app.py

import streamlit as st
import os
from transformers import AutoTokenizer, pipeline
import torch

# get the Hugging Face API token from environment variables
hugging_face_token = os.getenv('HUGGINGFACEHUB_API_TOKEN')

def getLLamaresponse(input_text, no_words, blog_style):
    model = "daryl149/llama-2-7b-chat-hf"
    tokenizer = AutoTokenizer.from_pretrained(model, use_auth_token=hugging_face_token)

    # Configure the model's device (CPU or GPU) based on the environment
    device = 0 if torch.cuda.is_available() else -1

    text_generation_pipeline = pipeline(
        "text-generation",
        model=model,
        tokenizer=tokenizer,
        use_auth_token=hugging_face_token,
        torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
        device=device
    )

    # prompt
    prompt = f"Write a blog for a {blog_style} job profile about {input_text} within {no_words} words.\n"

    # generate
    sequences = text_generation_pipeline(
        prompt,
        do_sample=True,
        top_k=10,
        num_return_sequences=1,
        max_length=int(no_words),
    )

    # print and return the text of the first generated sequence
    for seq in sequences:
        print(f"Result: {seq['generated_text']}")
        return seq['generated_text']

# streamlit
st.set_page_config(page_title="Generate Blogs",
                   page_icon='🤖',
                   layout='centered',
                   initial_sidebar_state='collapsed')

st.header("Generate Blogs 🤖")
input_text = st.text_input("Enter the Blog Topic")

# creating to more columns for additonal 2 fields
col1, col2 = st.columns([5, 5])
with col1:
    no_words = st.text_input('No of Words')
with col2:
    blog_style = st.selectbox('Writing the blog for',
                              ('Researchers', 'Data Scientist', 'Common People'), index=0)

submit = st.button("Generate")

# Final response
if submit:
    st.write(getLLamaresponse(input_text, no_words, blog_style))