Each time I start my script it just gets killed on my VPS

🚀 Deployment
1

So I made a script to run through streamlit that accepts a phrase and uses Parrot to paraphrase it and then it gives it to GPT Neo to create an article.
here is my code

from parrot import Parrot
from transformers import pipeline, GPTNeoForCausalLM, GPT2Tokenizer
import torch
import warnings
import streamlit as st


@st.cache
def test_generate(input_str: str):
    tokenizerfile = GPT2Tokenizer.from_pretrained("EleutherAI/gpt-neo-125M")
    tokenizerfile.pad_token = tokenizerfile.eos_token
    gradient_ckpt = True
    model = GPTNeoForCausalLM.from_pretrained("EleutherAI/gpt-neo-125M", pad_token_id=tokenizerfile.eos_token_id, gradient_checkpointing=gradient_ckpt, use_cache=not gradient_ckpt)
    input_ids = tokenizerfile.encode(input_str, add_special_tokens=False, return_tensors="pt")
    attention_mask = torch.where(input_ids == tokenizerfile.eos_token_id, torch.zeros_like(input_ids), torch.ones_like(input_ids)).to(model.device)
    output_ids = model.generate(input_ids, attention_mask=attention_mask, max_new_tokens=750, num_return_sequences=10)
    output_str = tokenizerfile.decode(output_ids[0], skip_special_tokens=False, clean_up_tokenization_spaces=False)
    st.write(output_str)


parrot = Parrot(model_tag="prithivida/parrot_paraphraser_on_T5", use_gpu=False)
phrases = []
phrases.append(st.text_input("insert phrase to search"))
for phrase in phrases:
    para_phrases = parrot.augment(input_phrase=phrase, use_gpu=False)
    st.write(para_phrases)
    selected_items = [item[0] for item in para_phrases]
test_generate(selected_items[0])

Each time I run it It just gets killed and exits

2

This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.