app.py

import streamlit as st
from transformers import pipeline
from io import StringIO

unmasker = pipeline('fill-mask', model='dsfsi/zabantu-bantu-250m')

st.set_page_config(layout="wide")

def fill_mask(sentences):
    results = {}
    warnings = []
    for language, sentence in sentences.items():
        if "<mask>" in sentence:
            masked_sentence = sentence.replace('<mask>', unmasker.tokenizer.mask_token)
            unmasked = unmasker(masked_sentence)
            results[language] = unmasked
        else:
            warnings.append(f"Warning: No <mask> token found in sentence: {sentence}")
    return results, warnings

def replace_mask(sentence, predicted_word):
    return sentence.replace("<mask>", f"**{predicted_word}**")

st.title("Fill Mask | Zabantu-XLM-Roberta")
st.write(f"")

st.markdown("Zabantu-XLMR refers to a fleet of models trained on different combinations of South African Bantu languages. It supports the following languages Tshivenda, Nguni languages (Zulu, Xhosa, Swati), Sotho languages (Northern Sotho, Southern Sotho, Setswana), and Xitsonga.")

col1, col2 = st.columns(2)

if 'text_input' not in st.session_state:
    st.session_state['text_input'] = ""

if 'warnings' not in st.session_state:
    st.session_state['warnings'] = []

with col1:
    with st.container(border=True):
        st.markdown("Input :clipboard:")

        select_options = ['Choose option', 'Enter text input', 'Upload a file(csv/txt)']
        sample_sentence = {'tshivenda': "Rabulasi wa <mask> u khou bvelela nga u lima.",
                "tsonga": "N'wana wa xisati u <mask> ku tsaka."
        }
        language_options = ['Choose language', 'Zulu', 'Tshivenda', 'Sepedi', 'Tswana', 'Tsonga']
        
        option_selected = st.selectbox(f"Select an input option:", select_options, index=0)
        input_sentences = {}
        
        if option_selected == 'Enter text input':
            st.session_state['warnings'].clear()
        
            # Initialize session state to preserve language and sentence inputs between reruns
            if 'input_sentences' not in st.session_state:
                st.session_state['input_sentences'] = {}
            if 'languages_selected' not in st.session_state:
                st.session_state['languages_selected'] = {}
        
            input1, input2 = st.columns(2)
        
            for i in range(5):
                # Get the previously selected language and sentence, if available
                previous_language = st.session_state['languages_selected'].get(f'language_{i}', 'Choose language')
                previous_sentence = st.session_state['input_sentences'].get(f'text_input_{i}', '')
        
                # Select language in column 1
                with input1:
                    language = st.selectbox(
                        f"Select language for input {i+1}:",
                        language_options,
                        key=f'language_{i}',
                        index=language_options.index(previous_language) if previous_language in language_options else 0
                    )
                    # Store selected language in session state
                    st.session_state['languages_selected'][f'language_{i}'] = language
        
                # Enter sentence in column 2
                with input2:
                    sentence = st.text_input(
                        f"Enter sentence for input {i+1} (with <mask>):", 
                key=f'text_input_{i}', 
                value=previous_sentence
            )
            # Store input sentence in session state
            st.session_state['input_sentences'][f'text_input_{i}'] = sentence

            if sentence:
                if language and language != 'Choose language':
                    # Add valid language and sentence to input_sentences
                    st.session_state['input_sentences'][f'{language.lower()}_{i+1}'] = sentence
                else:
                    st.session_state['warnings'].append(f"Warning: Choose the language for input {i+1}")

            # Submit button
            if st.button("Submit"):
                if st.session_state['warnings']:
                    # Show warnings if any
                    for warning in st.session_state['warnings']:
                        st.warning(warning)
                else:
                    # Process the sentences if no warnings
                    result, warnings = fill_mask(st.session_state['input_sentences'])
                    st.session_state['warnings'] = warnings


        if option_selected == 'Upload a file(csv/txt)':
            
            uploaded_file = st.file_uploader("Choose a file-(one sentence per line)")
            if uploaded_file is not None:
                
                stringio = StringIO(uploaded_file.getvalue().decode("utf-8"))
                string_data = stringio.read()
                
                input_sentences = string_data.split("\n")
    
                if st.button("Submit",use_container_width=True):
                    if st.session_state['warnings']:
                        for warning in st.session_state['warnings']:
                            st.warning(warning)
                    else:
                        result, warnings = fill_mask(input_sentences)
                        st.session_state['warnings'] = warnings 

