df") |
|
|
|
df_sententces = pd.read_parquet( wiki_sentence_path , engine='fastparquet') |
|
|
|
|
|
|
|
my_index = faiss.read_index(indexpath, faiss.IO_FLAG_MMAP | faiss.IO_FLAG_READ_ONLY) |
|
|
|
query ="" |
|
|
|
while query != "q": |
|
|
|
query=input("Type in your query: ") |
|
print("Query Text: ", query) |
|
inputs = tokenizer([query], padding=True, truncation=True, return_tensors="pt").to(device) |
|
|
|
outputs = contriever(**inputs) |
|
embeddings = mean_pooling(outputs[0], inputs['attention_mask']) |
|
|
|
query_vector = np.asarray(embeddings .cpu().detach().numpy()).reshape(1, 768) |
|
|
|
|
|
|
|
k = 5 |
|
distances, indices = my_index.search(query_vector, k) |
|
|
|
print(f"Top {k} elements in the dataset for max inner product search:") |
|
for i, (dist, indice) in enumerate(zip(distances[0], indices[0])): |
|
print(f"{i+1}: Vector number {indice:4} with distance {dist}") |
|
|
|
text = str( df_sententces.iloc[[indice]]['text_snippet'] ) |
|
|
|
try: |
|
inputs = tokenizer([str( df_sententces.iloc[[indice-1]]['text_snippet'] ), str( df_sententces.iloc[[indice]]['text_snippet']), str( df_sententces.iloc[[indice+1]]['text_snippet'] ) ], padding=True, truncation=True, return_tensors="pt").to(device) |
|
outputs = contriever(**inputs) |
|
embeddings = mean_pooling(outputs[0], inputs['attention_mask']) |
|
embeddings = np.asarray(embeddings .cpu().detach().numpy()) |
|
|
|
|
|
if cos_sim_2d(embeddings[0].reshape(1, 768), embeddings[1].reshape(1, 768)) > 0.7: |
|
text = str( df_sententces.iloc[[indice-1]]['text_snippet'] ) +" "+ str( df_sententces.iloc[[indice]]['text_snippet'] ) |
|
|
|
|
|
if cos_sim_2d(embeddings[0].reshape(1, 768), embeddings[1].reshape(1, 768)) > 0.7: |
|
text += str( df_sententces.iloc[[indice+1]]['text_snippet'] ) |
|
|
|
except: |
|
pass |
|
|
|
print(text) |
|
|
|
|
|
|
|
|
