Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -7,7 +7,7 @@ semantic_model = SentenceTransformer('all-MiniLM-L6-v2')
|
|
7 |
|
8 |
@st.cache(allow_output_mutation=True)
|
9 |
def get_model(model):
|
10 |
-
return pipeline("fill-mask", model=model, top_k=100)#
|
11 |
|
12 |
|
13 |
HISTORY_WEIGHT = 100 # set history weight (if found any keyword from history, it will priorities based on its weight)
|
@@ -15,28 +15,37 @@ HISTORY_WEIGHT = 100 # set history weight (if found any keyword from history, it
|
|
15 |
st.caption("This is a simple auto-completion where the next token is predicted per probability and a weigh if appears in user's history")
|
16 |
|
17 |
history_keyword_text = st.text_input("Enter users's history keywords (optional, i.e., 'Gates')", value="")
|
|
|
18 |
|
19 |
text = st.text_input("Enter a text for auto completion...", value='Where is Bill')
|
20 |
-
|
21 |
|
22 |
-
|
|
|
23 |
|
24 |
-
|
|
|
25 |
nlp = get_model(model)
|
|
|
|
|
|
|
26 |
|
27 |
if text:
|
28 |
-
data_load_state = st.text('Inference to model...')
|
29 |
result = nlp(text+' '+nlp.tokenizer.mask_token)
|
30 |
-
data_load_state.text('')
|
31 |
-
|
32 |
-
if len(semantic_text):
|
33 |
-
|
34 |
-
|
|
|
35 |
cosine_scores = util.cos_sim(predicted_embeddings, semantic_history_embeddings)
|
36 |
|
37 |
for index, r in enumerate(result):
|
38 |
if len(semantic_text):
|
39 |
-
|
|
|
|
|
40 |
if r['token_str'].lower().strip() in history_keyword_text.lower().strip() and len(r['token_str'].lower().strip())>1:
|
41 |
#found from history, then increase the score of tokens
|
42 |
result[index]['score']*=HISTORY_WEIGHT
|
@@ -44,5 +53,5 @@ if text:
|
|
44 |
#sort the results
|
45 |
df=pd.DataFrame(result).sort_values(by='score', ascending=False)
|
46 |
|
47 |
-
|
48 |
st.table(df)
|
|
|
7 |
|
8 |
@st.cache(allow_output_mutation=True)
|
9 |
def get_model(model):
|
10 |
+
return pipeline("fill-mask", model=model, top_k=100)#set the maximum of tokens to be retrieved after each inference to model
|
11 |
|
12 |
|
13 |
HISTORY_WEIGHT = 100 # set history weight (if found any keyword from history, it will priorities based on its weight)
|
|
|
15 |
st.caption("This is a simple auto-completion where the next token is predicted per probability and a weigh if appears in user's history")
|
16 |
|
17 |
history_keyword_text = st.text_input("Enter users's history keywords (optional, i.e., 'Gates')", value="")
|
18 |
+
#history_keyword_text=''
|
19 |
|
20 |
text = st.text_input("Enter a text for auto completion...", value='Where is Bill')
|
21 |
+
#text='Where is Bill'
|
22 |
|
23 |
+
semantic_text = st.text_input("Enter users's history semantic (optional, i.e., 'Microsoft or President')", value="Microsoft")
|
24 |
+
#semantic_text='President'
|
25 |
|
26 |
+
model = st.selectbox("choose a model", ["roberta-base", "bert-base-uncased"])
|
27 |
+
#model='roberta-base'
|
28 |
nlp = get_model(model)
|
29 |
+
#data_load_state = st.text('Loading model...')
|
30 |
+
|
31 |
+
|
32 |
|
33 |
if text:
|
34 |
+
# data_load_state = st.text('Inference to model...')
|
35 |
result = nlp(text+' '+nlp.tokenizer.mask_token)
|
36 |
+
# data_load_state.text('')
|
37 |
+
sem_list=[_.strip() for _ in semantic_text.split(',')]
|
38 |
+
if len(semantic_text):
|
39 |
+
predicted_seq=[rec['sequence'] for rec in result]
|
40 |
+
predicted_embeddings = semantic_model.encode(predicted_seq, convert_to_tensor=True)
|
41 |
+
semantic_history_embeddings = semantic_model.encode(sem_list, convert_to_tensor=True)
|
42 |
cosine_scores = util.cos_sim(predicted_embeddings, semantic_history_embeddings)
|
43 |
|
44 |
for index, r in enumerate(result):
|
45 |
if len(semantic_text):
|
46 |
+
# for j_index in range(len(sem_list)):
|
47 |
+
if len(r['token_str'])>2: #skip spcial chars such as "?"
|
48 |
+
result[index]['score']+=float(sum(cosine_scores[index]))
|
49 |
if r['token_str'].lower().strip() in history_keyword_text.lower().strip() and len(r['token_str'].lower().strip())>1:
|
50 |
#found from history, then increase the score of tokens
|
51 |
result[index]['score']*=HISTORY_WEIGHT
|
|
|
53 |
#sort the results
|
54 |
df=pd.DataFrame(result).sort_values(by='score', ascending=False)
|
55 |
|
56 |
+
# show the results as a table
|
57 |
st.table(df)
|