Spaces:
Sleeping
Sleeping
Federico Galatolo
commited on
Commit
β’
168a4de
1
Parent(s):
9532cd7
first commit
Browse files- .gitignore +4 -0
- README.md +1 -3
- app.py +120 -0
- embedders/__pycache__/labse.cpython-38.pyc +0 -0
- embedders/labse.py +26 -0
- requirements.txt +19 -0
.gitignore
ADDED
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
1 |
+
/env
|
2 |
+
/__pycache__/
|
3 |
+
|
4 |
+
.env
|
README.md
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
---
|
2 |
title: Serica Semantic Search
|
3 |
-
emoji:
|
4 |
colorFrom: indigo
|
5 |
colorTo: pink
|
6 |
sdk: streamlit
|
@@ -9,5 +9,3 @@ app_file: app.py
|
|
9 |
pinned: false
|
10 |
license: agpl-3.0
|
11 |
---
|
12 |
-
|
13 |
-
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
|
|
1 |
---
|
2 |
title: Serica Semantic Search
|
3 |
+
emoji: π
|
4 |
colorFrom: indigo
|
5 |
colorTo: pink
|
6 |
sdk: streamlit
|
|
|
9 |
pinned: false
|
10 |
license: agpl-3.0
|
11 |
---
|
|
|
|
app.py
ADDED
@@ -0,0 +1,120 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import streamlit as st
|
3 |
+
from elasticsearch import Elasticsearch
|
4 |
+
|
5 |
+
from embedders.labse import LaBSE
|
6 |
+
|
7 |
+
def search():
|
8 |
+
status_indicator.write(f"Loading model {model_name}...")
|
9 |
+
model = globals()[model_name]()
|
10 |
+
|
11 |
+
status_indicator.write(f"Computing query embeddings...")
|
12 |
+
query_vector = model(query)[0, :].tolist()
|
13 |
+
|
14 |
+
status_indicator.write(f"Performing query...")
|
15 |
+
target_field = f"{model_name}_features"
|
16 |
+
results = es.search(
|
17 |
+
index="sentences",
|
18 |
+
query={
|
19 |
+
"script_score": {
|
20 |
+
"query": {"match_all": {}},
|
21 |
+
"script": {
|
22 |
+
"source": f"cosineSimilarity(params.query_vector, '{target_field}') + 1.0",
|
23 |
+
"params": {"query_vector": query_vector}
|
24 |
+
}
|
25 |
+
}
|
26 |
+
},
|
27 |
+
size=limit
|
28 |
+
)
|
29 |
+
|
30 |
+
for result in results["hits"]["hits"]:
|
31 |
+
sentence = result['_source']['sentence']
|
32 |
+
score = result['_score']
|
33 |
+
document = result['_source']['document']
|
34 |
+
number = result['_source']['number']
|
35 |
+
|
36 |
+
previous = es.search(
|
37 |
+
index="sentences",
|
38 |
+
query={
|
39 |
+
"bool": {
|
40 |
+
"must": [{
|
41 |
+
"term": {
|
42 |
+
"document": document
|
43 |
+
}
|
44 |
+
},{
|
45 |
+
"range": {
|
46 |
+
"number": {
|
47 |
+
"gte": number-3,
|
48 |
+
"lt": number,
|
49 |
+
}
|
50 |
+
}
|
51 |
+
}
|
52 |
+
]
|
53 |
+
}
|
54 |
+
}
|
55 |
+
)
|
56 |
+
|
57 |
+
previous_hits = sorted(previous["hits"]["hits"], key=lambda e: e["_source"]["number"])
|
58 |
+
previous_context = "".join([r["_source"]["sentence"] for r in previous_hits])
|
59 |
+
|
60 |
+
|
61 |
+
subsequent = es.search(
|
62 |
+
index="sentences",
|
63 |
+
query={
|
64 |
+
"bool": {
|
65 |
+
"must": [{
|
66 |
+
"term": {
|
67 |
+
"document": document
|
68 |
+
}
|
69 |
+
},{
|
70 |
+
"range": {
|
71 |
+
"number": {
|
72 |
+
"lte": number+3,
|
73 |
+
"gt": number,
|
74 |
+
}
|
75 |
+
}
|
76 |
+
}
|
77 |
+
]
|
78 |
+
}
|
79 |
+
}
|
80 |
+
)
|
81 |
+
|
82 |
+
subsequent_hits = sorted(subsequent["hits"]["hits"], key=lambda e: e["_source"]["number"])
|
83 |
+
subsequent_context = "".join([r["_source"]["sentence"] for r in subsequent_hits])
|
84 |
+
|
85 |
+
|
86 |
+
document_name_results = es.search(
|
87 |
+
index="documents",
|
88 |
+
query={
|
89 |
+
"bool": {
|
90 |
+
"must": [{
|
91 |
+
"term": {
|
92 |
+
"id": document
|
93 |
+
}
|
94 |
+
}
|
95 |
+
]
|
96 |
+
}
|
97 |
+
}
|
98 |
+
)
|
99 |
+
|
100 |
+
document_name_data = document_name_results["hits"]["hits"][0]["_source"]
|
101 |
+
document_name = f"{document_name_data['title']} - {document_name_data['author']}"
|
102 |
+
|
103 |
+
results_placeholder.markdown(f"#### {document_name} (score: {score:.2f})\n{previous_context} **{sentence}** {subsequent_context}")
|
104 |
+
|
105 |
+
|
106 |
+
status_indicator.write(f"Results ready...")
