app.py

import streamlit as st
import time
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

@st.cache(allow_output_mutation=True)
def define_model():
    model = AutoModelForCausalLM.from_pretrained("facebook/opt-1.3b", torch_dtype=torch.float16).cuda()
    tokenizer = AutoTokenizer.from_pretrained("facebook/opt-1.3b", use_fast=False)
    return model, tokenizer

@st.cache(allow_output_mutation=True)
def opt_model(prompt, model, tokenizer, num_sequences = 1, max_length = 50):  
    input_ids = tokenizer(prompt, return_tensors="pt").input_ids.cuda()
    generated_ids = model.generate(input_ids, num_return_sequences=num_sequences, max_length=max_length)
    answer = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
    return answer


model, tokenizer = define_model()

prompt= st.text_area('Your prompt here',
 '''Hello, I'm am conscious and''')
 
answer = opt_model(prompt, model, tokenizer,)
#lst = ['ciao come stai sjfsbd dfhsdf  fuahfuf  feuhfu wefwu ']
lst = ' '.join(answer)

t = st.empty()
for i in range(len(lst)):
    t.markdown("### %s..." % lst[0:i])
    time.sleep(0.04)