Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -2,18 +2,22 @@ import streamlit as st
|
|
2 |
import pandas as pd
|
3 |
|
4 |
from transformers import pipeline
|
|
|
|
|
5 |
|
6 |
@st.cache(allow_output_mutation=True)
|
7 |
def get_model(model):
|
8 |
return pipeline("fill-mask", model=model, top_k=100)#seto maximum of tokens to be retrieved after each inference to model
|
9 |
|
|
|
10 |
HISTORY_WEIGHT = 100 # set history weight (if found any keyword from history, it will priorities based on its weight)
|
11 |
|
12 |
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")
|
13 |
|
14 |
-
history_keyword_text = st.text_input("Enter users's history keywords (optional, i.e., 'Gates')", value="
|
15 |
|
16 |
text = st.text_input("Enter a text for auto completion...", value='Where is Bill')
|
|
|
17 |
|
18 |
model = st.selectbox("choose a model", ["roberta-base", "bert-base-uncased"])
|
19 |
|
@@ -24,12 +28,26 @@ if text:
|
|
24 |
data_load_state = st.text('Inference to model...')
|
25 |
result = nlp(text+' '+nlp.tokenizer.mask_token)
|
26 |
data_load_state.text('')
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
for index, r in enumerate(result):
|
|
|
28 |
if r['token_str'].lower().strip() in history_keyword_text.lower().strip() and len(r['token_str'].lower().strip())>1:
|
29 |
#found from history, then increase the score of tokens
|
30 |
result[index]['score']*=HISTORY_WEIGHT
|
31 |
|
|
|
|
|
32 |
#sort the results
|
33 |
df=pd.DataFrame(result).sort_values(by='score', ascending=False)
|
|
|
|
|
|
|
|
|
|
|
34 |
#show the results as a table
|
35 |
st.table(df)
|
|
|
2 |
import pandas as pd
|
3 |
|
4 |
from transformers import pipeline
|
5 |
+
from sentence_transformers import SentenceTransformer, util
|
6 |
+
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)#seto 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)
|
14 |
|
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 |
+
semantic_text = st.text_input("Enter users's history semantic (optional, i.e., 'Microsoft')", value="Microsoft")
|
21 |
|
22 |
model = st.selectbox("choose a model", ["roberta-base", "bert-base-uncased"])
|
23 |
|
|
|
28 |
data_load_state = st.text('Inference to model...')
|
29 |
result = nlp(text+' '+nlp.tokenizer.mask_token)
|
30 |
data_load_state.text('')
|
31 |
+
|
32 |
+
predicted_embeddings = model.encode(result['sequence'], convert_to_tensor=True)
|
33 |
+
semantic_history_embeddings = model.encode(semantic_text.spllit(','), convert_to_tensor=True)
|
34 |
+
|
35 |
+
cosine_scores = util.cos_sim(embeddings1, embeddings2)
|
36 |
+
|
37 |
for index, r in enumerate(result):
|
38 |
+
result[index]['score']=cosine_scores[index][index]
|
39 |
if r['token_str'].lower().strip() in history_keyword_text.lower().strip() and len(r['token_str'].lower().strip())>1:
|
40 |
#found from history, then increase the score of tokens
|
41 |
result[index]['score']*=HISTORY_WEIGHT
|
42 |
|
43 |
+
|
44 |
+
|
45 |
#sort the results
|
46 |
df=pd.DataFrame(result).sort_values(by='score', ascending=False)
|
47 |
+
|
48 |
+
|
49 |
+
|
50 |
+
|
51 |
+
|
52 |
#show the results as a table
|
53 |
st.table(df)
|