app.py

#import streamlit as st

#x = st.slider('Select a value')
#st.write(x, 'squared is', x * x)
import streamlit as st
from transformers import pipeline, AutoModelForMaskedLM, AutoTokenizer


st.title("Completamento del testo in Latino con Latin BERT")
st.write("Inserisci un testo con il token [MASK] per vedere le previsioni del modello.")


st.write("Esempi di testo:");
st.write("Asdrubal, frater Annibalis, qui secundo Punico bello [MASK] ingentibus copiis ab Hispania veniens > cum");
st.write("hanno et mago qui [MASK]  punico bello cornelium consulem aput liparas ceperunt > primo");
st.write("Lorem ipsum dolor sit amet, [MASK] adipiscing elit. > consectetur");
st.write("Populus Romanus cum Macedonibus [MASK] ter gessit => bellum");
#Asdrubal, frater Annibalis, qui secundo Punico bello [MASK] ingentibus copiis ab Hispania veniens => cum
#hanno et mago qui [MASK]  punico bello cornelium consulem aput liparas ceperunt => primo
#Lorem ipsum dolor sit amet, [MASK] adipiscing elit. => consectetur
input_text = st.text_input("Testo:", value="Lorem ipsum dolor sit amet, [MASK] adipiscing elit.")

# Model based on BERT
#modelname = "./models/latin_bert/"
#Hugging face LuisAVasquez/simple-latin-bert-uncased
#modelname_lv = "LuisAVasquez/simple-latin-bert-uncased"
#https://github.com/dbamman/latin-bert
modelname = "./models/bert-base-latin-uncased"



#tokenizer_roberta = AutoTokenizer.from_pretrained("pstroe/roberta-base-latin-cased3")
#model_roberta = AutoModelForMaskedLM.from_pretrained("pstroe/roberta-base-latin-cased3")
#fill_mask_roberta = pipeline("fill-mask", model=model_roberta, tokenizer=tokenizer_roberta)

tokenizer_robertaclasscat = AutoTokenizer.from_pretrained("ClassCat/roberta-base-latin-v2")
model_robertaclasscat = AutoModelForMaskedLM.from_pretrained("ClassCat/roberta-base-latin-v2")
fill_mask_robertaclasscat = pipeline("fill-mask", model=model_robertaclasscat, tokenizer=tokenizer_robertaclasscat)

tokenizer = AutoTokenizer.from_pretrained(modelname)
model = AutoModelForMaskedLM.from_pretrained(modelname)
fill_mask = pipeline("fill-mask", model=model, tokenizer=tokenizer)

#tokenizer_lv = AutoTokenizer.from_pretrained(modelname_lv)
#model_lv = AutoModelForMaskedLM.from_pretrained(modelname_lv)
#fill_mask_lv = pipeline("fill-mask", model=model_lv, tokenizer=tokenizer_lv)

if input_text:
    predictions = fill_mask(input_text)
    st.subheader("Risultati delle previsioni con Bert:")
    for pred in predictions:
        st.write(f"**Parola**: {pred['token_str']}, **Probabilità**: {pred['score']:.4f}, **Sequence**: {pred['sequence']}")
    input_text_roberta = input_text.replace("[MASK]", "<mask>")
    #predictions_roberta = fill_mask_roberta(input_text_roberta)    
    #st.subheader("Risultati delle previsioni con Roberta Base Latin Cased 3:")
    #for pred_roberta in predictions_roberta:
    #    st.write(f"**Parola**: {pred_roberta['token_str']}, **Probabilità**: {pred_roberta['score']:.4f}, **Sequence**: {pred_roberta['sequence']}")
    predictions_robertaclasscat = fill_mask_robertaclasscat(input_text_roberta)    
    st.subheader("Risultati delle previsioni con Roberta:")
    for pred_robertaclasscat in predictions_robertaclasscat:
        st.write(f"**Parola**: {pred_robertaclasscat['token_str']}, **Probabilità**: {pred_robertaclasscat['score']:.4f}, **Sequence**: {pred_robertaclasscat['sequence']}")