Update app.py
Browse files
app.py
CHANGED
@@ -9,6 +9,7 @@ from optimum.onnxruntime import ORTModelForSequenceClassification
|
|
9 |
from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification
|
10 |
from sentence_transformers import SentenceTransformer, CrossEncoder, util
|
11 |
import streamlit as st
|
|
|
12 |
|
13 |
nltk.download('punkt')
|
14 |
|
@@ -50,18 +51,28 @@ auth_token = os.environ.get("auth_token")
|
|
50 |
|
51 |
progress_bar = st.sidebar.progress(0)
|
52 |
|
53 |
-
@st.experimental_singleton()
|
54 |
def load_models():
|
55 |
asr_model = whisper.load_model("small")
|
56 |
q_model = ORTModelForSequenceClassification.from_pretrained("nickmuchi/quantized-optimum-finbert-tone")
|
|
|
57 |
q_tokenizer = AutoTokenizer.from_pretrained("nickmuchi/quantized-optimum-finbert-tone")
|
|
|
58 |
sent_pipe = pipeline("text-classification",model=q_model, tokenizer=q_tokenizer)
|
59 |
sum_pipe = pipeline("summarization",model="facebook/bart-large-cnn", tokenizer="facebook/bart-large-cnn")
|
|
|
|
|
60 |
cross_encoder = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-12-v2')
|
61 |
|
62 |
-
return asr_model, sent_pipe, sum_pipe, cross_encoder
|
|
|
|
|
|
|
|
|
|
|
63 |
|
64 |
-
|
|
|
65 |
|
66 |
@st.experimental_memo(suppress_st_warning=True)
|
67 |
def inference(link, upload):
|
@@ -131,6 +142,147 @@ def preprocess_plain_text(text,window_size=3):
|
|
131 |
print(f"Passages: {len(passages)}")
|
132 |
|
133 |
return passages
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
134 |
|
135 |
def display_df_as_table(model,top_k,score='score'):
|
136 |
'''Display the df with text and scores as a table'''
|
|
|
9 |
from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification
|
10 |
from sentence_transformers import SentenceTransformer, CrossEncoder, util
|
11 |
import streamlit as st
|
12 |
+
import en_core_web_lg
|
13 |
|
14 |
nltk.download('punkt')
|
15 |
|
|
|
51 |
|
52 |
progress_bar = st.sidebar.progress(0)
|
53 |
|
54 |
+
@st.experimental_singleton(suppress_st_warning=True)
|
55 |
def load_models():
|
56 |
asr_model = whisper.load_model("small")
|
57 |
q_model = ORTModelForSequenceClassification.from_pretrained("nickmuchi/quantized-optimum-finbert-tone")
|
58 |
+
ner_model = AutoModelForTokenClassification.from_pretrained("xlm-roberta-large-finetuned-conll03-english")
|
59 |
q_tokenizer = AutoTokenizer.from_pretrained("nickmuchi/quantized-optimum-finbert-tone")
|
60 |
+
ner_tokenizer = AutoTokenizer.from_pretrained("xlm-roberta-large-finetuned-conll03-english")
|
61 |
sent_pipe = pipeline("text-classification",model=q_model, tokenizer=q_tokenizer)
|
62 |
sum_pipe = pipeline("summarization",model="facebook/bart-large-cnn", tokenizer="facebook/bart-large-cnn")
|
63 |
+
ner_pip = pipeline("ner", model=model, tokenizer=tokenizer, grouped_entities=True)
|
64 |
+
sbert = SentenceTransformer("all-mpnet-base-v2")
|
65 |
cross_encoder = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-12-v2')
|
66 |
|
67 |
+
return asr_model, sent_pipe, sum_pipe, ner_pipe, sbert, cross_encoder
|
68 |
+
|
69 |
+
@st.experimental_singleton(suppress_st_warning=True)
|
70 |
+
def get_spacy():
|
71 |
+
