app.py

import streamlit as st

from transformers import T5Tokenizer, T5ForConditionalGeneration


@st.cache_resource
def init_model():
    model = T5ForConditionalGeneration.from_pretrained("t5-small")
    tokenizer = T5Tokenizer.from_pretrained("t5-small")
    return model, tokenizer


max_source_length = 512
max_target_length = 128

model, tokenizer = init_model()

st.title('T5-Small')
with st.form('my_form'):
    text = st.text_area('Enter text:', '')
    cols = st.columns(3)
    submitted = cols[0].form_submit_button('translate')
    task_prefix = cols[1].text_input("input language", "translate Chinese to English: ")

placeholder = st.markdown("", unsafe_allow_html=True)
if submitted:
    with st.spinner("Translating..."):
        input_ids = tokenizer(f"{task_prefix}{text}", return_tensors="pt").input_ids
        outputs = model.generate(input_ids)
        placeholder.markdown(tokenizer.decode(outputs[0], skip_special_tokens=True))