app.py

from InstructorEmbedding import INSTRUCTOR
import streamlit as st
import pandas as pd
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np

# if 'model' not in st.session_state:
#     st.session_state['model'] = INSTRUCTOR('hkunlp/instructor-large')

if 'query_bool' not in st.session_state:
    st.session_state['query_bool'] = False

if 'df' not in st.session_state:
    st.session_state['df'] = pd.DataFrame()

@st.cache_resource
def load_model():
    return INSTRUCTOR('hkunlp/instructor-large')

model = load_model()

@st.cache_data(persist='disk')
def embed_data(data):
    return model.encode(data)
    

def process_data(df, desc, message, query):
    data = [
        [
            f'Represent the document for retrieval of {x[desc]} information : ',
            x[message]
        ] for _,x in df.iterrows()
    ]

    corpus_embeddings = embed_data(data)
    
    query_embeddings = model.encode(query)

    with st.spinner('Wait for it...'):
        similarities = cosine_similarity(query_embeddings,corpus_embeddings)
        retrieved_doc_id = np.argmax(similarities)
    st.markdown(f"{data[retrieved_doc_id][-1]}",unsafe_allow_html=True)
    
    # question = st.text_input("Question (Press Enter to query) :")
    # query  = [['Represent the question for retrieving supporting documents: ',question]]
    # btn_q = st.button("Submit", key="submit_query")

    # if btn_q :
    #     query  = [['Represent the question for retrieving supporting documents: ',question]]
    #     query_embeddings = model.encode(query)

    #     with st.spinner('Wait for it...'):
    #         similarities = cosine_similarity(query_embeddings,corpus_embeddings)
    #         retrieved_doc_id = np.argmax(similarities)
    #     st.markdown(f"{data[retrieved_doc_id][-1]}",unsafe_allow_html=True)
    

opt = st.radio("Choose Data : ", ["intent.csv", "upload file CSV"], captions=["LMD CSV intent data", "Custom upload CSV data"])


if opt == "intent.csv":
    df = pd.read_csv("intent.csv", delimiter=";")
    st.dataframe(df)
    question = st.text_input("Question (Press Enter to query) :")
    btn_q = st.button("Submit", key="submit_query")
    if btn_q:
        query  = [['Represent the question for retrieving supporting documents: ',question]]
        process_data(df, desc='description', message='message', query=query)

else :
    f = st.file_uploader("Upload CSV File with at least 2 columns", ['xlsx', 'csv'])
    delim = st.text_input('CSV File Delimiter')
    btn = st.button("Submit", key="submit_first")


    if btn:
        # df = pd.read_csv(f, delimiter=delim)
        st.session_state.df = pd.read_csv(f, delimiter=delim)
        
        if len(list(st.session_state.df.columns)) < 2 :
            st.write("FAILED! At least 2 columns needed. Please check your dataset")
        else :
            st.session_state.query_bool = True
            

    if st.session_state.query_bool:
        st.dataframe(st.session_state.df)
        with st.form("my_form"):
            cols = list(st.session_state.df.columns)
            desc = st.radio("Description Column (e.g. description)", cols)
            message = st.radio("Response Template Column (e.g. message)", cols)
            question = st.text_input("Question (Press Enter to query) :")
            submitted = st.form_submit_button("submit")
            
            if submitted:
                query  = [['Represent the question for retrieving supporting documents: ',question]]
                process_data(st.session_state.df, desc, message, query=query)