Update app.py
Browse files
app.py
CHANGED
@@ -1,75 +1,53 @@
|
|
1 |
-
import json
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
# TODO: improve layout (columns, sidebar, forms)
|
6 |
-
# st.set_page_config(layout='wide')
|
7 |
-
|
8 |
-
|
9 |
st.title('Question Answering example')
|
10 |
-
|
11 |
-
|
12 |
-
##########################################################
|
13 |
st.subheader('1. A simple question (extractive, closed domain)')
|
14 |
-
##########################################################
|
15 |
-
|
16 |
-
|
17 |
-
WIKI_URL = 'https://en.wikipedia.org/w/api.php'
|
18 |
-
WIKI_QUERY = "?format=json&action=query&prop=extracts&explaintext=1"
|
19 |
-
WIKI_BERT = "&titles=BERT_(language_model)"
|
20 |
-
WIKI_METHOD = 'GET'
|
21 |
-
|
22 |
response = req.request(WIKI_METHOD, f'{WIKI_URL}{WIKI_QUERY}{WIKI_BERT}')
|
23 |
resp_json = json.loads(response.content.decode("utf-8"))
|
24 |
wiki_bert = resp_json['query']['pages']['62026514']['extract']
|
25 |
-
paragraph =
|
26 |
-
|
27 |
-
written_passage = st.text_area(
|
28 |
-
'Paragraph used for QA (you can also edit, or copy/paste new content)',
|
29 |
-
paragraph,
|
30 |
-
height=250
|
31 |
-
)
|
32 |
if written_passage:
|
33 |
paragraph = written_passage
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
written_question = st.text_input(
|
39 |
-
'Question used for QA (you can also edit, and experiment with the answers)',
|
40 |
-
question
|
41 |
-
)
|
42 |
if written_question:
|
43 |
question = written_question
|
44 |
-
|
45 |
-
QA_URL = "https://api-inference.huggingface.co/models/deepset/roberta-base-squad2"
|
46 |
-
QA_METHOD = 'POST'
|
47 |
-
|
48 |
-
|
49 |
if st.button('Run QA inference (get answer prediction)'):
|
50 |
if paragraph and question:
|
51 |
inputs = {'question': question, 'context': paragraph}
|
52 |
payload = json.dumps(inputs)
|
53 |
prediction = req.request(QA_METHOD, QA_URL, data=payload)
|
54 |
answer = json.loads(prediction.content.decode("utf-8"))
|
55 |
-
# >>> answer structure:
|
56 |
-
#
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
73 |
else:
|
74 |
-
st.write('Write some passage of text and a question')
|
75 |
-
|
|
|
1 |
+
import json; import streamlit as st; import requests as req; from transformers import pipeline
|
2 |
+
WIKI_URL = 'https://en.wikipedia.org/w/api.php'; WIKI_BERT = "&titles=BERT_(language_model)"
|
3 |
+
WIKI_QUERY = "?format=json&action=query&prop=extracts&explaintext=1"; WIKI_METHOD = 'GET'
|
4 |
+
pipe_exqa = pipeline("question-answering") #, model="distilbert-base-cased-distilled-squad"
|
|
|
|
|
|
|
|
|
5 |
st.title('Question Answering example')
|
|
|
|
|
|
|
6 |
st.subheader('1. A simple question (extractive, closed domain)')
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
response = req.request(WIKI_METHOD, f'{WIKI_URL}{WIKI_QUERY}{WIKI_BERT}')
|
8 |
resp_json = json.loads(response.content.decode("utf-8"))
|
9 |
wiki_bert = resp_json['query']['pages']['62026514']['extract']
|
10 |
+
paragraph = wiki_bert
|
11 |
+
par_text = 'Paragraph used for QA (you can also edit, or copy/paste new content)'
|
12 |
+
written_passage = st.text_area(par_text, paragraph, height=250)
|
|
|
|
|
|
|
|
|
13 |
if written_passage:
|
14 |
paragraph = written_passage
|
15 |
+
question = 'How many attention heads does Bert have?' # question = 'How many languages does bert understand?'
|
16 |
+
query_text = 'Question used for QA (you can also edit, and experiment with the answers)'
|
17 |
+
written_question = st.text_input(query_text, question)
|
|
|
|
|
|
|
|
|
|
|
18 |
if written_question:
|
19 |
question = written_question
|
20 |
+
QA_URL = "https://api-inference.huggingface.co/models/deepset/roberta-base-squad2"; QA_METHOD = 'POST'
|
|
|
|
|
|
|
|
|
21 |
if st.button('Run QA inference (get answer prediction)'):
|
22 |
if paragraph and question:
|
23 |
inputs = {'question': question, 'context': paragraph}
|
24 |
payload = json.dumps(inputs)
|
25 |
prediction = req.request(QA_METHOD, QA_URL, data=payload)
|
26 |
answer = json.loads(prediction.content.decode("utf-8"))
|
27 |
+
# >>> answer structure: # { "answer": "over 70", "score": 0.240, "start": 35, "end": 62 }
|
28 |
+
answer_dict = dict(answer) # st.write(answer_dict)
|
29 |
+
print(answer_dict)
|
30 |
+
if "answer" in answer_dict.keys():
|
31 |
+
answer_span, answer_score = answer_dict["answer"], answer_dict["score"]
|
32 |
+
st.write(f'Answer: **{answer_span}**')
|
33 |
+
start_par, stop_para = max(0, answer_dict["start"]-86), min(answer_dict["end"]+90, len(paragraph))
|
34 |
+
answer_context = paragraph[start_par:stop_para].replace(answer_span, f'**{answer_span}**')
|
35 |
+
st.write(f'Answer context (and score): ... _{answer_context}_ ... (score: {format(answer_score, ".3f")})')
|
36 |
+
st.write(f'Answer JSON: '); st.write(answer)
|
37 |
+
else:
|
38 |
+
try:
|
39 |
+
qa_result = pipe_exqa(question=question, context=paragraph)
|
40 |
+
except Exception as e:
|
41 |
+
qa_result = str(e)
|
42 |
+
|
43 |
+
if "answer" in qa_result.keys():
|
44 |
+
answer_span, answer_score = qa_result["answer"], qa_result["score"]
|
45 |
+
st.write(f'Answer: **{answer_span}**')
|
46 |
+
start_par, stop_para = max(0, qa_result["start"]-86), min(qa_result["end"]+90, len(paragraph))
|
47 |
+
answer_context = paragraph[start_par:stop_para].replace(answer_span, f'**{answer_span}**')
|
48 |
+
st.write(f'Answer context (and score): ... _{answer_context}_ ... (score: {format(answer_score, ".3f")})')
|
49 |
+
|
50 |
+
st.write(f'Answer JSON: '); st.write(qa_result)
|
51 |
else:
|
52 |
+
st.write('Write some passage of text and a question'); st.stop()
|
53 |
+
# x = st.slider('Select a value'); st.write(x, 'squared is', x * x)
|