import streamlit as st
import torch
from transformers import GPT2LMHeadModel, GPT2Tokenizer

model_path = "finetuned_model/"
model_name = "sberbank-ai/rugpt3small_based_on_gpt2"
tokenizer = GPT2Tokenizer.from_pretrained(model_name)
model = GPT2LMHeadModel.from_pretrained(model_path)


promt = st.text_input("Ask a question")
generate = st.button("Generate")
if generate:
    if not promt:
        st.write("42")
    promt = tokenizer.encode(promt, return_tensors="pt")
    model.eval()
    with torch.no_grad():
        out = model.generate(
            promt,
            do_sample=True,
            num_beams=2,
            temperature=1.5,
            top_p=0.9,
        )
    out = list(map(tokenizer.decode, out))[0]
    st.write(out)