app.py

""" Script for streamlit demo
    @author: AbinayaM02
"""

# Install necessary libraries
from transformers import AutoTokenizer, AutoModelWithLMHead, pipeline
import streamlit as st
import json

# Read the config
with open("config.json") as f:
    config = json.loads(f.read())

# Set page layout
st.set_page_config(
        page_title="Tamil Language Models",
        page_icon="U+270D",
        layout="wide",
        initial_sidebar_state="expanded"
    )

# Load the model
@st.cache(allow_output_mutation=True)
def load_model(model_name):
    with st.spinner('Waiting for the model to load.....'):
        model = AutoModelWithLMHead.from_pretrained(model_name)
        tokenizer = AutoTokenizer.from_pretrained(model_name)
    return model, tokenizer

# Side bar
img = st.sidebar.image("images/tamil_logo.jpg", width=300)

# Choose the model based on selection
st.sidebar.title("கதை சொல்லி!")
page = st.sidebar.selectbox(label="Select model", 
                            options=config["models"],
                            help="Select the model to generate the text")
data = st.sidebar.selectbox(label="Select data",
                            options=config[page],
                            help="Select the data on which the model is trained")
if page == "Text Generation" and data == "Oscar + IndicNLP":
    st.sidebar.markdown(
        "[Model tracking on wandb](https://wandb.ai/wandb/hf-flax-gpt2-tamil/runs/watdq7ib/overview?workspace=user-abinayam)"
        "[Model card](https://huggingface.co/abinayam/gpt-2-tamil)"
        )
elif page == "Text Generation" and data == "Oscar":
    st.sidebar.markdown(
        "[Model tracking on wandb](https://wandb.ai/abinayam/hf-flax-gpt-2-tamil/runs/1ddv4131/overview?workspace=user-abinayam)",
        "[Model card](https://huggingface.co/flax-community/gpt-2-tamil)"
    )

# Main page
st.title("Tamil Language Demos")
st.markdown(
    "Built as part of the Flax/Jax Community week, this demo uses [GPT2 trained on Oscar dataset](https://huggingface.co/flax-community/gpt-2-tamil) "
    "and [GPT2 trained on Oscar & IndicNLP dataset] (https://huggingface.co/abinayam/gpt-2-tamil) "
    "to show language generation!"
)

# Set default options for examples
prompts = config["examples"] + ["Custom"]

if page == 'Text Generation' and data == 'Oscar':
    st.header('Tamil text generation with GPT2')
    st.markdown('A simple demo using gpt-2-tamil model trained on Oscar dataset!')
    model, tokenizer = load_model(config[data])
elif page == 'Text Generation' and data == "Oscar + Indic Corpus":
    st.header('Tamil text generation with GPT2')
    st.markdown('A simple demo using gpt-2-tamil model trained on Oscar + IndicNLP dataset')
    model, tokenizer = load_model(config[data])
else:
    st.title('Tamil News classification with Finetuned GPT2')
    st.markdown('In progress')

if page == "Text Generation":
    # Set default options
    prompt = st.selectbox('Examples', prompts, index=0)
    if prompt == "Custom":
        prompt_box = "",
        text = st.text_input(
        'Add your custom text in Tamil',
        "",
        max_chars=1000)
    else:
        prompt_box = prompt
        text = st.text_input(
        'Selected example in Tamil',
        prompt,
        max_chars=1000)
    max_len = st.slider('Select length of the sentence to generate', 25, 300, 100)
    gen_bt = st.button('Generate')

    # Generate text
    if gen_bt:
            try:
                with st.spinner('Generating...'):
                    generator = pipeline('text-generation', model=model, tokenizer=tokenizer)
                    seqs = generator(prompt_box, max_length=max_len)[0]['generated_text']
                st.write(seqs)
            except Exception as e:
                st.exception(f'Exception: {e}')