|
107 |
+
|
108 |
+
es = Elasticsearch(os.environ["ELASTIC_HOST"], basic_auth=os.environ["ELASTIC_AUTH"].split(":"))
|
109 |
+
|
110 |
+
st.header("Serica Semantic Search")
|
111 |
+
st.write("Perform a semantic search using a Sentence Embedding Transformer model on the SERICA database")
|
112 |
+
model_name = st.selectbox("Model", ["LaBSE"])
|
113 |
+
limit = st.number_input("Number of results", 10)
|
114 |
+
query = st.text_input("Query", value="")
|
115 |
+
status_indicator = st.empty()
|
116 |
+
do_search = st.button("Search")
|
117 |
+
results_placeholder = st.container()
|
118 |
+
|
119 |
+
if do_search:
|
120 |
+
search()
|
embedders/__pycache__/labse.cpython-38.pyc
ADDED
Binary file (1.27 kB). View file
|
|
embedders/labse.py
ADDED
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from transformers import BertModel, BertTokenizerFast
|
3 |
+
import torch.nn.functional as F
|
4 |
+
|
5 |
+
class LaBSE:
|
6 |
+
def __init__(self):
|
7 |
+
self.tokenizer = BertTokenizerFast.from_pretrained("setu4993/LaBSE")
|
8 |
+
self.model = BertModel.from_pretrained("setu4993/LaBSE")
|
9 |
+
self.model.eval()
|
10 |
+
|
11 |
+
@torch.no_grad()
|
12 |
+
def __call__(self, sentences):
|
13 |
+
if not isinstance(sentences, list):
|
14 |
+
sentences = [sentences]
|
15 |
+
tokens = self.tokenizer(sentences, return_tensors="pt", padding=True)
|
16 |
+
outputs = self.model(**tokens)
|
17 |
+
embeddings = outputs.pooler_output
|
18 |
+
return F.normalize(embeddings, p=2).cpu().numpy()
|
19 |
+
|
20 |
+
@property
|
21 |
+
def dim(self):
|
22 |
+
return 768
|
23 |
+
|
24 |
+
if __name__ == "__main__":
|
25 |
+
labse = LaBSE()
|
26 |
+
print(labse(["odi et amo", "quare id faciam"]).shape)
|
requirements.txt
ADDED
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
certifi==2022.6.15
|
2 |
+
charset-normalizer==2.1.0
|
3 |
+
elastic-transport==8.1.2
|
4 |
+
elasticsearch==8.3.3
|
5 |
+
filelock==3.7.1
|
6 |
+
huggingface-hub==0.8.1
|
7 |
+
idna==3.3
|
8 |
+
numpy==1.23.1
|
9 |
+
packaging==21.3
|
10 |
+
pyparsing==3.0.9
|
11 |
+
PyYAML==6.0
|
12 |
+
regex==2022.7.25
|
13 |
+
requests==2.28.1
|
14 |
+
tokenizers==0.12.1
|
15 |
+
tqdm==4.64.0
|
16 |
+
transformers==4.21.0
|
17 |
+
typing-extensions==4.3.0
|
18 |
+
urllib3==1.26.11
|
19 |
+
torch==1.12.0
|