nlp = en_core_web_lg.load()
|
72 |
+
return nlp
|
73 |
|
74 |
+
nlp = get_spacy()
|
75 |
+
asr_model, sent_pipe, sum_pipe, ner_pipe, sbert, cross_encoder = load_models()
|
76 |
|
77 |
@st.experimental_memo(suppress_st_warning=True)
|
78 |
def inference(link, upload):
|
|
|
142 |
print(f"Passages: {len(passages)}")
|
143 |
|
144 |
return passages
|
145 |
+
|
146 |
+
@st.experimental_memo(suppress_st_warning=True)
|
147 |
+
def chunk_clean_text(text):
|
148 |
+
|
149 |
+
"""Chunk text longer than 500 tokens"""
|
150 |
+
|
151 |
+
article = nlp(text)
|
152 |
+
sentences = [i.text for i in list(article.sents)]
|
153 |
+
|
154 |
+
current_chunk = 0
|
155 |
+
chunks = []
|
156 |
+
|
157 |
+
for sentence in sentences:
|
158 |
+
if len(chunks) == current_chunk + 1:
|
159 |
+
if len(chunks[current_chunk]) + len(sentence.split(" ")) <= 500:
|
160 |
+
chunks[current_chunk].extend(sentence.split(" "))
|
161 |
+
else:
|
162 |
+
current_chunk += 1
|
163 |
+
chunks.append(sentence.split(" "))
|
164 |
+
else:
|
165 |
+
chunks.append(sentence.split(" "))
|
166 |
+
|
167 |
+
for chunk_id in range(len(chunks)):
|
168 |
+
chunks[chunk_id] = " ".join(chunks[chunk_id])
|
169 |
+
|
170 |
+
return chunks
|
171 |
+
|
172 |
+
def summary_downloader(raw_text):
|
173 |
+
|
174 |
+
b64 = base64.b64encode(raw_text.encode()).decode()
|
175 |
+
new_filename = "new_text_file_{}_.txt".format(time_str)
|
176 |
+
st.markdown("#### Download Summary as a File ###")
|
177 |
+
href = f'<a href="data:file/txt;base64,{b64}" download="{new_filename}">Click to Download!!</a>'
|
178 |
+
st.markdown(href,unsafe_allow_html=True)
|
179 |
+
|
180 |
+
def get_all_entities_per_sentence(text):
|
181 |
+
doc = nlp(''.join(text))
|
182 |
+
|
183 |
+
sentences = list(doc.sents)
|
184 |
+
|
185 |
+
entities_all_sentences = []
|
186 |
+
for sentence in sentences:
|
187 |
+
entities_this_sentence = []
|
188 |
+
|
189 |
+
# SPACY ENTITIES
|
190 |
+
for entity in sentence.ents:
|
191 |
+
entities_this_sentence.append(str(entity))
|
192 |
+
|
193 |
+
# FLAIR ENTITIES (CURRENTLY NOT USED)
|
194 |
+
# sentence_entities = Sentence(str(sentence))
|
195 |
+
# tagger.predict(sentence_entities)
|
196 |
+
# for entity in sentence_entities.get_spans('ner'):
|
197 |
+
# entities_this_sentence.append(entity.text)
|
198 |
+
|
199 |
+
# XLM ENTITIES
|
200 |
+
entities_xlm = [entity["word"] for entity in ner_model(str(sentence))]
|
201 |
+
for entity in entities_xlm:
|
202 |
+
entities_this_sentence.append(str(entity))
|
203 |
+
|
204 |
+
entities_all_sentences.append(entities_this_sentence)
|
205 |
+
|
206 |
+
return entities_all_sentences
|
207 |
+
|
208 |
+
def get_all_entities(text):
|
209 |
+
all_entities_per_sentence = get_all_entities_per_sentence(text)
|
210 |
+
return list(itertools.chain.from_iterable(all_entities_per_sentence))
|
211 |
+
|
212 |
+
def get_and_compare_entities(article_content,summary_output):
|
213 |
+
|
214 |
+
all_entities_per_sentence = get_all_entities_per_sentence(article_content)
|
215 |
+
entities_article = list(itertools.chain.from_iterable(all_entities_per_sentence))
|
216 |
+
|
217 |
+
all_entities_per_sentence = get_all_entities_per_sentence(summary_output)
