app.py

import streamlit as st
import pandas as pd

from transformers import pipeline
from sentence_transformers import SentenceTransformer, util
semantic_model = SentenceTransformer('all-MiniLM-L6-v2')

@st.cache(allow_output_mutation=True)
def get_model(model):
	return pipeline("fill-mask", model=model, top_k=100)#set the maximum of tokens to be retrieved after each inference to model


HISTORY_WEIGHT = 100 # set history weight (if found any keyword from history, it will priorities based on its weight)

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")

history_keyword_text = st.text_input("Enter users's history keywords (optional, i.e., 'Gates')", value="")
#history_keyword_text=''

text = st.text_input("Enter a text for auto completion...", value='Where is Bill')
#text='Where is Bill'

semantic_text = st.text_input("Enter users's history semantic (optional, i.e., 'Microsoft or President')", value="Microsoft")
#semantic_text='President'

model = st.selectbox("choose a model", ["roberta-base", "bert-base-uncased"])
#model='roberta-base'
nlp = get_model(model)
#data_load_state = st.text('Loading model...')



if text:
#    data_load_state = st.text('Inference to model...')
    result = nlp(text+' '+nlp.tokenizer.mask_token)
#    data_load_state.text('')
    sem_list=[_.strip() for _ in semantic_text.split(',')]
    if len(semantic_text):
        predicted_seq=[rec['sequence'] for rec in result]
        predicted_embeddings = semantic_model.encode(predicted_seq, convert_to_tensor=True)
        semantic_history_embeddings = semantic_model.encode(sem_list, convert_to_tensor=True)
        cosine_scores = util.cos_sim(predicted_embeddings, semantic_history_embeddings)
    
    for index, r in enumerate(result):
        if len(semantic_text):
#            for j_index in  range(len(sem_list)):
                if len(r['token_str'])>2: #skip spcial chars such as "?"
                    result[index]['score']+=float(sum(cosine_scores[index]))
        if r['token_str'].lower().strip() in history_keyword_text.lower().strip() and len(r['token_str'].lower().strip())>1:
            #found from history, then increase the score of tokens
            result[index]['score']*=HISTORY_WEIGHT
            
    #sort the results        
    df=pd.DataFrame(result).sort_values(by='score', ascending=False)
    
#    show the results as a table
    st.table(df)