app.py

# coding=utf-8
# Copyright 2023 The GIRT Authors.
# Lint as: python3


# This space is built based on AMR-KELEG/ALDi and cis-lmu/GlotLID space.
# GIRT Space

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import streamlit as st
import pandas as pd
import base64


@st.cache_data
def render_svg(svg):
    """Renders the given svg string."""
    b64 = base64.b64encode(svg.encode("utf-8")).decode("utf-8")
    html = rf'<p align="center"> <img src="data:image/svg+xml;base64,{b64}", width="40%"/> </p>'
    c = st.container()
    c.write(html, unsafe_allow_html=True)


@st.cache_resource
def load_model(model_name):
    model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
    return model

@st.cache_resource
def load_tokenizer(model_name):
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    return tokenizer

with st.spinner(text="Please wait while the model is loading...."):

    model = load_model('nafisehNik/girt-t5-base')
    tokenizer = load_tokenizer('nafisehNik/girt-t5-base')


def create_instruction(name, about, title, labels, assignees, headline_type, headline, summary):
    val_list = [name, about, title, labels, assignees, headline_type, headline]

    val_list = ['<|MASK|>' if not element else element for element in val_list]
    if not summary:
        summary = '<|EMPTY|>'
    
    instruction = f'name: {value_list[0]}\nabout: {value_list[1]}\ntitle: {value_list[2]}\nlabels: {value_list[3]}\nassignees: {value_list[4]}\nheadlines_type: {value_list[5]}\nheadlines: {value_list[6]}\nsummary: {summary}'
    return instruction

def compute(sample, top_p, top_k, do_sample, max_length, min_length):

    inputs = tokenizer(sample, return_tensors="pt").to('cpu')

    outputs = model.generate(
        **inputs,
        min_length= min_length,
        max_length=max_length,
        do_sample=do_sample,
        top_p=top_p,
        top_k=top_k).to('cpu')

    generated_texts = tokenizer.batch_decode(outputs, skip_special_tokens=False)
    generated_text = generated_texts[0]
    
    replace_dict = {
        '\n ': '\n',
        '</s>': '',
        '<pad> ': '',
        '<pad>': '',
        '<unk>': ''
    }
    
    postprocess_text = generated_text
    for key, value in replace_dict.items():
        postprocess_text = postprocess_text.replace(key, value)


    return postprocess_text

st.markdown("[![Duplicate Space](https://img.shields.io/badge/-Duplicate%20Space-blue?labelColor=white&style=flat&logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAAAXNSR0IArs4c6QAAAP5JREFUOE+lk7FqAkEURY+ltunEgFXS2sZGIbXfEPdLlnxJyDdYB62sbbUKpLbVNhyYFzbrrA74YJlh9r079973psed0cvUD4A+4HoCjsA85X0Dfn/RBLBgBDxnQPfAEJgBY+A9gALA4tcbamSzS4xq4FOQAJgCDwV2CPKV8tZAJcAjMMkUe1vX+U+SMhfAJEHasQIWmXNN3abzDwHUrgcRGmYcgKe0bxrblHEB4E/pndMazNpSZGcsZdBlYJcEL9Afo75molJyM2FxmPgmgPqlWNLGfwZGG6UiyEvLzHYDmoPkDDiNm9JR9uboiONcBXrpY1qmgs21x1QwyZcpvxt9NS09PlsPAAAAAElFTkSuQmCC&logoWidth=14)](https://huggingface.co/spaces/nafisehNik/girt-space?duplicate=true)")

render_svg(open("assets/logo.svg").read())

st.markdown(
    """
    <style>
    [data-testid="stSidebar"][aria-expanded="true"]{
        min-width: 450px;
        max-width: 450px;
    }
    """,
    unsafe_allow_html=True)


with st.sidebar:
    st.title(" 🔧 Settings")

    with st.expander("🏗 Issue Template Inputs", True):
        
        in_name = st.text_input("Name Metadata: ", placeholder="e.g., Bug Report or Feqture Request or Question", on_change=None)
        in_about = st.text_input("About Metadata: ", placeholder="e.g., File a bug report", on_change=None)

        empty_title = st.checkbox('without title')
        if empty_title == False:
            in_title = st.text_input("Title Metadata: ", placeholder="e.g., [Bug]: ", on_change=None)
        else:
            in_title = '<|EMPTY|>'    
        
        empty_labels = st.checkbox('without labels')
        if empty_labels == False:
            in_labels = st.text_input("Labels Metadata: ", placeholder="e.g., feature, enhancement", on_change=None)
        else:
            in_labels = '<|EMPTY|>'

        empty_assignees = st.checkbox('without Assignees')
        if empty_assignees == False:
            in_assignees = st.text_input("Assignees Metadata: ", placeholder="e.g., USER_1, USER_2", on_change=None)
        else:
            in_assignees = '<|EMPTY|>'
            
        # if no headlines is selected, force the headlines to be empty as well.
        in_headline_type = st.selectbox(
        'How would you like to be your Headlines?',
        ('**Emphasis**', '# Header', 'No headlines'))

        if in_headline_type!='No headlines':
            in_headlines = st.text_area("Headlines: ", placeholder="Enter each headline in one line. e.g.,\nWelcome\nConcise Description\nAdditional Info", on_change=None, height=200)
            in_headlines = in_headlines.split('\n').strip()
        else:
            in_headline_type = '<|EMPTY|>'
            in_headlines = '<|EMPTY|>'

        # df = pd.DataFrame(
        # [{"headline": "Welcome"},{"headline": "Concise Description"}, {"headline": "Additional Info"}])
        # in_headlines = st.experimental_data_editor(df, num_rows="dynamic")

        in_summary = st.text_area("Summary: ", placeholder="This Github Issue Template is ...", on_change=None, height=200)


    with st.expander("🎛 Model Configs", False):
        max_length = st.slider("max_length", 30, 512, 300)
        min_length = st.slider("min_length", 0, 300, 30)
        top_p = st.slider("top_p", 0.0, 1.0, 0.92)
        top_k = st.slider("top_k", 0, 100, 0)

    
    clicked = st.button("Submit", key='prompt')

    with st.spinner("Please Wait..."):
        prompt = create_instruction(in_name, in_about, in_title, in_labels, in_assignees, in_headline_type, in_headlines, in_summary)

        res = compute(prompt, top_p, top_k, do_sample=True, max_length, min_length)
        st.code(res, language="python")

tab1, tab2 = st.tabs(["Design GitHub Issue Template", "Manual Prompt"])

with tab1:

    template_prompt = "name:"
    filled_prompt = "name:"

    clicked = st.button("Submit", key='design')

    with st.spinner("Please Wait..."):

        if filled_prompt!=template_prompt:
            res = compute(prompt, top_p=0.92, top_k=0, do_sample=True, max_length=300, min_length=0)
            st.code(res, language="python")


with tab2:

    st.markdown('This part is only based on the prompt you provide here and not the issue template inputs.')

    prompt = st.text_area("Prompt: ", placeholder="Enter your prompt.", on_change=None, height=200)

    clicked = st.button("Submit", key='prompt')

    with st.spinner("Please Wait..."):

        if prompt:
            res = compute(prompt, top_p=0.92, top_k=0, do_sample=True, max_length=300, min_length=0)
            st.code(res, language="python")