|
218 |
+
entities_summary = list(itertools.chain.from_iterable(all_entities_per_sentence))
|
219 |
+
|
220 |
+
matched_entities = []
|
221 |
+
unmatched_entities = []
|
222 |
+
for entity in entities_summary:
|
223 |
+
if any(entity.lower() in substring_entity.lower() for substring_entity in entities_article):
|
224 |
+
matched_entities.append(entity)
|
225 |
+
elif any(
|
226 |
+
np.inner(sentence_embedding_model.encode(entity, show_progress_bar=False),
|
227 |
+
sentence_embedding_model.encode(art_entity, show_progress_bar=False)) > 0.9 for
|
228 |
+
art_entity in entities_article):
|
229 |
+
matched_entities.append(entity)
|
230 |
+
else:
|
231 |
+
unmatched_entities.append(entity)
|
232 |
+
|
233 |
+
matched_entities = list(dict.fromkeys(matched_entities))
|
234 |
+
unmatched_entities = list(dict.fromkeys(unmatched_entities))
|
235 |
+
|
236 |
+
matched_entities_to_remove = []
|
237 |
+
unmatched_entities_to_remove = []
|
238 |
+
|
239 |
+
for entity in matched_entities:
|
240 |
+
for substring_entity in matched_entities:
|
241 |
+
if entity != substring_entity and entity.lower() in substring_entity.lower():
|
242 |
+
matched_entities_to_remove.append(entity)
|
243 |
+
|
244 |
+
for entity in unmatched_entities:
|
245 |
+
for substring_entity in unmatched_entities:
|
246 |
+
if entity != substring_entity and entity.lower() in substring_entity.lower():
|
247 |
+
unmatched_entities_to_remove.append(entity)
|
248 |
+
|
249 |
+
matched_entities_to_remove = list(dict.fromkeys(matched_entities_to_remove))
|
250 |
+
unmatched_entities_to_remove = list(dict.fromkeys(unmatched_entities_to_remove))
|
251 |
+
|
252 |
+
for entity in matched_entities_to_remove:
|
253 |
+
matched_entities.remove(entity)
|
254 |
+
for entity in unmatched_entities_to_remove:
|
255 |
+
unmatched_entities.remove(entity)
|
256 |
+
|
257 |
+
return matched_entities, unmatched_entities
|
258 |
+
|
259 |
+
def highlight_entities(article_content,summary_output):
|
260 |
+
|
261 |
+
markdown_start_red = "<mark class=\"entity\" style=\"background: rgb(238, 135, 135);\">"
|
262 |
+
markdown_start_green = "<mark class=\"entity\" style=\"background: rgb(121, 236, 121);\">"
|
263 |
+
markdown_end = "</mark>"
|
264 |
+
|
265 |
+
matched_entities, unmatched_entities = get_and_compare_entities(article_content,summary_output)
|
266 |
+
|
267 |
+
print(summary_output)
|
268 |
+
|
269 |
+
for entity in matched_entities:
|
270 |
+
summary_output = re.sub(f'({entity})(?![^rgb\(]*\))',markdown_start_green + entity + markdown_end,summary_output)
|
271 |
+
|
272 |
+
for entity in unmatched_entities:
|
273 |
+
summary_output = re.sub(f'({entity})(?![^rgb\(]*\))',markdown_start_red + entity + markdown_end,summary_output)
|
274 |
+
|
275 |
+
print("")
|
276 |
+
print(summary_output)
|
277 |
+
|
278 |
+
print("")
|
279 |
+
print(summary_output)
|
280 |
+
|
281 |
+
soup = BeautifulSoup(summary_output, features="html.parser")
|
282 |
+
|
283 |
+
return HTML_WRAPPER.format(soup)
|
284 |
+
|
285 |
+
nlp = get_spacy()
|
286 |
|
287 |
def display_df_as_table(model,top_k,score='score'):
|
288 |
'''Display the df with text and scores as a table'''
|