theachyuttiwari
commited on
Commit
·
8a678af
1
Parent(s):
53a7a2f
Upload create_faiss_index.py
Browse files- create_faiss_index.py +67 -0
create_faiss_index.py
ADDED
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
import os
|
3 |
+
|
4 |
+
import faiss
|
5 |
+
import torch
|
6 |
+
from datasets import load_dataset
|
7 |
+
from transformers import AutoTokenizer, DPRContextEncoder
|
8 |
+
|
9 |
+
from common import articles_to_paragraphs, embed_passages
|
10 |
+
|
11 |
+
|
12 |
+
def create_faiss(args):
|
13 |
+
dims = 128
|
14 |
+
min_chars_per_passage = 200
|
15 |
+
device = ("cuda" if torch.cuda.is_available() else "cpu")
|
16 |
+
|
17 |
+
ctx_tokenizer = AutoTokenizer.from_pretrained(args.ctx_encoder_name)
|
18 |
+
ctx_model = DPRContextEncoder.from_pretrained(args.ctx_encoder_name).to(device)
|
19 |
+
_ = ctx_model.eval()
|
20 |
+
|
21 |
+
kilt_wikipedia = load_dataset("kilt_wikipedia", split="full")
|
22 |
+
kilt_wikipedia_columns = ['kilt_id', 'wikipedia_id', 'wikipedia_title', 'text', 'anchors', 'categories',
|
23 |
+
'wikidata_info', 'history']
|
24 |
+
|
25 |
+
kilt_wikipedia_paragraphs = kilt_wikipedia.map(articles_to_paragraphs, batched=True,
|
26 |
+
remove_columns=kilt_wikipedia_columns,
|
27 |
+
batch_size=512,
|
28 |
+
cache_file_name=f"../data/wiki_kilt_paragraphs_full.arrow",
|
29 |
+
desc="Expanding wiki articles into paragraphs")
|
30 |
+
|
31 |
+
# use paragraphs that are not simple fragments or very short sentences
|
32 |
+
# Wikipedia Faiss index needs to fit into a 16 Gb GPU
|
33 |
+
kilt_wikipedia_paragraphs = kilt_wikipedia_paragraphs.filter(
|
34 |
+
lambda x: (x["end_character"] - x["start_character"]) > min_chars_per_passage)
|
35 |
+
|
36 |
+
if not os.path.isfile(args.index_file_name):
|
37 |
+
def embed_passages_for_retrieval(examples):
|
38 |
+
return embed_passages(ctx_model, ctx_tokenizer, examples, max_length=128)
|
39 |
+
|
40 |
+
paragraphs_embeddings = kilt_wikipedia_paragraphs.map(embed_passages_for_retrieval,
|
41 |
+
batched=True, batch_size=512,
|
42 |
+
cache_file_name="../data/kilt_embedded.arrow",
|
43 |
+
desc="Creating faiss index")
|
44 |
+
|
45 |
+
paragraphs_embeddings.add_faiss_index(column="embeddings", custom_index=faiss.IndexFlatIP(dims))
|
46 |
+
paragraphs_embeddings.save_faiss_index("embeddings", args.index_file_name)
|
47 |
+
else:
|
48 |
+
print(f"Faiss index already exists {args.index_file_name}")
|
49 |
+
|
50 |
+
|
51 |
+
if __name__ == "__main__":
|
52 |
+
parser = argparse.ArgumentParser(description="Creates Faiss Wikipedia index file")
|
53 |
+
|
54 |
+
parser.add_argument(
|
55 |
+
"--ctx_encoder_name",
|
56 |
+
default="vblagoje/dpr-ctx_encoder-single-lfqa-base",
|
57 |
+
help="Encoding model to use for passage encoding",
|
58 |
+
)
|
59 |
+
|
60 |
+
parser.add_argument(
|
61 |
+
"--index_file_name",
|
62 |
+
default="../data/kilt_dpr_wikipedia.faiss",
|
63 |
+
help="Faiss index file with passage embeddings",
|
64 |
+
)
|
65 |
+
|
66 |
+
main_args, _ = parser.parse_known_args()
|
67 |
+
create_faiss(main_args)
|