apps/inference.py

from pandas.io.formats.format import return_docstring
import streamlit as st
import pandas as pd
from transformers import AutoTokenizer, AutoModelForMaskedLM
from transformers import pipeline
import os
import json
import random

with open("config.json") as f:
    cfg = json.loads(f.read())


@st.cache(show_spinner=False, persist=True)
def load_model(masked_text, model_name):

    model = AutoModelForMaskedLM.from_pretrained(model_name)
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    nlp = pipeline("fill-mask", model=model, tokenizer=tokenizer)

    MASK_TOKEN = tokenizer.mask_token

    masked_text = masked_text.replace("<mask>", MASK_TOKEN)
    result_sentence = nlp(masked_text)

    return result_sentence[0]["sequence"], result_sentence[0]["token_str"]


def app():
    st.markdown(
        "<h1 style='text-align: center; color: green;'>RoBERTa Hindi</h1>",
        unsafe_allow_html=True,
    )
    st.markdown(
        "This demo uses multiple hindi transformer models for Masked Language Modelling (MLM)."
    )

    models_list = list(cfg["models"].keys())

    models = st.multiselect("Choose models", models_list, models_list)

    target_text_path = "./mlm_custom/mlm_targeted_text.csv"
    target_text_df = pd.read_csv(target_text_path)

    texts = target_text_df["text"]

    st.sidebar.title("Hindi MLM")

    results_df = pd.DataFrame(columns=["Model Name", "Filled Token", "Filled Text"])

    model_names = []
    filled_masked_texts = []
    filled_tokens = []
    pick_random = st.checkbox("Pick any random text")
    if pick_random:
        random_text = texts[random.randint(0, texts.shape[0] - 1)]
        masked_text = st.text_area("Please type a masked sentence to fill", random_text)
    else:
        select_text = st.sidebar.selectbox("Select any of the following text", texts)
        masked_text = st.text_area("Please type a masked sentence to fill", select_text)

    # pd.set_option('max_colwidth',30)
    if st.button("Fill the Mask!"):
        with st.spinner("Filling the Mask..."):

            for selected_model in models:

                filled_sentence, filled_token = load_model(
                    masked_text, cfg["models"][selected_model]
                )
                model_names.append(selected_model)
                filled_tokens.append(filled_token)
                filled_masked_texts.append(filled_sentence)

            results_df["Model Name"] = model_names
            results_df["Filled Token"] = filled_tokens
            results_df["Filled Text"] = filled_masked_texts

            st.table(results_df)