                        if st.session_state['warnings']:
                            for warning in st.session_state['warnings']:
                                st.warning(warning)

        st.markdown("Example")
        st.code(sample_sentence, wrap_lines=True)
        if st.button("Test Example",use_container_width=True):
            result, warnings = fill_mask(sample_sentence)

with col2:
    with st.container(border=True):
        st.markdown("Output :bar_chart:")
        if 'result' in locals() and result:  
            if len(result) == 1:
                for language, predictions in result.items():
                    for prediction in predictions:
                        predicted_word = prediction['token_str']
                        score = prediction['score'] * 100
    
                        st.markdown(f"""
                        <div class="bar">
                            <div class="bar-fill" style="width: {score}%;"></div>
                        </div>
                        <div class="container">
                            <div style="align-items: left;">{predicted_word}</div>
                            <div style="align-items: center;">{score:.2f}%</div>
                        </div>
                        """, unsafe_allow_html=True)

            else:
                for language, predictions in result.items():
                    if predictions:
                        top_prediction = predictions[0]
                        predicted_word = top_prediction['token_str']
                        score = top_prediction['score'] * 100
    
                        st.markdown(f"""
                        <div class="bar">
                            <div class="bar-fill" style="width: {score}%;"></div>
                        </div>
                        <div class="container">
                            <div style="align-items: left;">{predicted_word} ({language})</div>
                            <div style="align-items: right;">{score:.2f}%</div>
                        </div>
                        """, unsafe_allow_html=True)

                
if 'result' in locals():  
    if result:
        line = 0
        for sentence, predictions in result.items():
            line += 1
            predicted_word = predictions[0]['token_str']
            full_sentence = replace_mask(sentence, predicted_word)
            st.write(f"**Sentence {line}:** {full_sentence }")

css = """
<style>
footer {display:none !important;}

.gr-button-primary {
    z-index: 14;
    height: 43px;
    width: 130px;
    left: 0px;
    top: 0px;
    padding: 0px;
    cursor: pointer !important; 
    background: none rgb(17, 20, 45) !important;
    border: none !important;
    text-align: center !important;
    font-family: Poppins !important;
    font-size: 14px !important;
    font-weight: 500 !important;
    color: rgb(255, 255, 255) !important;
    line-height: 1 !important;
    border-radius: 12px !important;
    transition: box-shadow 200ms ease 0s, background 200ms ease 0s !important;
    box-shadow: none !important;
}
.gr-button-primary:hover{
    z-index: 14;
    height: 43px;
    width: 130px;
    left: 0px;
    top: 0px;
    padding: 0px;
    cursor: pointer !important;
    background: none rgb(66, 133, 244) !important;
    border: none !important;
    text-align: center !important;
    font-family: Poppins !important;
    font-size: 14px !important;
    font-weight: 500 !important;
    color: rgb(255, 255, 255) !important;
    line-height: 1 !important;
    border-radius: 12px !important;
    transition: box-shadow 200ms ease 0s, background 200ms ease 0s !important;
    box-shadow: rgb(0 0 0 / 23%) 0px 1px 7px 0px !important;
}
.hover\:bg-orange-50:hover {
    --tw-bg-opacity: 1 !important;
    background-color: rgb(229,225,255) !important;
}
.to-orange-200 {
    --tw-gradient-to: rgb(37 56 133 / 37%) !important;
}
.from-orange-400 {
    --tw-gradient-from: rgb(17, 20, 45) !important;
    --tw-gradient-to: rgb(255 150 51 / 0);
    --tw-gradient-stops: var(--tw-gradient-from), var(--tw-gradient-to) !important;
}
.group-hover\:from-orange-500{
    --tw-gradient-from:rgb(17, 20, 45) !important; 
    --tw-gradient-to: rgb(37 56 133 / 37%);
    --tw-gradient-stops: var(--tw-gradient-from), var(--tw-gradient-to) !important;
}
.group:hover .group-hover\:text-orange-500{
    --tw-text-opacity: 1 !important;
    color:rgb(37 56 133 / var(--tw-text-opacity)) !important;
}

.container {
    display: flex;
    justify-content: space-between;
    align-items: center;
    margin-bottom: 5px;
    width: 100%;
}
.bar {
    # width: 70%;
    background-color: #e6e6e6;
    border-radius: 12px;
    overflow: hidden;
    margin-right: 10px;
    height: 5px;
}
.bar-fill {
    background-color: #17152e;
    height: 100%;
    border-radius: 12px;
}

</style>
"""

st.markdown(css, unsafe_allow_html=True)