| 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) | |