Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from sentence_transformers import SentenceTransformer, util
|
3 |
+
from transformers import (AutoModelForQuestionAnswering,
|
4 |
+
AutoTokenizer, pipeline)
|
5 |
+
|
6 |
+
import pandas as pd
|
7 |
+
import regex as re
|
8 |
+
from urllib import request
|
9 |
+
|
10 |
+
# Select model for question answering
|
11 |
+
model_name = "deepset/roberta-base-squad2"
|
12 |
+
|
13 |
+
# Load model & tokenizer
|
14 |
+
model = AutoModelForQuestionAnswering.from_pretrained(model_name)
|
15 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
16 |
+
|
17 |
+
# Create pipeline
|
18 |
+
pipe = pipeline('question-answering', model=model_name, tokenizer=model_name)
|
19 |
+
|
20 |
+
# Load Harry Potter book corpus from link
|
21 |
+
url = ("https://raw.githubusercontent.com/formcept/whiteboard/master/nbviewer/notebooks/data/harrypotter/Book%201%20-%20The%20Philosopher's%20Stone.txt")
|
22 |
+
response = request.urlopen(url)
|
23 |
+
book1_raw_0 = response.read().decode('utf8')
|
24 |
+
|
25 |
+
# Text pre-processing
|
26 |
+
# Remove page statements
|
27 |
+
book1_raw_1 = re.sub(r'Page \| [0-9]+ Harry Potter [a-zA-Z \-]+J.K. Rowling',
|
28 |
+
'', book1_raw_0)
|
29 |
+
|
30 |
+
# Remove newlines
|
31 |
+
book1_raw_1 = re.sub(r'\n', '', book1_raw_1)
|
32 |
+
|
33 |
+
# Remove periods; this will relevant in the regrouping later
|
34 |
+
book1_raw_1 = re.sub(r'Mr. ', 'Mr ', book1_raw_1)
|
35 |
+
book1_raw_1 = re.sub(r'Ms. ', 'Ms ', book1_raw_1)
|
36 |
+
book1_raw_1 = re.sub(r'Mrs. ', 'Mrs ', book1_raw_1)
|
37 |
+
|
38 |
+
# Group into 3 sentences-long parts
|
39 |
+
paragraphs = re.findall("[^.?!]+[.?!][^.?!]+[.?!][^.?!]+[.?!]", book1_raw_1)
|
40 |
+
|
41 |
+
# Type in HP-related query here
|
42 |
+
query = st.text_area("Hello muggle! What is your question?")
|
43 |
+
|
44 |
+
# Perform sentence embedding on query and sentence groups
|
45 |
+
model_embed_name = 'sentence-transformers/multi-qa-MiniLM-L6-cos-v1'
|
46 |
+
|
47 |
+
model_embed = SentenceTransformer(model_embed_name)
|
48 |
+
doc_emb = model_embed.encode(paragraphs)
|
49 |
+
query_emb = model_embed.encode(query)
|
50 |
+
|
51 |
+
#Compute dot score between query and all document embeddings
|
52 |
+
scores = util.cos_sim(query_emb, doc_emb)[0].cpu().tolist()
|
53 |
+
|
54 |
+
#Combine docs & scores
|
55 |
+
doc_score_pairs = list(zip(paragraphs, scores))
|
56 |
+
|
57 |
+
#Sort by decreasing score and get only 3 most similar groups
|
58 |
+
doc_score_pairs = sorted(doc_score_pairs, key=lambda x: x[1],
|
59 |
+
reverse=True)[:3]
|
60 |
+
|
61 |
+
# Join these similar groups to form the context
|
62 |
+
context = "".join(x[0] for x in doc_score_pairs)
|
63 |
+
|
64 |
+
# Perform the querying
|
65 |
+
QA_input = {'question': query, 'context': context}
|
66 |
+
out = pipe(QA_input)
|
67 |
+
st.json(out)
|